Europe’s loss of metal smelting and refining capacity is undermining its ability to produce gallium and germanium, chip materials made as by-products of metal processing.

China produces about 98% of the world’s primary gallium, and both metals have needed Chinese export licenses since August 2023.

“We need the smelters. We need the zinc smelters and the alumina refineries to produce the gallium, germanium and antimony,” said Bernd Schäfer, CEO and managing director of EIT RawMaterials, an EU-backed innovation network. “No midstream, no mine. We have already been losing a lot of midstream.”

He said Europe’s raw materials debate has shifted from supplying the twin digital and green transition to securing materials for defense, and that Europe must first determine what midstream capacity each value chain needs to stay resilient.

Citing Trafigura CEO Richard Holtum, he said Europe’s aluminum smelting capacity fell 30% in a decade and 50% since 2022.

Gallium goes into gallium nitride chips for power supplies in artificial intelligence (AI) data centers, and into chips for radar and 5G networks, while germanium is used in fiber optics and infrared optics. Both are recovered in small quantities during alumina and zinc refining, so supply depends on those plants running.

“This cannot go on, because we are moving in growing markets. So the net loss is much bigger than what has been shown there, and that’s where we are in a make-or-break situation now in Europe,” he said.

James Watson, director general of European Metals, which represents Europe’s non-ferrous metals producers, said a company choosing where to build a smelter today might not pick Europe because electricity is much cheaper elsewhere, even where the carbon footprint is far higher.

He gave Indonesia as an example of where producers could build smelters and then export to the EU.

“We should stay the course on the ETS (Emissions Trading System). We should support those organizations that are decarbonizing Europe by giving them indirect cost compensation until the electricity prices come down,” he said.

The EU’s ETS has long given energy-intensive smelters free allowances and compensation for carbon costs passed through in power prices. He said these are being phased out in favor of the carbon border adjustment mechanism (CBAM), which he called unproven and far from perfect.

He cited the Mosjøen smelter in Norway, which is 91% electrified and receives indirect cost compensation.

America moves first

Schäfer and Watson spoke at the FT Metals & Mining Summit, organized by FT Live in London on October 15. Dan Thomas, the Financial Times' natural resources editor, moderated the fireside chat on carbon costs and shifting trade flows.

Europe relies on China for up to 99% of some critical raw materials and over 70% on average, Schäfer said.

“What we’re doing all starts with facts and figures, not with imposed regulations. We want to know what is Europe’s raw materials equation, to give a very clear answer: how much of what, by when and from where,” he said.

He said the analysis should cover all 34 critical and strategic raw materials, plus several defense metals not on the list. It would show:

• how much capacity Europe must create at home

• how much it must secure through offtakes with reliable partners

• how to cut dependence on single sources

The United States has moved faster, investing in Project Vault and guaranteeing offtakes, he said.

“Look what the Americans have been doing. They’ve been building this rare earth permanent magnet value chain, and while we were still talking, they bought a couple of companies from Europe and are strengthening this value chain,” he said. “Everything that we have been saying over the last years, the Americans have taken and implemented.”

Project Vault, launched in February, is a $12 billion stockpile of minerals including rare earths and gallium, backed by a $10 billion loan from the US Export-Import Bank.

Schäfer said EIT RawMaterials proposed an action plan for a European rare-earth magnet value chain in 2020, estimating that about €2 billion in subsidies would be needed to level the playing field with rivals and offset China’s pricing, often based on figures declared by state administrations. It refreshed the plan late last year.

Europe makes far less than 5% of the permanent magnets it uses today, he said.

“We have no proof point in Europe of one functioning value chain where we are very dependent. But if we manage to get this to 20% or 50% self-sufficiency, then definitely this will create credibility and followership,” he said.

Watson said Project Vault and Canada’s critical minerals funding send investors clear signals that Europe has yet to match.

“Strategic autonomy is the pillar of the new competitiveness agenda that we’ve talked about in Europe. But talk is one thing, and delivery is another,” he said.

Money for the midstream

Watson said the planned European Critical Raw Materials (CRM) Center will test whether Europe is serious about securing these materials. On October 20, the European Parliament voted to make the center a standalone EU agency with a €2 billion fund under the bloc’s 2028-2034 budget.

“Will it have money or not? If it doesn’t have money, it’s not credible,” he said.

He said the European Commission’s RESourceEU Action plan, adopted last December, assembled about €3 billion by plucking figures from existing funding pots that companies must still apply to and struggle to access.

The plan lists AI chips and data centers among the industries it aims to protect from supply shocks, alongside cars, industrial motors, consumer electronics, aerospace and defense. He said Brussels still focuses on the ETS, CBAM and permitting, offering little support on competitiveness.

“We never asked for handouts. It’s not just about subsidies. It’s like a kind of addiction: you start, then everyone gets used to it, and no one wants to give it up,” he said.

He said the center must have finance to help build smelter capacity, and called for a critical raw materials bank.

Schäfer said public money for the center could be leveraged eight to 12 times with private capital, turning a fund of billions into a three-digit billion sum.

He said electricity costs about $100 per megawatt hour in Europe, against about $25 in places such as Canada, and a ton of aluminum takes about 14 megawatt hours to make. That leaves European smelters $700 to $1,000 a ton behind, even though they already run on 78% renewable power.

“The origin of the problem that we are having in Europe is not our CO2 efficiency. We are world-class. The real issue is power costs,” he said.

He said imported aluminum carries far higher carbon emissions than metal made in Europe. With half of Europe’s roughly 12 million tons of annual demand imported, that amounts to about 20 million tons of CO2 a year, similar to Croatia’s total emissions.

Both speakers said CBAM, which entered its definitive phase in January, does little to close the power-cost gap. Schäfer noted that aluminum accounts for about 2% of global greenhouse gas emissions, against 8% for steel.

“It’s a bit like a sieve rather than a border. It’s not particularly watertight,” Watson said.

He said loopholes such as a 50-ton threshold and the zero-carbon treatment of post-consumer scrap let suppliers keep resource shuffling. Because CBAM covers only imports, exporters must apply for temporary rebates, which he described as a bureaucratic puzzle that only adds to the complexity.

“CBAM does not create a business. It’s the cart before the horse,” Schäfer said.

He said Europe should first decide how much capacity it needs and discuss CBAM afterward, and that it will have to rethink its entire raw materials strategy if the CRM Center does not happen.

Technology companies building artificial intelligence (AI) data centers are knocking on miners’ doors for long-term copper supply, but few are ready to buy into a mine.

Most seek offtake agreements, which lock in a share of a mine’s future output, rather than equity stakes.

“It’s high on the mind of most of the people in the downstream, certainly in the new forms of demand such as data centers and AI,” said Duncan Wanblad, chief executive of Anglo American. “Those conversations are absolutely being had.”

“When people start to get their heads around the shape of the mining business, the capital intensity and the nature of the cycles in it, it’s less familiar to those types of businesses,” he said. “But given the scarcity issue, a lot of people are looking for ways to secure supply.”

Some technology firms have shown deep interest in taking equity stakes, though none has done so yet.

China has gone further in consolidating value chains. In nickel, for example, the mine and the processing plant are part of a much bigger chain, and financial viability is calculated over the whole rather than the parts.

Data centers use copper in electrical cables, busbars, transformer windings, circuit boards and processor heat sinks. In January, Amazon Web Services became the first customer for copper from Nuton, Rio Tinto’s bioleaching venture in Arizona, under a two-year deal to supply its US data centers.

Wanblad said Chile’s copper output is down about 8% year on year so far in 2026, largely because of weather, though part of the fall is structural.

“Resources are hard to come by; they’re very hard to permit and bring into production, and you are starting to see the impact of that going through,” he said. “This is not only a demand story; it’s probably as much of a supply story.”

Labor unrest is another pressure point. With copper trading near record highs, unions are seeking a bigger share of the gains, and strikes or threats of strikes have hit mines in several countries recently.

“With copper prices pretty much at an all-time high today, no one would be surprised that labor negotiations seek to attract some of that value,” Wanblad said. “The reality is that we have to plan these businesses through the long run, not just the short run.”

Anglo American is a London-based miner of copper, iron ore and other metals. In 2025, it agreed to merge with Canada’s Teck Resources to form Anglo Teck, a combined group focused on copper.

Storms wipe out buffer

Wanblad spoke at the FT Metals & Mining Summit, organized by the Financial Times in London on October 15. Camilla Hodgson, the newspaper’s commodities correspondent, interviewed him in a keynote session on deals, supply risks and the copper market.

He said climate change is clearly affecting everyone, and extreme weather hit Anglo’s operations across several continents this year:

• Floods in Australia early in the year took out large parts of its coal business

• Floods in South Africa’s Northern Cape in May shut the Kolomela and Sishen iron ore mines

• Snowstorms hit Los Bronces and Collahuasi in Chile in late July and early August

• A cyclone-like storm damaged infrastructure at Minas-Rio in Brazil in September

“There’s been a lot of work over several decades now on improving the resilience to climate change in the business, so the vast majority of these events didn’t have a material production impact,” he said. “Certainly any contingency that we built up during the course of the year has pretty much been wiped out.”

Depending on how rain in Chile plays out, Wanblad is not concerned about meeting this year’s guidance.

Several of this year’s events were rated worse than one-in-200 or one-in-300-year occurrences. They are hitting every business in the sector, not just Anglo, and the whole industry needs to rethink the resilience of its underlying assets, he said.

Geopolitics adds another layer of risk, as governments from Washington to Beijing seek more control over critical minerals.

Wanblad said political uncertainty between a deal’s announcement and its completion turns into financial uncertainty, making boards more cautious about pulling the trigger. He called it a risk-management issue given today’s market volatility.

“There is no doubt that the world as we have it today has a material impact on the board’s view of whether a deal can be executed,” he said. “There seems to be an enormous amount of interest in doing deals, but there are not many deals that are being activated.”

Waiting on Beijing

China is the only authority yet to approve the Anglo Teck merger, and Wanblad said he has no plans to visit Beijing unless he needs to.

He said talks are proceeding as expected, with completion still due between September 2026 and March 2027, but declined to discuss details. A key part of the deal’s logic is combining Teck’s Quebrada Blanca copper mine in Chile with neighboring Collahuasi, co-owned by Anglo, Glencore and a Japanese consortium.

Talks with Glencore and the other shareholders are in the very early stages. Both mines have attractive standalone options, which they will weigh against a combination over the next few months before commercial negotiations begin.

“The rationale and the logic of it remain completely sound. This is one of the world’s most amazing adjacencies. This is growth that has got a very reasonable risk profile, so it’s one that doesn’t break the bank from a capital intensity point of view,” he said.

In Brazil, Anglo’s planned sale of its nickel business to China-backed MMG still faces concerns from the European Commission.

Wanblad said well over 100 parties had shown interest before Anglo narrowed the field to fewer than a handful and picked MMG. That makes a viable sale to another buyer unlikely, and if the MMG deal fails and markets do not improve, the business would very probably be shut.

Anglo runs an asset-backed trading business, mainly trading its own mines’ output. He said the merger’s much bigger equity book will add value, which forms part of the deal’s expected synergies, and Anglo plans to keep growing the business.

He declined to predict whether the US will impose further copper tariffs, but said any tariffs feed through into copper prices and where the metal ends up. The US is now well stocked with copper, and its Project Vault stockpile affects Anglo only indirectly because Anglo mostly produces concentrate rather than refined metal.

Trading will shape copper markets over time, but with new supply scarce and slow to bring on, Wanblad expects copper prices to remain well supported until the gap between supply and demand closes.

Demand from artificial intelligence (AI), chip and defense companies is turning smelter by-products such as germanium and indium into strategic metals in short supply.

These metals go into fiber optics, silicon photonics, infrared and targeting systems.

“It began originally with people that were just consuming finished metal, but now it is fiber optics manufacturers, AI and semiconductor users, large defense primes, all of which are far more interested in this industry than they have been,” said Ian Anderson, executive vice president and chief commercial officer of Teck Resources.

“We’re having semiconductor manufacturers come to us, particularly interested in indium and silicon photonics. That landscape has changed. It’s wider and more nodes down the chain than it has been traditionally, and the awareness is very high,” he said.

Teck is the largest germanium producer in the West, and its Trail smelter in British Columbia, North America’s largest integrated lead-zinc smelter, also produces indium, antimony and cadmium. Anderson said the fiber-optics industry faces a severe shortage of germanium tetrachloride.

He cited Nvidia’s investment in Corning as an effort to lock up fiber-optic supply over the long term. The US chip giant put $500 million into the glassmaker in May to expand fiber production for AI data centers.

He said buyers face shortages now and expect them to last.

“What they’re looking for is a trusted, high-quality producer who behaves the right way, who’s deeply engaged in transparency and producing sustainably, but also someone they can trust for the long term,” Anderson said. “That’s the space the game is played in, as opposed to just a commodity market.”

Matt Hadfield, head of metals and minerals assets at commodities trader Trafigura, said its zinc smelting arm Nyrstar is putting such metals at the center of its plans in Europe and Australia.

“Those critical metals come in today in the raw material that the smelters process. Some of those smelters don’t recover those, so they end up in waste,” he said. “If you want to have these critical minerals processed and available in your country, you need to invest and support the processing capacity you have today.”

He said Nyrstar’s French smelter has produced indium since 2009, and its European plants could also recover germanium, antimony and bismuth. In Australia, antimony and germanium projects are under development as part of a possible rebuild.

In July, Teck signed an agreement with the Canada Growth Fund for an investment of up to C$400 million at Trail. Anderson said the money would mainly expand germanium output, and that customers are keen to back such capacity through long-term purchase agreements.

A national security question

The comments came at the FT Metals & Mining Summit, organized by FT Live in London on October 15, during a panel on smelting and energy costs. Camilla Hodgson, commodities correspondent at the Financial Times, moderated the panel.

The panelists offered different cases for why governments should support smelters.

“It’s an aspect of national security. If countries want to retain the processing capacity within their countries, there needs to be the support structure in order to create a level playing field,” Hadfield said.

He said governments must provide the policy and funding to make smelters globally cost-competitive.

“I would challenge both governments and capital markets to think about investment in processing capacity in a way that doesn’t focus just on individual returns for that specific asset,” Anderson said.

He said that approach is fundamentally wrong if Western smelters are to compete globally. Governments should first support the mines that feed smelters, through faster permitting and investment in rail and terminals. He said China builds smelters close to supply, refineries and grid infrastructure in a hub-and-spoke model.

When judging whether state support is sustainable, governments should weigh what is lost by not acting rather than whether a single smelter can stand on its own, he said.

South Africa once lacked the power to run its smelters but now has a surplus.

Agit Singh, executive head of processing operations at South African platinum group metals (PGMs) producer Valterra Platinum, said cost is the next challenge, and the government is working with industry on renewables.

“That partnership between government and the private sector is what makes the difference,” he said. “Once we understand each other, what our concerns are and what it means for the country, there is a collective understanding that this is going to benefit all of us.”

He said producing green metal is costly because green energy must run through the whole value chain.

“If we want to compete on the playing field on the basis of equality, then every metal needs to be faced with the same stringent requirements,” he said.

Hadfield said Europe is the region hit hardest by energy costs. Power prices in the Netherlands and Belgium are approaching €200 per megawatt hour (MWh), up to four times what other regions pay, as the Middle East crisis drives up gas prices.

“Historical European power prices were something in the range of around €50 per MWh. We’re now looking at €200 per MWh in the forthcoming months,” he said.

He said grid connections in the Netherlands cost four times as much as in Belgium. Nyrstar is reviewing the future of its Budel zinc smelter in the Netherlands, which can produce 315,000 tonnes a year. Hadfield said he met Dutch ministers in The Hague two days before the summit.

“They’re taking the issue very seriously and doing everything that they can within their power to be able to alleviate some of the cost issues,” he said.

Protect what exists

Panelists agreed the West should protect existing smelters before building new ones.

“From a Western perspective, the first priority needs to be protecting what already exists today. Building a new smelter and refinery is an extremely costly and time-consuming exercise,” Hadfield said.

He said no new zinc smelter has been built in the West since 1999, while China has added nearly 2 million tonnes of capacity. He said adding units to recover critical metals at existing smelters is the most efficient way to bring them to market.

Anderson agreed, saying national interests will inevitably focus on supply security and little can be done about global overcapacity. Existing smelters are the natural place to lift output and upgrade technology.

“We have to be focused on brownfields [existing sites]. What we have has got to be full,” Singh said. “The best way to run a smelter is full. If we get half-full smelters around the world, it is going to be a nightmare with regard to costs.”

He said PGM smelting capacity is already sufficient on paper, so consolidation is needed.

“If you directly take everything you’ve got coming out of a mining operation, concentrate it and bring it to the smelter, your efficiencies will be extremely low and your energy costs will be astronomical,” he said.

Valterra consolidated its own smelting about three years ago and cut waste by about 10% by rejecting it before smelting. About 30% of its power now comes from renewables.

Singh said energy costs were not going away, so the company had to work from that basis.

“It’s not just about moving commodities around. These are supply chains that function together, and that is fundamentally how the market is changing,” Anderson said.

He said a small mine ramping up can strengthen its investment case with a 10-year agreement to sell its output to a nearby smelter, and customer prepayments are becoming longer-term. Recycled feed will also grow, with suppliers offering slags and residues smelters can use quickly.

Singh said buyers are not yet paying premiums for greener metal, but he expects them to.

Hadfield said EU policy on critical metals is moving in the right direction but now needs funding, and member states can currently move faster than Brussels.

The metals that feed batteries, chips and defense systems are being split into rival Eastern and Western markets as national security reviews reshape mining deals.

Chinese and Western buyers are now largely shut out of each other’s auctions for critical minerals. Regulators that once ran simple antitrust checks now examine supply chains and market access, and their expected reaction shapes how deals are structured from the outset.

“There’s clearly a divergence now between East and West on critical minerals around the world,” said Shea Small, partner, co-head of mergers and acquisitions (M&A) and co-leader of global metals and mining at McCarthy Tétrault, a Canadian law firm. “In anything that is critical minerals, gold and precious metals excluded, we are seeing China and Western countries largely excluded from each other’s auction processes.”

“Governments are more savvy regarding the overall flows and balances,” said Vance Scott, partner and managing director at AlixPartners and global co-leader of its energy and process industries practice. “If you think about the disruption that exists in the world right now, with the change in flows of supply chains and the concerns around transportation and market access, that’s causing regulators to think at another deeper level.”

He said scrutiny used to fall mainly on large mergers, depending on whether a country produced or consumed the commodity and how important it was to downstream industries. Individual asset sales are now examined too, and that risk feeds into transaction costs and the premium buyers demand.

“The companies that are going to succeed on the transaction front will anticipate that and try to manage ahead of that, and think about how that risk will be factored into their valuation and how that will impact their ability to secure and deploy capital,” he said.

Small said M&A practice had changed drastically over the past decade. Even on smaller deals, regulators’ likely response now drives the structure, with national security and supply-chain reviews added to antitrust. He said the White House is a new wild card that did not exist two or three years ago.

In Canada, 11 of the 16 investments sent to full national security review in the 2024-25 fiscal year involved Chinese investors. Similar reviews are rising in the US and Europe.

“Historically, China was willing to put more money on the table than others to acquire assets,” Small said. “That was the perception, and that has now disappeared in critical minerals.”

He said Africa and South America remain a different mix, where Chinese and Western bidders still compete for the same assets. The EU is examining Anglo American’s nickel sale in Brazil, while Chinese regulators are reviewing the Anglo-Teck merger.

Prices without benchmarks

The panel, titled “Dealmaking 2026: executing deals under tighter capital and rising intervention,” was held at the FT Metals & Mining Summit 2026 in London on October 15. Alex Hamer, the Financial Times’ assistant companies news editor, moderated it.

Carlos Francisco Fernandez, chief executive of FinkaMetals, said metals fall into two camps, depending on how the market prices them.

“On one hand, you have copper, which is a metal that has a lot of depth and liquidity with Western benchmarks, and you have a similar story with gold,” he said. “In these markets, the capacity for buyers or acquirers to underwrite an asset and compete for it with confidence is a lot easier.”

Strategic metals such as lithium, cobalt, antimony and tungsten are a completely different story. They have no liquid Western benchmarks, most sales are bilateral, processing is heavily concentrated in China and export restrictions sit on top. That makes it far harder for buyers to bid with confidence.

“With that lack of transparency and understanding of the market, it’s more difficult to bid for an asset,” Fernandez said. “You don’t agree on the value of the asset because you cannot even agree on the value of the product that asset is selling.”

“What you need in the market is infrastructure that ultimately closes that gap, because the price discovery gap is becoming the valuation gap,” he said. “When you have a better understanding between buyer and seller of what the value of the product is, that will allow for transactions to happen a lot easier.”

Transparency also matters inside joint ventures (JVs), which miners use to share the cost and risk of developing large projects.

“Once the JV is set up, the sharing of the capex (capital expenditure) is one thing; how that material is then sold is a different question,” Fernandez said. “The successful JVs are those that can provide visibility on how material is being sold and on what terms to every JV counterparty.”

Scott said combining portfolios along the development cycle can create real synergy value. A partner further down the path can bring processing capacity and a downstream sales function that reaches markets, so producers avoid taking an unnecessary discount on the commodity price.

“You don’t have the outlay for processing kit, and you can repurpose, refurbish and bring those assets back to performance, and then get much faster access to markets,” he said. “There is value in speeding all that up.”

State plays four roles

Governments are moving deeper into the sector, with the US government among those getting involved upstream. Fernandez said state capital funds projects that private money alone would not, citing Chile’s state-owned copper miner Codelco.

“Today you have the state as a shareholder, as a lender, as a regulator and as a buyer at the same time,” he said. “That makes it a lot more complicated, and that’s new for most investors.”

Transparent pricing and full traceability of where metal is sold would protect acquirers, co-investors and governments spending taxpayers’ money, because they show assets were bought at fair market value, Fernandez said.

Sibusiso Ndlovu, director of metals and mining M&A at Standard Chartered, said boards would ask whether such partnerships stay stable.

“Investors will always reward stability,” she said. “The question that will remain is: is this going to provide stability over my operations? Not just when things are going well, but even when things go bad, I still need to know that I’m able to manage the situation in the event of a crisis.”

The geopolitical split is also changing where companies choose to be based. Small said some Canadian clients are weighing a move abroad, because being domiciled in one jurisdiction can automatically create a connection with that country for investment purposes.

“It is changing how people view where to set themselves up,” he said. “I’m not sure it’s going to get much bigger than being on the fringes, but it’s definitely a consideration.”

Valuations add another hurdle. Spot prices for most commodities have climbed far above long-term consensus forecasts, and junior developers often treat them as a proxy for their own worth.

“Valuation is probably one of the toughest things we’re facing in all the deals we’re trying to do right now,” Small said. “If you look at the implied valuation off those spot prices versus the long-term consensus, you almost can’t make a deal work.”

Buyers are bridging the gap with tools such as contingent value rights, he said. Gold deals are mostly share-for-share because prices can swing either way, while copper deals tend to be in cash, based on long-term consensus prices.

Ndlovu said some junior miners’ valuations have expanded on commodity prices rather than project progress, and that investors underwriting a mine must look 10 to 13 years ahead rather than 12 months, weighing today’s prices against the long-term demand that critical-mineral supply will have to meet.

Soaring demand for critical minerals from artificial intelligence (AI), data centers and defense is pushing financiers to test whether new supply chains can last.

Yet new mines are slow to come on stream, and some materials need careful handling.

“The energy transition, the rapid adoption of AI and even the defense sector where it is today have created the spike in demand for critical minerals,” said Biro Condé, managing director for West and Central Africa at the US International Development Finance Corporation (DFC).

“When it comes to rare earths, you have to be very careful about not only processing capacity, but even the durability and the sustainability of the supply chain,” he said.

He said DFC is very selective on rare earth investments, which are highly important for the defense sector.

“We go as far as even creating chains, connecting processing plants where we’ve invested with mines that are potential suppliers for some of these plants. So we play a catalyzing role in the sector to unlock this value,” he said.

DFC keeps a public list of critical minerals and a separate list of strategic minerals.

Beyond copper, Condé said the agency is interested in manganese, lithium and zinc, and is closely watching iron ore projects and where it can diversify those value chains.

Some in the market worry that the AI story has been overplayed and that global copper supply is healthier than it looks. Inflation and central bank interest rate rises add to the uncertainty.

Phesto Musonda, board chairperson of ZCCM Investments Holdings (ZCCM-IH) in Zambia, said the projected copper demand is still there.

He said defense contractors and data center operators need to plan ahead, so they want supply security and look for market makers and traders that can hold stocks and balance supply.

Few proper new deposits have been found in recent years, and developing a copper asset takes 10 to 15 years from exploration to delivery. As grades fall from 1% to 0.5%, a mine producing 250,000 tons must excavate twice as much material for the same copper.

“We are not developing assets as fast as we should. That in itself will continue to make copper a sought-after commodity, a critical mineral,” he said.

Mining meets national security

The FT Metals & Mining Summit 2026, organized by the Financial Times, took place in London on October 15. Camilla Hodgson, commodities correspondent at the Financial Times, moderated a panel on what it will take for Africa to capture more value from its minerals.

“Let’s not be naive. The elephant in the room on this subject is that mining is becoming an important part of not only national security, but also competitiveness,” Condé said.

“When we think about partner countries like the DRC (Democratic Republic of Congo) or any other countries in Southern and Central Africa, eventually as they move up the value chain, they will naturally start thinking more in that regard, especially from a national security standpoint,” he said.

Gécamines, the DRC’s state-owned miner, and Swiss commodity trader Mercuria set up a trading joint venture (JV) in December 2025 with DFC support. US buyers get first refusal on its copper, but only at fair market prices.

Condé said DFC’s financing for the JV is going through its normal underwriting process.

“We see an opportunity to not only meet US priorities as far as diversifying the mineral supply chain, but creating supply chains that are durable and sustainable,” he said.

Condé said DFC is also working on deals to secure metal for Project Vault, a US strategic minerals reserve.

Kostas Bintas, global head of metals and minerals at Mercuria Energy Trading, said the first 12,000 tons of DRC copper had gone into Project Vault that week. He said the DRC has committed to ship at least 100,000 tons of copper to the US through the JV.

Project Vault is a $12 billion stockpile launched in February and backed mainly by Export-Import Bank of the United States loans. Mercuria is one of its anchor investors.

Bintas said all JV sales are on market terms, and copper that US buyers turn down can go anywhere.

“Now this copper finds its way not only to the United States but to the strategic reserve initiative of the United States. A lot of policy initiatives have started to have real effects on the market and on the flows,” he said.

He said about 80% of the world’s copper stocks, roughly 1.8 million tons, now sit in the US because of the risk of a tariff on copper cathode. That is more than two years of US import needs.

“Policy that impacts price in critical minerals will continue in a broader way. We’re going to see the kind of tension we have seen on critical minerals in the last two to four years, and it will be stronger,” he said, adding that without a tariff, the market would send the metal back mainly to China.

Beyond the mine

Condé said DFC’s strategy goes beyond the mine, to whether partner countries can also build the energy, infrastructure and productive capabilities around it.

“Beyond the tax revenue, beyond the royalties, critical infrastructure has a tremendous impact in transforming these countries where we’re investing. That falls right in line with our development objectives for some of these countries, but also securing our value chain and making sure that they’re durable,” he said.

African state miners want a bigger share of that value.

“One of our priorities is to build our own production capacity, because Gécamines has evolved to become more like a passive investor, waiting for dividends,” said Baraka Kabemba, chief executive of Gécamines.

He said Gécamines will look to acquire existing assets in the Katanga region, pursue large greenfield exploration projects and build processing plants in the DRC to capture more value locally. The DRC, now the world’s second-largest copper producer, aims to become the largest by 2030 with 6 million tons.

Musonda said Zambian and Congolese copper long went to China as cathode and came back to Africa as finished goods.

“As Africa, we are saying things are changing. We need to add value to our copper. We are so close to it,” he said.

He said Africa’s young population will form a middle class buying electric vehicles, phones and laptops, though adding value locally will take time, technology and skills. Zambia has already deployed the first electric haul trucks of their kind to carry copper ore.

Bintas said staff from the African state partners sit in Mercuria’s Geneva offices, follow the transactions, and review the negotiations and documents so the knowledge can be transferred to their governments.

Kabemba said the JV is not meant to control all of the DRC’s copper.

“Everybody knows that China is our largest partner right now. But we have to make choices sometimes. For the time being, as far as copper is concerned, commercial agreements are with the United States,” he said.

Musonda said ZCCM-IH works with partners from Singapore, China, India and the US across its energy and copper assets, and a new trading platform does not override their rights, adding that even copper sent to China ends up in cables used in the US, Paris and the UK.

“The geopolitics is neither here nor there. Copper is a commodity for all of us here,” he said.

Kabemba said the DRC remains a multilateralist country and aims to bring in as many partners as possible to increase competition in the country.

A mining giant best known for iron ore aims to nearly double its copper output as data centers and electrification lift demand for the metal.

Copper is essential to data centers, power grids and electric vehicles, and demand is expected to keep rising.

Annual copper production stands at about 380 kilotons, against a target of 700 kilotons within five years from the Carajás region in northern Brazil. Rail lines and other infrastructure are already in place there, which the miner says halves the capital needed compared with a typical greenfield project.

“We’ll continue to be a diversified company, but certainly with a higher share coming from our copper story,” said Gustavo Pimenta, Vale's chief executive.

“We believe iron ore, medium to long term, has a role to play in urbanization, decarbonization and population growth. But certainly our copper story will grow substantially in the coming years. We have two projects under construction as we speak,” he said.

Often asked about acquisitions, Pimenta said Vale would rather develop its existing deposits in Brazil than buy rivals.

Vale Base Metals (VBM), the group’s copper and nickel unit, accounts for about 30% of group earnings before interest, taxes, depreciation and amortization (EBITDA). Pimenta expects that share to rise above 40% in the coming years.

He said a VBM listing could help fund growth but is a means to an end and not imminent.

Founded in 1942 and based in Rio de Janeiro, Vale is the world’s largest iron ore producer. Carajás, in the northern state of Pará, hosts its biggest iron ore mines and its Salobo and Sossego copper operations. Within five years, Vale expects iron ore output to reach about 360 million tons.

Balanced despite Simandou

Pimenta was speaking at the FT Metals & Mining Summit, organized by the Financial Times (FT) in London on October 15. FT commodities correspondent Camilla Hodgson moderated the keynote interview on iron ore, copper and Brazil’s mining ambitions.

Rio Tinto and Chinese partners have begun shipping from Simandou in Guinea, designed to produce up to 120 million tons a year. Pimenta said depletion removes 60 to 80 million tons of supply each year, which Simandou will help offset over the next few years.

“The view we have is that the market will continue to be balanced because of the amount of depletion that we are seeing in our space,” he said.

Alumina is an impurity that raises energy use and slag volumes in blast furnaces, so producers blend ores to keep it in check. Pimenta said Simandou’s ore has higher alumina content, which will raise demand for low-alumina ore, such as Vale’s Brazilian supply, to blend with it.

China’s crude steel output is slightly down this year at just over 1 billion tons, as manufacturing replaces property.

“Demand in China continues to be solid. We’ve seen that shift in the Chinese economy, but we are also seeing growth outside China,” he said.

Elsewhere, Pimenta, who recently visited Vietnam, said the country is tripling its crude steel production. India is growing too, and he cited a recent report projecting that its steel output could rise from 170 million tons to about 600 million tons. He said this growth outside China is encouraging for long-term demand.

“We are seeing China reducing but not substantially, continuing to be a very important market for us. Our view is that medium to long term, even with Simandou, the market remains balanced,” he said.

Brazil’s untapped endowment

Luiz Inácio Lula da Silva won Brazil’s presidential runoff against Flávio Bolsonaro on October 25. At the summit 10 days earlier, Pimenta said mining accounts for about 4% of Brazil’s gross domestic product, compared with more than 10% in Australia. A recent Vale study set out how Brazil could unlock its potential.

“Mining is already very relevant in Brazil. Brazil has an enormous endowment, which could be unlocked. For the country, it would be an enormous opportunity given the endowment that we have there,” he said.

Mining is increasingly discussed in Brazil, including how to speed up licenses while ensuring the industry develops responsibly. Pimenta said both candidates had been focused on unlocking that potential and that the mining agenda was aligned with either side of politics.

“Doesn’t matter if it’s right or left. This is good for the country, and that’s our wish,” Pimenta said.

Vale has held talks with the US, Europe and others keen on Brazilian minerals, from rare earths to copper. He said global interest is high, and Brazil will also see projects from other mining companies developed there.

Washington has been seeking critical mineral supplies outside China, which dominates rare-earth processing and many battery metals.

“Brazil has this unique characteristic of having the scale, quality, the good jurisdiction and infrastructure to supply the US, Europe and many others. So we’ve been in conversations with all of them,” he said.

Mining is essential to the energy transition and digitalization, yet the industry still struggles to win public trust.

“If you run any survey where we compare the perception of mining vis-à-vis other sectors, unfortunately, we are one of the lowest in terms of public perception,” Pimenta said.

He said the industry must show that it improves lives and communities while preserving the environment. He cited Carajás, where Vale says it uses only 2% to 3% of the territory and helps protect 800,000 hectares of forest, about five times the size of London, while land outside it has been used for other purposes.

“We are convinced that we can work towards a future where we can not only provide what is essential, but we can make the world better,” he said.

China buys most of the world’s seaborne iron ore, and its state-backed buyer, China Mineral Resources Group (CMRG), held protracted talks with BHP last year. Pimenta said Vale’s talks with CMRG cover decarbonization and ore-blending solutions, and he is optimistic as they continue in the coming years.

Western governments can cut their reliance on Chinese rare earths for artificial intelligence (AI), cars and defense, but only if they fast-track new processing plants.

China mines about 60% of the world’s magnet rare earths, refines more than 90% of them and makes almost 95% of permanent magnets, according to the International Energy Agency (IEA). Breaking that grip has become a priority for Washington since Beijing began restricting rare earth exports.

“It can be done, but the thing you have to realize is the situation we are in today developed incrementally over a period of years. You just cannot create a new ecosystem around rare earths by way of magic wand,” said Jim Rutherford, non-executive director of Ionic Rare Earths.

“When you’re staring at the possibility of your car industry and your defense industry being shut down within six months, it does tend to focus the mind.”

He said the US government had woken up to the problem. At the BMO mining conference in February, he lost count of how many US government departments were present, all seriously seeking investment. He said the Trump administration had brought heavyweight business experience into government to drive the effort.

Rutherford said Western miners had made a rational choice to put their capital upstream. Further down the chain, the capital employed rises while the economic rent shrinks, and winning environmental approval for a smelting and refining operation can take years.

Asked whether breaking the dependency could take 20 years, he said the answer hinges on downstream processing.

“Unless something is done to shift and enable fast-tracking of a lot of downstream processing operations, it will take 20 years,” he said.

He said a conventional banker would hesitate to finance a rare earth project with opaque pricing, one dominant player and high volatility. That is where three tools come in:

• Government underwriting

• Strategic stockpiles

• Minimum floor prices

Europe and the UK, including the UK National Wealth Fund, are catching up but need joined-up thinking, he said.

Rutherford separated such structural problems from war-related shocks. The Middle East conflict has squeezed fertilizer makers out of the sulfur market because copper producers need sulfuric acid, a dislocation he hoped would be temporary.

Asked by an audience member whether the Trump administration’s recent measures to build rare earth supply chains could deliver results within three to five years, Laura Konkel, senior counsel at law firm Akin Gump, said success would come down to the sheer number of deals.

“It’s a volume game. Not everything is going to work. But when you are investing in a significant number of companies in an industry, the statistical likelihood of success obviously increases,” she said.

She said it was now up to the industry to deliver on due diligence, project timelines and milestones.

Governments as anchor investors

The panel discussion, titled “Minerals diplomacy: responding to the impact of war on trade and supply chains,” took place at the FT Metals & Mining Summit, organized by the Financial Times in London on October 15. Gideon Rachman, the newspaper’s chief foreign affairs commentator, moderated it.

Konkel said she does significant work for US government agencies active in the sector.

The Department of War’s Office of Strategic Capital alone can provide up to $100 billion in loans and guarantees for critical minerals, backed by a $500 million credit subsidy. The Export-Import Bank of the United States (EXIM) and the US International Development Finance Corporation (DFC) are also investing.

She said state money is now pulling in private investors, with $500 million of government debt sometimes matched by the same sum in private capital.

“No longer are governments just regulators. They are one of the anchor investors in the space,” she said. “The US government may ask for $500 million in private capital to be raised, and the crazy part about it is that they’re able to raise it. That’s how frothy the markets are right now.”

She said governments are also creating trading blocs through restricted-entity rules and offtake limits. Companies can raise money only from specific investors in specific countries, which can slow broader mergers and acquisitions.

Many of the people now driving these deals come from private equity and hedge fund backgrounds. The deals are publicly filed, and the government takes warrants and equity stakes to secure both upside and downside protection for taxpayers, something Konkel said had not happened before.

“They’re not just coming in saying let’s talk about a project. They’re saying let’s talk about a project, and what’s your ROI (return on investment)? There are no free handouts,” she said.

Rutherford said agencies such as the DFC had hired people who understand capital allocation and carry out serious due diligence. He added that the IEA’s estimate of $60 billion for rare earths alone shows why such large sums are needed once other commodities are added.

Kuswantoro Pranabudi, director of corporate strategy, business development and portfolio at Indonesia’s state-owned mining holding company MIND ID, said the shift is visible from the other side of the negotiating table.

“Five years ago, when we discussed with partners, normally they were asking about possible costs, what the grades are, and what the price is. Now it’s totally changing,” he said. “When we discuss with the potential partner, even before we talk about commercial terms, they ask about where the product is going to go.”

He said minerals have almost become a security issue over the past 18 months.

“We would like to be diversifying. We’d like to have our partners as long as this is mutually beneficial,” he said.

About a decade ago, Indonesia launched a policy to move beyond extraction into processing, and MIND ID, whose group includes Antam’s bauxite operations, has since expanded into refining.

Recycling as a shortcut

Rutherford said the panel had so far focused on primary supply, but a very substantial pool of rare earth material will come from end-of-life wind turbines and electric cars.

“Where the West does have an inbuilt comparative advantage, and has yet to be fully exploited, is on the recycling side,” he said. “With proper economic incentives and infrastructure put in place, the West could catch up more quickly in certain areas.”

Ionic Rare Earths plans to roll out its magnet recycling technology in Northern Ireland first, before scaling it to other countries.

Rebecca Tomkinson, agent general for Western Australia in the UK and Europe, said critical minerals do not follow traditional market mechanisms and must be approached differently as an economic concept. She said an F-35 fighter jet needs about 417 kg of rare earths and critical minerals, while a submarine needs four tonnes.

She said Western Australia is reindustrializing former industrial hubs, where communities already accept heavy industry and want jobs back.

“When we used to have a gallstone, we used to go in to extract that gallstone with open abdominal surgery. Today, you go in with a keyhole, and it’s a day procedure. Mining is the same,” she said.

In the Pilbara region, people living in the city remotely run mines 2,000 km away.

Tomkinson said the US has approached the state about projects, including Alcoa’s plan to produce gallium with US backing. Asked whether Washington wants Western Australia to do less with China, she said the new partnerships add to, rather than replace, existing trade.

“I don’t think any of us would be sitting in this room five years ago talking about the importance of gallium, yet that is how significantly those commodity mixes have changed,” she said.

Konkel said the minerals boom is drawing young people into sophisticated manufacturing jobs that they see as protected from AI because they are building the AI revolution. She pointed to North Carolina, where Vulcan Elements is building a magnet production facility with the help of tax incentives.

“You either grow it or you mine it, and so the first stop is mining and critical materials for the AI revolution,” she said.

She said private equity and hedge fund figures, such as the founders of MP Materials, are helping to kickstart the industry, while strong environmental and social protections in countries such as Canada and Australia mean people will not allow past harms to be repeated.

Tomkinson said aging smelters and other infrastructure need reinvestment, and the required capital will mean governments and companies work in partnership so new supply comes online as quickly as possible.

Artificial intelligence (AI) is helping make mining a national priority, as governments begin to treat metals and minerals as infrastructure rather than tradable commodities.

Without metals and minerals, AI would not be possible, the energy system would not work, defense industries could not rearm and supply chains could not be restocked. That dependence could bring state financing, subsidies and even public ownership to an industry long left to market cycles.

“Governments increasingly see metals and minerals as public goods. Historically, we’ve been privately traded commodities, driven by demand and supply. No longer,” said Rohitesh Dhawan, president and chief executive of the International Council on Mining and Metals (ICMM).

“The public benefit of metals and minerals means that we are now being seen much more like national infrastructure than a privately traded commodity. This isn’t just semantics. It’s a fundamental mindset shift.”

He said the change will open up policy tools so far reserved for ports, railways and bridges.

“I’m talking about things like financing support, help with land acquisition, subsidy and, in some cases, even public ownership of metals and minerals,” he said.

Washington has already moved in that direction. It announced a $400 million equity investment in rare earths producer MP Materials in July 2025, paired with a 10-year price floor, and has since agreed stakes in Lithium Americas, Trilogy Metals and USA Rare Earth. It has also backed a Korea Zinc smelter project in Tennessee.

“Overall this is positive for mining, but it’s uncharted territory. We need to prepare ourselves by changing how we see ourselves, not as a supplier of commodities, but as a part of critical national infrastructure,” he said.

Founded in 2001 and based in London, ICMM brings together the chief executives of 26 mining and metals companies. It estimates that its membership accounts for about a third of the global industry. Dhawan has led it since April 2021.

Tools of statecraft

Dhawan gave opening remarks at the FT Metals & Mining Summit 2026, which the Financial Times organized in London on October 15. He outlined three fundamental shifts that he said have brought generations of change to mining in just five years: geopolitics, supply, and reputation.

He said these shifts set the agenda for the two-day summit.

“By understanding what has changed and why it’s changed, we can operate from a place of wisdom,” he said.

He opened by pointing to a week of headlines about missiles in Riyadh, protests in Delhi and treason in Germany. He said much has been said about geopolitics, but two aspects of how it intersects with mining remain underappreciated. The first is how closely governments now link minerals to their own security.

“Minerals are no longer just commodities. They are now tools of statecraft,” he said. “From price floors to trade wars, tariffs to investments, when you equate mineral security with national security, the game fundamentally changes.”

Dhawan said the other is the public-goods view of metals and minerals, which sits beneath the trade disputes.

“Behind the headlines, a deeper shift has occurred that we must recognize for what it is, name it and act accordingly,” he said.

The second of his three big shifts is on the supply side. For generations, the industry’s fortunes rose and fell with the economic cycle, with demand setting the pace. Dhawan said that holds less true today.

“If mining was a circus, then demand has been our ringmaster. Our fortunes have waxed and waned over generations to the tune of the economic cycle,” he said. “By no means are we immune to what changes on the demand side, but what happens on the supply side now has a much bigger impact on our fortunes than it ever has before.”

He said several pressures now compound one another to create a real tension point for supply:

• Long permitting times for new mines

• Environmental constraints

• Social complexity

• Deeper ores that are harder to reach

Copper stockout fears

Dhawan said nowhere is this truer than in copper.

On October 11, heavy rain triggered a landslide at Codelco’s El Teniente mine in the Andes, about 75 km southeast of Santiago, killing two contractor workers and forcing a partial suspension of operations.

The century-old mine, one of Chile’s largest, had already suffered a collapse that killed six workers in July 2025.

“Other operations are offline or curtailed because of strikes, extreme weather, mudslides or mud rushes, and combined they have created the almost unthinkable scenario of a physical stockout of copper,” he said.

“US tariffs haven’t helped by pulling metal into the US, which may not be exported back anytime soon. But this just shows you how important the supply side is. We need to rise to the challenge and meet this moment by investing in resilient and, most importantly, safe production.”

The third shift is the industry’s reputation. Dhawan said surveys show public understanding and perception of mining have kept improving over the past five years, citing other signs of progress:

• Demand for mining jobs has soared

• Countries are competing to bring mining onto their shores

• Generalist investors now see the sector as investable

“This has been our Achilles’ heel for as long as I can remember, but we have turned a corner across the world. This has been very hard won. So we mustn’t avoid taking stock and celebrating the success we’ve had,” he said.

“But there is no time for complacency because this could be very short-lived unless we continue to tell our story, continue to be in open dialogue with people and, most importantly, continue to demonstrate safe and responsible production.”

Dhawan will step down at the end of the year and hand over to Froydis Cameron-Johansson, chosen by the 26 chief executives who make up ICMM. He said her 30-year career in natural resources makes her well placed to lead ICMM and the industry into its next era.

Cameron-Johansson has held executive roles at Rio Tinto, Anglo American and Shell, covering health, safety and security, sustainability and geopolitics. Her career spans three decades in both emerging and established markets.

“The job of the ICMM CEO (chief executive officer) is a genuine privilege. You get a ringside seat to the action and a bird’s-eye view of the industry,” he said.

He said the next era will be defined by the industry’s adoption of a common global mining framework, and he hoped people would remember mining not only for what it took from the earth but for what it gave back and the trust it built along the way.

Thin margins push logistics firms to shift the risk of robot investment onto suppliers, leaving them exposed when a vendor disappears or drops support for older machines.

Physical artificial intelligence (AI) is making warehouse robots faster to test and reprogram. But early returns are often low, as companies must first change processes, upskill workers, and reassure staff, which drains energy and resources. Economists call this pattern the productivity J-curve. For logistics operators, the question is who carries that early cost.

“Especially in logistics, our margins are thin, so we’re risk-averse,” said Arjan van Baarle, global head of contract logistics technology adoption and transformation at Kuehne+Nagel. “We try to bring the risk back to the supplier.”

He said the company always wants a back-to-back arrangement. When it signs a customer contract, Kuehne+Nagel invests in the technology and matches the supplier agreement to exactly the same length, including maintenance.

“It also means for the supplier that they need to go all in. Sometimes after three years the supplier is gone, and then we still need to work with the equipment, and they’re not there anymore,” he said.

He said some suppliers decide to upgrade their technology and then stop supporting the older systems still running in customers’ warehouses. At the same time, AI is changing how quickly his team can test robots and fix problems once they are installed.

“With AI entering into the equation, we can be much faster in testing,” he said. “We can be much faster in reprogramming, but that also means that if something goes wrong, we are much faster at adjusting at the right moment in time. It will help us tremendously in bringing more of those solutions live.”

Kuehne+Nagel was founded in Bremen, Germany, in 1890 and is headquartered in Schindellegi, Switzerland. It runs sea, air, road and contract logistics businesses in more than 100 countries. Its contract logistics arm provides warehousing and distribution services.

Don’t be the first

Van Baarle spoke at a fireside chat moderated by Eddie Milev, editorial lead of Tech Frontiers at Economist Enterprise, during the Economist Enterprise Physical AI & Robotics Summit in London on October 13. The session focused on building a robotics strategy that survives beyond the pilot stage.

Asked what advice he would give to companies that try to scale robotics too quickly, van Baarle offered three recommendations:

• Stress-test the equipment

• Look at the supplier’s strategy, structure and culture, not just the device

• Check whether the supplier already works with system integrators and get their feedback

“The first advice is to test the hell out of it. Try to do all the crazy stuff that you’re officially not allowed to,” he said. “Try to jump in front of the AGV (automated guided vehicle) or the AMR (autonomous mobile robot). See if it really stops. Understand how it works.”

“Not only look at the device itself, but also look at the company’s strategy, structure and culture behind it, because that will always reflect what you see in the product itself,” he said.

The company sees its role in the industry as safeguarding its customers.

“If it’s a Chinese company new in the market and does not have anything running outside China, then don’t be the first. I want to be the second, not the first,” he said.

Van Baarle runs teams in Singapore, the US, and Europe, and compares lessons across regions, including what China's progress in physical AI can teach. Robot suppliers in China offer different products at different prices from those used in Europe because of differences in regulation and safety rules.

“China is still a little bit ahead technology-wise, but behind on safety and those kinds of things,” he said. “By bringing this team together, we have the opportunity to look further in the future. It will give us a much more stable view, but also gives us the opportunity to translate to our customers what works and what doesn’t work in what area and what region.”

Robots moved into corners

Technology is only half of the challenge. Five years ago, the company tested a cleaning robot at a site in Canada that stopped working every night at around midnight or 1 am. A camera installed to find out why showed that forklift truck drivers, afraid for their jobs, were moving it into a corner.

“Change management is important. Make sure you bring the people along the journey and let them understand where it works and where it doesn’t work,” van Baarle said.

He said companies must also take away workers’ fear of losing their jobs, but middle management is the second hurdle. In his experience, frontline staff often adapt more easily than the managers above them, who worry more about their jobs or about no longer being seen as the knowledge center.

“Your middle management layer needs to think differently, work differently and adapt to the new technology,” he said.

On the technology side, warehouse systems have shifted to goods-to-person models that store bins, cartons and pallets and bring them to workers. But the picking itself is still not easy. For the past five years, the group has worked with vision systems to pick items from one carton and place them in another.

“Today we still have 99.4% or 99.5% precision, but that still means that if you pick 500 or 600 per hour, you still have two items falling down,” he said. “The minute we hit this one, this portion is going to be fully automated.”

A person still has to step in to help the robot. He said a generalizable vision model is already much easier to deploy across different enterprise functions, and automating other repetitive tasks will make a huge difference.

“It’s picking, but it’s also palletizing and unloading a truck. A lot of those repetitive tasks where humans still make a difference because we still understand how to palletize a pallet right with all those different carton sizes,” he said.

In the past, the company brought automation only to its biggest and bulkiest warehouses.

Cheaper mobile robots, easier to deploy and more scalable, are now bringing automation to mid-sized customers. Low-cost Chinese robots are spreading fast in e-commerce bin picking and carton movement, a trend he expects to continue, while pallet handling for fast-moving consumer goods (FMCG) firms remains largely old-fashioned.

“If I look at a pipeline three years ago, I saw only a couple of projects with big numbers on it. Today our pipeline looks completely different,” van Baarle said.

Kuehne+Nagel now has a much longer list of smaller automation projects for many more customers, rather than a handful of large ones. He expects the shift to let the company commoditize automation and grow even faster.

ABB Robotics, the robot-making arm of Switzerland-based ABB, is turning simulation into a lab for what-if experiments, from evening glare to dust and scratched parts.

Simulation once mainly checked that systems were ready to go live. Now it generates cheaper synthetic data.

“Synthetic datasets are not going to replace the real deployment, but if we reduce the time a robot needs to spend in the real world, that’s profitability,” said Boris Fiedler, head of operations and portfolio for software and digital at ABB Robotics. “It reduces risk and increases the payoff for the customer.”

Fiedler said no simulation perfectly copies the real world, at least not yet.

He said the industry is shifting toward using simulation to run experiments, such as testing how a vision system copes with evening sun. The resulting synthetic data, including pictures, videos and virtual robot forces, feeds the neural networks behind physical artificial intelligence (AI).

He said that when a deployment at a large fashion-industry customer goes wrong, ABB feeds the real data back into the original simulation. It then builds a richer set of synthetic variations around that single failure to train better AI.

“If you invest in simulation early and only think about the initial deployment, you’re kind of wrong because the system keeps working,” he said. “You will keep having a payoff every time there’s a new situation. There is a continuous payoff from having really good simulation on your systems.”

The technology now cuts across industries, from construction, where metal is bent and joined, to laboratory automation in life sciences. Lab tasks such as pipetting are simple, but one robot must handle many of them.

SoftBank Group agreed in October 2025 to buy ABB’s robotics business for $5.375 billion.

The unstructured frontier

Fiedler spoke at a fireside chat on digital twins and the gap between simulation and real life at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. Laveena Iyer, a senior analyst at the Economist Intelligence Unit, moderated the session.

Fiedler cited kitting, where one worker handles a chain of tasks:

• Opens the box

• Removes the cushions

• Takes out a part

• Inspects it for defects

• Puts it away

“Unstructured task is the frontier. Even though the environment is structured, the task itself is very dense, which is a massive challenge for automation,” he said.

The next challenge comes after deployment, when conditions keep changing, such as dust, a new stock-keeping unit (SKU) or defective parts. He said teleoperation, in which humans control robots remotely, is one way to close the gap, while ABB also uses its synthetic world to create more cases and more variety.

“I don’t think there’s one single answer, but the different technologies together will help to get it closer to generating the three nines in the end for the customer,” he said.

“Three nines” means 99.9% reliability. He said physical AI makes projects less predictable than traditional automation.

“We need to understand it more deeply. We need to work closer with our customers on certain problems, because problems and processes which we thought couldn’t be solved before can be solved today,” he said.

Consumer electronics poses a different test. Fiedler said the sector is now as heavily automated as automotive, yet some steps remain manual. ABB is working with Foxconn, the Taiwanese contract manufacturer behind many of the world’s smartphones, at one of its electronics manufacturing services (EMS) sites to raise yields further.

Tiny parts must meet extremely high quality standards, and product cycles are short. Once manufacturers finalize a new phone or tablet design, they usually have only about three months before launch to scale up the entire production line.

“If you have your phone in your hand, you want to have it perfect,” he said. “This is where simulation is so powerful, because you can do tons of this validation early in the process, so that the time you need to ramp up the production volumes is decreased significantly.”

Asia dominates robot installations

Global industrial robot installations rose 11% to a record 603,000 units in 2025, according to the International Federation of Robotics (IFR). China accounted for 59% of the total, and Asia for about three-quarters; installations fell 11% in the European Union (EU) and rose 12% in the US. Automotive installations in the EU fell 25% to 14,900 units.

“Why is Europe weak? Because automotive is weak, and there’s a huge transformation around the tier ones [suppliers], which struggle,” Fiedler said.

He said Europe remains strong in healthcare and other growing sectors that may eventually overtake automotive. Flexible physical AI systems are also growing fast, though it is early for this to show in the numbers. Random bin picking, in which robots pick randomly piled parts from a bin, has grown about 20% over the years.

“When you look into the industry, where growth is and what kind of application is equally important as the regions themselves, because that will tell you a lot about where investment is being directed,” he said.

He said scaling would remain a hurdle even if the gap between simulation and reality were largely closed.

“If we bring those [research systems] into pilots or production, it’s PhDs who are doing the work, and that’s not scalable,” he said. “The next big challenge is how to make this technology accessible to a robot engineer who is used to running around with a big screwdriver and bolting the metal to the ground.”

That means packaging autonomous systems so people who know the factory process, but aren't AI or software specialists, can deploy them.

His advice to companies planning a rollout is not to copy exactly how a human does a task, which he noted sounds counterintuitive, since autonomous systems are often expected to mimic humans. The process may not need as much redesign as in the past, because today’s systems adapt more.

“Very often that might not be the most efficient thing to do. Maybe if you change the process slightly, you can generate a much bigger return and have a much more effective production system,” he said.

His second piece of advice was to be open to agentic frameworks and “agentic engineering.”

“A simulation experimentation engine is the ideal tool for agentic engineering because the agents have something to work against, to test and to iterate quickly,” he said.

These tools aim to get companies into real-world testing faster.

Many successful physical artificial intelligence (AI) pilots fail at scale because the networks and data behind the robots cannot keep up.

Factories are adding sensors, control systems, and video analytics to networks built for an earlier era, while cloud-trained models must run at the edge, where delays and security gaps carry real costs. Without clean, contextualized and time-stamped data, these systems cannot correlate what is happening on the production line.

“The main question is not if AI technology is scalable. We know it is. The basic question is if the companies have the operational infrastructure required to scale it up,” said Germán Fernández, vice president of global strategy and ecosystems at Belden.

“The first assumption we need to break down is that the network that you built 10 years ago is going to support the new traffic that is required for physical AI or agentic AI,” he said.

Fernández said many pilots fail at scale for three reasons:

• There is no solid, resilient and secure network

• Companies lack real-time, clean and contextualized data

• No governance model integrates different technology providers with clear accountability

He said companies typically underestimate the complexity, and the biggest mistake is to treat each deployment as a separate project, creating fragmented architectures that are hard to maintain and manage.

“We are moving from systems of insights that were connected to ERPs (enterprise resource planning systems), MES (manufacturing execution systems), etc. to systems of action, and here the role of the network changes completely. This will be a foundational layer of the autonomous decision-making process,” Fernández said.

Aaron Boasman-Patel, vice president of innovation at TM Forum, a global telecoms standards body, said most deployment plans overlook the networks that power robots.

“Get your security right, get your network right. If you get those two wrong, you get carried away with the fancy robotics, you’re on a pathway to failure,” he said.

He said companies must first decide which data is processed in the cloud and which on a core server, as latency becomes critical and compute grows expensive. He added that hacking has become one of the biggest forms of terrorism, with attackers able to switch off hospital systems.

Ahmet Hasanbeseoglu, chief value partner for digital transformation in supply chain at Arla Foods, said a fleet of robots can orchestrate well and still miss its expected throughput without a strong data platform behind it.

“The challenge is not about orchestrating the physical robot, but building this data knowledge around that, so we can program the factory. That’s our biggest challenge,” he said.

He said companies should treat “soft AI” and “hard AI” together, linked by a “digital thread,” so a maintenance agent reading the next service cycle from SAP can trigger maintenance robots or workers on the shop floor.

Traceability over dashboards

The panel, titled “Increasing autonomy without losing control,” took place at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. Laveena Iyer, a senior analyst at the Economist Intelligence Unit, moderated the session on how companies can let AI systems adapt while keeping safeguards and human oversight.

Tom Hummel, robotics technical director at Ford Motor Company, said leading indicators are hard to track in autonomous systems, and quality problems usually surface only in a failed downline test.

“The most important thing that we look for at Ford is traceability, and that’s something that we are keenly aware of when we think about our physical AI deployments,” he said.

“If we’re building an engine, say the bearings were improperly installed, that engine is going to blow up on the test stand when we do its first spin, and all of a sudden that’s a rather large amount of value and time that’s wasted.”

More insidious faults show up later as warranty claims. He said real-time monitoring is valuable but very difficult, so Ford needs traceability from autonomous systems.

Boasman-Patel said the same applies to AI agents. Through agent-to-agent (A2A) communication, companies must be able to see which agent made each decision and when, with trust and safety guardrails built into every step. He had just returned from a TM Forum hackathon in Dallas with about 250 telecoms developers.

The panel also turned to who answers for a robot’s mistake.

Sara Filipčić, executive director and founder of the Institute for Humane Robotics, said more than 600 industry conversations across Japan, China, Korea, the US, and Europe revealed a reactive mindset.

“A lot of robotics companies are almost like waiting for a problem to emerge before they start creating the solution for that,” she said.

She urged companies to settle three questions in advance:

• Who decides what happens with the robot

• Who has the agency to stop its actions during a malfunction

• How to learn from the incident

“Who is responsible if something goes wrong? Is it the person who sourced the data? Is it the person who built the bot or the AI algorithm?” Boasman-Patel said. “That governance isn’t some kind of ethereal thing that sits on paper. It sits with people because people will go to jail over this.”

He said the industry is not taking enough accountability for who is responsible and what happens to those people when something goes wrong.

Every 15 seconds

Ford builds an F-150 truck every 15 seconds in North America, so even minutes of downtime are costly, Hummel said, adding that robots must be easy to service, with constant teleoperation or a support worker on call. Ford’s own teams act first, followed by external service support.

“Ford takes an approach where we want our skill teams at the plants to know how to fix and repair the robot as fast as possible,” he said.

Task design is another concern. Many physical AI systems target brownfield tasks that already exist, and he questioned whether those lines could still hit the correct takt time if the robot were removed. He said companies should judge each use case by its total value over the deployment's life.

“In many cases, in what I’ve seen in real-world deployments, a lot of times the task is modified for the robot at the expense of the human,” Hummel said.

“It’s very easy with physical AI, especially with humanoids or mobile manipulation, to say, ‘Hey, we can do it.’ There are very few times we really ask, even at a large organization like Ford, should we do it?”

At Arla Foods, robots are rolled out with safety as the top priority. Hasanbeseoglu said the dairy cooperative runs safety design tests during proof-of-concept and pilot phases because staff work alongside collaborative robots (cobots).

“The human is always there checking if the robot is acting correctly. Ideally, when we go to the future, these systems should be more and more human in the loop,” he said.

In that model, systems would alert a maintenance or site manager only when something unusual happens. He said governance must cover a mixed workforce of humans, “augmented” humans and robots, and that staff affected by automation should be reskilled into roles such as observers and supervisors.

“We have fewer and fewer people who want to work in a factory. We are somehow going into this dark factory mode,” Hasanbeseoglu said.

Filipčić said robotics companies must also understand their customers’ customers. In a hospital, staff and patients interact with robots most, so they should help co-design the technology.

Boasman-Patel said common interfaces, such as open application programming interfaces (APIs), are needed so every device shares the same data. He cited the fanfare around Waymo’s driverless cars in London, which depend on surrounding sensors and traffic lights sharing the same data.

“All of those devices must have the same data, and if they don’t, what happens? The car just stops and everybody else drives into it,” he said.

He said standards must also cover a common “expression of intent,” the way users now tell tools such as ChatGPT what they want instead of writing lines of code. When intent is expressed differently, things start to go awry.

Hasanbeseoglu said accountability, safety and cybersecurity are fundamentals that must be in place before Arla takes automation further.

Orbiting artificial intelligence (AI) data centers may sound like an easy escape from water-hungry cooling on Earth, but their heat has no easy way out.

A one-megawatt facility, about the size of a modest university computing cluster, would need more than 3,000 square meters of solar panels. That is roughly the size of the International Space Station (ISS) array, the largest ever launched, which took 23 years to assemble.

“This is an unfortunate thing that Hollywood has done for us. It’s so easy to imagine launching some gigantic thing up into Earth orbit, and it’s cold in space, so this is great,” said Lindy Elkins-Tanton, principal investigator of NASA's Psyche mission and director of the Space Sciences Laboratory at the University of California, Berkeley.

She said space offers no air or water to carry heat away, so all the collected energy must radiate back out into a vacuum. Each square centimeter of chip produces about 100 watts of heat, while a radiator of the same size sheds only about 0.1 watts.

“You need 1,000 times more area of radiators than you have chips, and so on the face of it, the energy balance doesn’t close,” she said. “I know that there are groups out there that claim they’ve figured it out, and I’m curious.”

She would love to see it work but said it needs more technology development than is publicly known.

Earth-orbiting communications satellites carry only 100 to 200 square meters of solar arrays, which makes a 3,000-square-meter array for a single megawatt enormous by comparison.

Elkins-Tanton has led Psyche, a NASA mission to a metal-rich asteroid of the same name, since 2011. The spacecraft launched on a SpaceX Falcon Heavy on October 13, 2023, and is now about halfway to its target. Its solar arrays cover 75 square meters, the second largest ever sent into deep space.

Chips generations behind

Elkins-Tanton spoke at a fireside chat moderated by Alex Hern, an AI writer at The Economist, at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. The session explored where the intelligence that runs robots should live.

She said the bigger barrier, which also affects low Earth orbit (LEO) and space data centers, is hardware.

“The radiation-hardened chips that we need to use in space are generations behind what we’re all using on the ground right now, and so there’s a real hardware limitation,” she said.

Cosmic rays add another hazard. A single-event upset (SEU) occurs when one strikes a chip and wipes out a function or piece of data. Earth’s magnetic field and atmosphere largely shield chips on the ground, but even extra hardening cannot fully protect them in space.

Science spacecraft are often bespoke, built once for a single use.

“Risk tolerance is a big deal. If we need to spend six years building this spacecraft and then we’re going to fly it for 15 years, it has to work because it cannot be repaired after it’s launched,” Elkins-Tanton said.

Psyche runs no generative AI, only older forms of machine learning long used in space. Its software was tested for years under every possible circumstance, with fault monitors that scale back the spacecraft’s functions if something goes wrong. Everything has to be fully deterministic.

“A lot of the generative AI that we’d want to run on board is non-deterministic, so you’re not quite sure what answer it’s going to come out with, and that doesn’t work so perfectly for spacecraft,” she said. “So you have to have some kind of post-processor that’s going to put guardrails up.”

AstroForge, an early-stage space mining company she advises, is developing AI-run autonomous spacecraft for Earth orbit.

Security favors keeping decisions on board, and she worries about snooping and tampered commands at all times.

“Any time that you’re communicating with something using radio waves or using laser, you have the possibility of interception. You have the possibility of spoofing,” she said. “It’s a huge security advantage if you can do all your decision-making on board, and none of it goes off and comes back on.”

Laser links are somewhat more secure than radio, which anyone can hear, because you can limit their geographic spread. A NASA laser communications demonstration aboard Psyche sent a cat video in 2023 from about 80 times farther away than the Moon.

“It saddens me because this is not the world that I would want to be in, but space is absolutely a place of nationalism now, and a proxy for war on Earth,” she said.

She said anti-satellite tests have already happened, and the threat that debris clouds from broken-up satellites could make launches difficult is real, though not as imminent as people say. She called such threats the biggest the sector faces.

Ground in the loop

Latency adds a further case for onboard intelligence. Psyche is about 15 light-minutes away, so a command and its reply take at least half an hour, and contact is impossible when the spacecraft is on the far side of the Sun.

“There’s no business case for anything in space that closes outside of Earth orbit. Earth orbit is where business happens in space, and latency is still a problem with that,” she said.

She said delays to LEO become undetectable once ground dishes cover the entire planet.

Collision avoidance is one of the biggest challenges. SpaceX has effectively been building a distributed compute facility in orbit, and thousands of satellites could soon run AI and make their own orbit decisions.

“It sounds great until you realize that the different constellations probably don’t talk to each other. Everyone else’s satellites won’t necessarily know what’s going on,” she said.

Satellites in the same constellation could coordinate orbit changes, but rivals would be left in the dark. That would force “ground in the loop” decision-making, with signals routed down to Earth and back up.

She sees more promise in onboard AI for science, where communications back to Earth are the bottleneck.

“What if you had a scientist on board who could look at the data that’s being gathered and say, let’s condense it in the following way?” she said.

Discovery often depends on serendipity and on what an expert knows to be unusual. Peer-reviewed studies have found that expert scientists watching a remote robot’s video feeds were often wrong about what they saw, so more data must be interpreted when no one is there.

“If you get a nicely trained AI on board, getting the information and changing what information is coming in, maybe some decisions can be made faster and more efficiently,” Elkins-Tanton said. “Then the data can be condensed and downloaded more efficiently.”

She said the stakes in planetary science are lower because no lives are at risk, unlike AI in drug production, where the human risks are gigantic.

Psyche is due to reach its target in 2029 and orbit it for 26 months. No photos of the asteroid exist yet. It will be the first visit to a world with a metal surface, and NASA's Deep Space Network will post unedited images within 30 minutes of arrival.

“We don’t edit them or censor them because we want to share with everyone in the world,” she said. “My dearest hope is that in the 100 days as we make our approach to the asteroid, all of you will be looking at the pictures with us, and we’ll be able to say this is something unusual and interesting.”

She expects the mission to reveal something about the materials that made Earth, which could help explain why Earth is habitable while Venus and Mars are not, and could point to economically valuable materials on smaller, Psyche-like asteroids that cross Earth’s orbit.

Ocado Group, a Hatfield-based grocery technology firm, is finding that a robot error too rare to notice in a lab becomes a weekly problem at scale.

Its robotic arms now pick about 10 million grocery items a week. A drop rate of one in 100,000 sounds like a rounding error, but it means around 100 falling items a week, each able to halt a bot, summon an engineer and dent uptime.

“At scale, a one-in-a-million edge case is not an edge case. You have to design your system and your processes assuming that the worst will happen,” said Matt Whelan, vice-president of engineering for automation at Ocado.

“Making the robot do the thing is only a small part of the problem. The big part is everything around it and the costs it can quietly push into different parts of your operation if you’re not paying attention.”

Whelan said physical artificial intelligence (AI) suddenly seems to be moving very quickly, and many people, himself included, feel a “ChatGPT moment” for robots is close. He expects many organizations to buy or build robots that look brilliant in a demo, only to discover how hard they are to run at scale.

Internally, Ocado uses the term “serviceable operational range” for what an arm can reliably pick in a real environment without causing problems. Speed and total addressable range, the obvious questions a buyer asks, miss what matters.

“You can get your robot metrics looking great, but the cost has moved somewhere else in the system. I’ll leave you to imagine the impact of spilling coconut oil on the grid. We know from experience,” he said.

The grid stores crates eight meters high and is served by washing machine-sized bots that can reach any stack.

Since deploying its first arm at the end of 2023, Ocado has gathered a growing dataset of edge cases and failures, along with human demonstrations of exemplary and recovery behavior. It uses the data to train and refine the arm’s controller.

Ocado was founded in 2000 as an online grocer whose warehouse staff pushed trolleys for miles down aisles. It now runs its automated system as a platform for retail partners.

The £5 double pick

Whelan spoke at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. His case-study session asked how companies can tell whether automation adds value or merely shifts costs from one part of a business to another.

A double pick, where a suction cup lifts two items instead of one, still counts as a robot success.

“Its KPIs (key performance indicators) are looking fantastic. If you’re in a grocery business with very low margins, you save a few pence per pick on everything you do successfully, then you give away a £5 item. You’ve undone all the value of hundreds of picks with one mistake,” Whelan said.

Because Ocado runs a platform business, it must weigh the arm’s impact on its partners’ economics as well as its own. Its costs include robotic damage, reliability hits when items land on the grid, arm repairs and remote support.

“Your robot labor savings sits on one side of the ledger, and everything else sits on the other. I improved the range that my robot could pick. My double-pick rate went up, and my overall P&L (profit and loss) got worse,” he said.

Whelan said companies should map everything their automation touches upstream and downstream, measure success across the whole operation and, whenever one metric improves, find out what quietly got worse.

Ocado’s task is relatively constrained: the same box size and the same goal every time, pick an item and place it carefully in a customer’s bag. But it unfolds in a complex industrial setting, with messy real-world challenges, in a low-margin industry.

“We are not trying to solve a fully generalized robotics problem. We’re trying to do one hard thing extremely well in a punishing industry,” he said.

Keeping the arms running is a job in itself. Ocado supports every arm worldwide from a single center, a 24/7 operation spanning time zones, peaks, troughs, and shift patterns, and its staff must be able to take remote control when a robot misbehaves.

“There is no such thing as a 100% success rate in hardware,” he said.

Range management, deciding which items a robot should touch, is another moving target as suppliers change packaging without warning.

Releases in harm’s way

Whelan’s answer is to learn in production. Ocado rolls changes out to a subset of arms, A/B tests them, deploys them gradually across sites and rolls them back if needed, with safeguards to match.

“You cannot find the one-in-a-million edge cases in a lab. You have to get into the real, chaotic world as early as possible, and that means your latest release is always in harm’s way. When you have hundreds of arms deployed, you can learn fast without putting the operation at risk,” he said.

He said this is how companies can move fast and break nothing that matters, even though breaking nothing at all is impossible in the physical world. He also urged a relentless focus on total cost of ownership until the system crosses its break-even line.

Beyond that point, better capability and growing scale can compound. Whelan said Ocado’s arms are approaching half a billion picks and should pass one billion around this time next year. For partners, the system cuts the labor needed to pick and pack a grocery order by about half.

The robotic picking system now runs for retailers in eight of Ocado’s markets, including the UK, the US, Japan and South Korea, and the company aims to push that labor saving higher.

Whelan said AI is moving robotics from traditional systems toward more generalized, more capable automation, and that getting a few aluminum tubes with actuators at the corners to perform a diverse task well is extraordinary. Such systems also need warehouse integration, remote-control interfaces and cloud infrastructure.

“It’s not just the control model either. It’s the perception layers on top that look for errors and intervene in the robot control without needing a human in the loop,” he said.

Classical robotics, such as the cells that weld Ocado’s metal totes, makes high-value goods and enjoys forgiving economics.

“You could build a super impressive system and never make a penny from it if you start with wrong assumptions. The robot that used to fit a car door is now trying to do something trivial at home. We are looking for much lower-value tasks at much higher volume,” he said.

He said grocery offers a preview of the problems the wider industry will face as physical AI spreads into everyday tasks, where every double pick and minute of remote intervention will be felt.

Companies deploying humanoid robots in warehouses and factories still haven't decided who owns the data those machines generate or the smarter robots they produce.

As pilots turn into full deployments, data captured when robots stumble has become the most valued training material.

“I don’t believe the entire industry has figured out who owns the data. Is it the customer, the operators or the robotic company?” said Nizar Trigui, chief technology officer of GXO.

“Right now we are willing to share data because we have an interest that the model is working well for us. But at some point in time, we want to be in a place that we can use our data to improve the model further for us,” said Patrick Lanners, innovation and AI lead at BMW Group. “Otherwise, we are just training the model for all our competitors as well.”

Trigui said GXO has deployed humanoids at seven sites for more than two years, working mainly with startups. Under current contracts, customers own their data. He said the best training material comes when the model is uncertain and a person teaches it.

“You teach one robot, but the entire family is learning at the same time. So it’s highly valuable,” he said. “As we move from proof of value into full deployment, we need to figure this part out and understand who owns the data.”

GXO plans to announce that move to full deployment by the end of November.

Lanners said he often hears debate over how much data BMW should share with frontier model providers. The more manufacturing data these models contain, the easier it is for BMW to adapt them, but the point at which it stops sharing has yet to be defined.

“If we are not able to use our data for our own purposes and we’re just training for everyone else as well, the question is really where’s our competitive advantage? Where can we be better in building cars than everyone else?” he said.

Christina Yan Zhang, chief executive of The Metaverse Institute, said nobody yet knows whether model providers will later use shared robot data for other commercial purposes. She said some enterprises are responding by building their own specialized small language models.

Open-source world models offer another route, such as Meta’s 8-billion-parameter RoboJEPA, released on October 8.

“Maybe the future is all about how we can use open-source models to provide support on the very specialized industrial robotic data we are training to serve our industry better. So we don’t need to worry too much about the data’s final ownership,” she said.

“The software and the model without the experience and the data are useless. The combination of both gives you only then the real power,” said Radosław Nowak, managing director for fast-moving consumer goods and supply chain at Greenyard.

Outcomes nobody recorded

The comments came at the Economist Enterprise Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. Alex Hern, AI writer at The Economist, moderated a panel on how to gather data good enough to train robots.

Lanners said BMW began work in 2025 on vision-language-action models, but the data its factory robots already produce is of little use for training them.

Trigui said crushing a box, or packing an item so the lid won’t fit, counts as a failure.

“Robots are not going to learn from visual; they need to experience it. So you need the context, and more importantly, you need the outcome. The outcome is the part that’s very difficult that we’ve never collected before,” he said.

Nowak said companies spent years collecting information-flow data, such as forecasts, demand and point-of-sale figures, which is already well automated. Robots instead need to know how much force to press a box, and no company system stores that.

Sertac Güzel, head of robotics, digital and specialty at Johnson & Johnson MedTech, said surgeons still learn by seeing one, doing one and teaching one, and AI can shorten those learning curves by identifying anatomy and tasks.

“There is still a lot of transferable know-how. When you think of surgery, you can think of it as a very underdeveloped space from a technological adoption standpoint,” he said.

Trigui said the best results come from combining data types, an approach borrowed from autonomous-vehicle programs:

• Real-world data comes first

• Teleoperation then lets humans handle exceptions

• Synthetic data fills in permutations nobody has observed

“You can build those synthetically, but the basis has to be the on-the-floor learning and understanding the perception and the outcome,” he said.

Güzel said surgical teams can study the operating room, patient anatomy and different approaches long before choosing any hardware.

“You could create the digital twin of the OR (operating room) and the procedure that’s going to be able to optimize the interventional approach for the robotics. So all of this you can do without a robot in the room,” he said.

Lanners said robot-collected data, from teleoperation and offline or online reinforcement learning, gives models the clearest signals, while egocentric and simulation data may help close the gap.

“We all learn right now what is the right amount of data and what are the different types of data. We haven’t figured it out fully, because if so, we would have these already in production,” he said.

Zhang said the largest open robot dataset, built by Google DeepMind with dozens of institutions, holds about 1 million episodes, while text models train on tens of trillions of tokens.

She said training costs for robot data have fallen from about $300 an hour to about $100 in two years.

But she said off-the-shelf generated data does little for regulated sectors such as healthcare, automotive and defense, which need physics-based data that complies with their rules.

Sorting avocados by algorithm

Greenyard buys and trades fruit and vegetables worldwide from growers with different standards.

Nowak said its sorting machines photograph each fruit 80 to 100 times and compare it with a reference image for mold, rot and color. Retailers demand different avocado shades, and the images settle disputes with quality-control staff.

“Imagine that you have a new employee, or you do this manually. It’s very subjective,” he said. “The learning is happening already after a few hours. I think the software can beat very experienced people who are doing this for many years.”

Car making sits at the other extreme. Lanners said BMW’s paint and body shops are already highly automated, but assembly and logistics are not, because the tasks are too complex.

“Where your part is not always in the same position, typically old classic automation fails. There we see still a lot of potential,” he said.

Güzel said sharing should rise in levels:

• Common standards and language at the bottom

• Data shared with customers and developers

• Data shared across industry players

• Shared models

“If you truly want to make a difference, then you need to be bold about this sharing idea,” he said. “We are sitting down with other industry players to at least align on the standards, which will leapfrog innovation.”

Trigui said sharing will start with common ground such as taxonomy and safety.

“The operational specifics for manufacturing and logistics, that’s the IP (intellectual property) that each company has, and that’s our comparative advantage. So I doubt that there will be a marketplace for that,” he said.

Nowak compared future robot standards to the EuroPallet, whose fixed dimensions make logistics easier for everyone, and said robots will need common standards to communicate with each other.

Zhang said the robotics industry needs a minimal interoperable mechanism, similar to the standards set by the International Telecommunication Union, that defines the minimum data automotive, healthcare, logistics and defense firms can share. She said that would move the industry forward and lower training costs.

The panel also weighed whether the firm with the most data will simply build the best model.

“Why is Reddit the most important data that helped the LLMs (large language models)? It’s that little thumbs up or down, because that gives you the feedback mechanism and gives you an outcome of people accepting those results or not,” Trigui said.

He said task-specific models fitted to each operation will scale first, before the industry eventually relies on generic models with broad perception and execution skills.

A British maker of expressive humanoid robots is building its first walking model, but it won't include its most sophisticated feature: the face.

The robot, due to be shown by the end of this year, will be smaller than a full-size humanoid and carry a simpler head than the expressive one on the company’s current machines. Very few legged humanoids are working in the real world yet, a point the summit’s previous panel also raised.

“We haven’t shown it off, and you guys are one of the first people to know that we’re doing that because I haven’t told anyone,” said Morgan Roe, chief operating officer of Engineered Arts.

“That’s not going to have the head with the facial features that we currently have because it’s going to fall on its face. That thing will fall down, and an expensive head with 32 motors in it is just going to get destroyed immediately,” he said.

Roe said the new robot will still be able to interact with people. It will stand 1.1 meters tall, and he has seen other humanoids that are shorter still.

When asked whether starting with the face made the route to a full humanoid harder, he said the company had approached the challenge differently. Very few markets exist for a robot with legs, while a desktop robot with just a head and neck can already serve front desks, care homes, and schools.

“We started with the face because that’s something that we can address right now. There are use cases for just a communication device,” he said.

Based in Falmouth, Cornwall, Engineered Arts has spent 20 years building humanoid robots that entertain, communicate and educate. Its customers include visitor attractions, science museums, university researchers and corporate labs that use its robots to test human-centric consumer devices such as virtual reality headsets and smart eyewear.

Small is safer

Roe spoke at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. Alex Hern, an artificial intelligence (AI) writer at The Economist, moderated the fireside chat on why human-robot interaction matters.

Homes pose a problem because a robot maker cannot control what is inside them, from fine china to small children.

“It’s an unconstrained environment. That is why it’s our vision, because it’s one of the most, if not the most, difficult areas to operate a humanoid robot in,” Roe said.

A full-size humanoid stands about 1.8 meters tall and, however it is built, will never weigh less than 35 to 40 kilograms, and even that is lightweight. When something goes wrong, it falls, and the task is to mitigate that risk.

“Something falling from that distance is a risk. The smaller the robot, the less the risk,” he said.

He said the company is always looking at the next steps it can take toward working in the home, and one of them is making a robot that acts, reacts and interacts like a human. Yet looking too human carries its own risks.

In 2017, Engineered Arts began making ultra-lifelike robots with painted skin and hair, and that's when it entered the uncanny valley.

“The uncanny valley is when something is uncannily like a human, but not quite. If it gets quite like a human but doesn’t move, act or react like a human, that’s when you get that sort of ‘This is scary. I don’t want to be around this,’” Roe said.

Over 20 years, the company has seen people warm to its robots and recoil from them.

“For us, for the engineers, we were looking at that, and we preferred it when we took the skin off and the hair off and the clothes off. It looked like a robot again,” he said.

When the pandemic shut down the businesses that bought its robots, the company used the time to build the character it wanted. The new robot had no hair and gray skin, so people could see the machine underneath and how it worked.

The robot, Ameca, drew tens of millions of YouTube views within weeks of its December 2021 debut.

“Some of the comments were, ‘This is the most lifelike thing I’ve ever seen.’ But what we did is we made it less lifelike, and what that did was make it more of a blank canvas, so people could project their idea of what a human is upon the robot,” he said.

Built to look robotic

Roe said a robot will never be exactly like a human, so it should act and react like one without trying to look like one.

“I don’t think trying to make it ultra lifelike is what we should be gunning for. It’s great in movies. That’s absolutely fine. We know that what we’re building is a robot. As a human interacting with that thing, I want to see that it’s a robot,” he said.

The human form is a high-bandwidth communication device that works through gesture and facial expression as well as speech, and everyone already knows how to use it.

Roe’s grandfather, born in 1922, lived to 98 and through the digital age, but never owned a computer or smartphone or used the internet. Learning to use the technology meant learning to live and communicate in a different way, with risks such as emails that should never be answered.

Faced with all that, his grandfather decided he did not want to take part.

“A humanoid robot is an incredibly complex machine, but if it says hello and greets you, my grandfather would be able to use that piece of technology because he’s used that since the moment he was born. It’s about opening up these complex technologies to the world,” he said.

Roe said the same human-centered thinking is showing up in AI agents such as Meta’s Muse, launched last month.

The personal agent works in the background, comes with a cute character and can buy products for its user through partners such as Shopify and PayPal.

“We believe that humanoid robots are going to be in the home. They’ll be everywhere. That humanoid robot in my home, it’s going to know what I buy. It’s going to know when I’m running out of loo roll, so it will buy it for me,” he said.

He said people will end up with personal AIs in physical form, and those assistants will carry into the workplace, as ChatGPT did after starting as a consumer product.

He cited a survey this year that found 89% of people who use AI at work first used it at home.

“The trust and the acceptance of robots is going to come from reducing the risk of physical harm, but also what the agent does. If you give an agent money, you want it to buy your loo roll. You don’t want it to buy a Lamborghini,” he said.

Engineered Arts plans to unveil its 1.1-meter walking robot by the end of this year, its next step toward a humanoid that can work safely in the home.

Robots deployed in shipyards and war zones are delivering patchy returns, as chaotic real-world conditions expose the gap between physical artificial intelligence (AI) hype and reality.

Practitioners’ estimates of how much deployed robotics is adding value range from about 15% to 50%.

“Soldiers, sailors and airmen will very quickly dispose of the robot if it’s not adding value to them and they can’t operate it quickly,” said Dan Anders-Brown, head of growth for advanced materials, electrification, autonomy and future systems at QinetiQ, a UK defense technology company.

“If it makes their life more complicated, they’ll fix the bayonet and they’ll do what they need to do rather than use some complicated system. Simplicity is key in my environment,” he said.

Defense robots face smoke, laser dazzle, obfuscation, bombs and people shooting at them on top of difficult terrain. QinetiQ has learned the cost of complexity through its robotics work in Ukraine.

Anders-Brown said only about half of the robotics deployed in defense works at the moment, with some adding no value at all. Autonomous vehicles running logistics resupply along a 2,000-kilometer road are among the success stories, as they take people out of vehicles and free them to go forward to the fight.

“You will find out very quickly how good your product is when you give it to a soldier, because they’ll either break it or they’ll find a different thing to do with it,” he said.

He said drone technology in Ukraine has advanced so far in three to five years that it has become a complete game changer, removing people from serious harm’s way.

Carlo Dentesano, head of robotics at the Digital Lab of Italian shipbuilder Fincantieri, said his company was at the beginning of the curve in getting returns from its robots. Asked to put a figure on it, he estimated 15% to 20%.

Fincantieri’s roadmap started with logistics robots, moved to quality monitoring and will eventually reach production tasks such as welding.

“Removing the man-hours is not how we intend to integrate robotics in our processes. We want the humans to collaborate alongside our robots,” he said.

He said Fincantieri wants its workers to build the skills to operate and collaborate with robots. Their knowledge of how the work is done would, in turn, help the machines become more efficient and adapt to edge cases that will always remain.

Robots in the nuclear sector are further from paying off.

Ertan Balaban, principal electronics engineer at the UK Atomic Energy Authority (UKAEA), said fusion is still a research effort rather than a reality, and robotics for the nuclear sector is itself a subject of research at a low technology readiness level.

“For the future fusion reactors, I can certainly say that robotics will be a big part of it. Most of the things have to be done by robots because of the hostility of the environment,” he said.

Pipe dreams of autonomy

The panelists spoke at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. Moderated by Christina Yan Zhang, chief executive of The Metaverse Institute, the session discussed how to keep robots running in variable environments and which hyped breakthroughs remain pipe dreams.

Dentesano doubts that one giant AI model will soon handle real factory work. He said robots should combine different ways of perceiving their environment rather than going directly from sensor to action.

“Having one single big VLA (vision-language-action) model like GPT style, based only on low-resolution images because they are heavy, will not deal with doing an actual job with the millimeter precision that we need in manufacturing. So you have to combine multiple systems,” he said.

“Things are changing and evolving very quickly, so I will be happy to be surprised. If we find out the ultimate model for AGI (artificial general intelligence) even in robotics, I will apply it immediately,” he said.

For Anders-Brown, the fantasy is senior commanders’ belief that robots will one day completely take soldiers out of harm’s way. He recalled seeing a drone in Kyiv a couple of weeks earlier that mistook a patch of grass for an obstacle, drove into a ditch and flipped over.

“Senior officers tend to think that some kind of advancement in AI is going to mean sentience. It’s just daft because they’re all being developed at about the same pace,” he said.

He compared it with the 1980s film WarGames, in which a computer concludes there is no point in fighting.

Balaban said fully autonomous robotic systems remain a distant prospect for super safety-critical settings, because the nuclear sector has zero tolerance for failure.

Fincantieri is studying flexible robot platforms, including humanoids, for its yards.

“You have to think about shipbuilding like metalworking and a construction site all together, a very chaotic environment. We have many phases of the process, many stages with different geometries and different environments,” Dentesano said.

He said designing a separate robot or mobile platform for each environment could leave the company with about 100 different robots within 10 years, each with its own maintenance, training and parts. Instead, it wants one platform flexible enough to work across all phases while keeping operations and maintenance scalable.

“We are mainly a team that acts as an integrator, so you have to find ways to solve complex problems and integrate different solutions from different vendors,” he said.

Start with the problem

Maria Clara Sayeg Ribeiro, who leads data and AI at the Centre for Advanced Manufacturing and Supply Chains of the World Economic Forum (WEF), said she would tell a plant manager with a budget for robots not to buy them just yet.

“The key question that we should ask is: we are acquiring robots and automation for what? What is the problem we are trying to solve? Is it productivity, quality or making the workplace safer?” she said.

She said the pace of new robots every year and new AI models every week has made companies forget to start with people. They should not go straight to the newest, shiniest model on the market.

She compared the gap between boardroom strategy and factory reality with Brazil’s 7-1 defeat by Germany at the 2014 World Cup, when the team had the best players and strategy. She said companies that fail to combine people, process and technology risk another 7-1 on the factory floor.

She said companies must then check that their factories have:

• The right data

• Sensors and perception

• Interconnected machines

• The right cybersecurity

The WEF’s Global Lighthouse Network has spent the past decade recognizing factories that escape “pilot purgatory” and scale up technology. Ribeiro said 94% of those lighthouses combine three technologies to reduce risk.

“It’s never only about physical AI or AI in isolation. It’s about combining and converging technology to make sure that you reduce this risk to the factory,” she said.

Anders-Brown said the UK’s Ministry of Defence doesn’t want to lead in every area of technology. It must lead where commercial industry has little presence, such as robotics for nuclear submarines, but elsewhere prefers to integrate ready-made commercial systems into complete solutions.

He said QinetiQ spends much of its time making commercial off-the-shelf systems work together, either as swarms or as independent systems. A recent one-way effector project fitted warheads and retrofit autonomy onto boats.

“The robotics will change. What we’re interested in is the ability to integrate new systems onto the sinew, the glue, the thing that holds it all together,” he said.

For fusion robots, the hurdle is the hardware itself.

“The problem with the current state of electronics is that it cannot withstand the high radiation level. This forces us to either explore new materials or find ways to protect our electronics,” Balaban said.

UKAEA works with universities on fundamental research into new materials and with startups on design techniques for radiation-hardened electronics. Where neither can do the job, it develops its own technology. For the past two years, it has been building a digital camera for its autonomous robots that can withstand one megagray of radiation.

The camera is designed to be integrated into UKAEA’s robotic systems, which will need to operate in the hostile environments of future fusion power plants.

Humanoid robots are entering factory floors, but no one can yet say who is liable when a robot assembled from parts from several suppliers hurts someone.

A single robot may combine one company’s body, another’s hand and a third firm’s artificial intelligence (AI) model.

“Let’s say the robot hits someone. Who takes charge? Is it the model? Is it the hardware? Nobody really knows,” said Aaron Rohmann, director of Atlas at MicroAGI, a Munich-based robotics startup. “That’s also stopping us at the moment from deploying quicker.”

“I couldn’t get a single company to come, primarily for insurance reasons,” said Richard Ahlfeld, senior vice president of physical AI at CoreWeave, recalling his attempt to bring live robot demos to the company’s latest summit. “So we ended up with pre-programmed robots that do the same things they’ve done for the last 30 years.”

He said the experience bore out Rohmann’s point. He had invited roughly 100 physical AI companies in CoreWeave’s network to see their robots at work on their own sites, and even the pre-programmed machines needed coverage. Insuring a robot for a single day is expensive.

Rohmann said the liability problem is built into how robots are assembled today. In MicroAGI’s own workshop, a Unitree robot is fitted with a Sharpa hand and runs on an Nvidia model.

He said this hardware is improving quickly, but much of it is still in prototype form.

“It’s a technology that is currently making big jumps, but it still has a very long tail. Before you get an autonomous car onto the street, you have years and years of validation ahead of you to make sure that it is reliable across a lot of different zones,” Ahlfeld said.

He said that is why it will still take a while before production use cases move forward.

The 99.7% question

Ahlfeld interviewed Rohmann at the Physical AI and Robotics Summit, organized by Economist Enterprise in London on October 13. Their session, supported by CoreWeave, focused on how companies can move physical AI from pilots into production.

CoreWeave, a US-based neocloud, provides computing power for training AI models. It agreed to acquire Monolith, a London startup that Ahlfeld founded as an Imperial College London spinout, in a deal announced in October 2025. Ahlfeld now runs its newly created physical AI department.

MicroAGI raised $55 million in a seed round, which Ahlfeld called the largest in German history. Its Atlas platform aims to help industrial companies move robots from demos into production.

Rohmann said the bigger open question is whether today’s AI models and hardware are good enough. Safety is a large part of that, and he was in London the week before the summit to discuss standards with the British Standards Institution (BSI).

“Are we there yet, that you can reliably deploy a robot to around 99.7%? That’s the question,” he said. “If you have to put the humanoid into a robot cell because it’s not safe, then it’s really hard to deploy a robot.”

He said a humanoid makes the most sense when it moves on a mobile platform and combines two, three or four tasks.

“If you have a task that’s 24/7, then it might not make sense to put a humanoid there. Just build a machine, maybe,” Rohmann said.

He said demand is not the problem. Some enterprises, such as German industrial supplier Schaeffler, already have dedicated teams and may supply humanoid robot makers or deploy their own robots. Others are only starting to ask whether physical AI is relevant to them and want to test it with a focused proof of concept.

“The customers are there. The overall interest is there,” he said.

Ahlfeld recently toured more than 20 Bay Area startups, from Stanford dorm founders to firms that have raised billions.

He said robots' ability to learn is at a really exciting stage, and the idea that more data and compute produce more intelligence, which transformed language models, is now reaching the physical world.

“Maybe not all the demos yet are real. Maybe it’s still really hard to get into production,” he said. “There are a lot of things we still need to do, but you can see that there is something happening.”

Start with the task

Rohmann said the first step is always to pick a task that today’s hardware and models can handle. MicroAGI is hardware-agnostic, working with robots from Unitree, AgiBot, Galbot and Sharpa, among others. He has spent recent weeks visiting shop floors across Europe, MicroAGI’s main focus, in logistics, automotive, manufacturing and retail.

He said the process then runs in four steps:

• Collect data on site

• Train the model

• Choose the right hardware

• Deploy on site

“Collecting data on site is the best data that you can collect. We collect the object, the sequence of how someone is doing the job and also the edge cases,” Rohmann said. “If something breaks or falls down, if there’s a production stop, what’s happening? What’s the fallback plan? You only have that if you collect data on site.”

He recalled a site in Switzerland where a 100-year-old machine is so worn that operators cannot simply load copper parts into it. They need a specific hand movement to make it work, and a model can learn that only from on-site data.

Ahlfeld gave the hypothetical example of a carmaker that has never been able to automate feeding cables into a car door and wants to see whether physical AI could make the job faster and cheaper.

He said every factory needs its own fine-tuning, even when a robot starts with a generic dataset.

“Everything is very unique, and a lot of things have gone into very clear rhythms over the years that you want to retrofit your tool into,” he said.

People are increasingly filming those generic datasets. MicroAGI, founded in 2025 by former Formula 1 engineer Bercan Kilic, set out to collect data first and train models second. It hands out head straps so people can record egocentric, or first-person, video in homes, warehouses and factories.

Other sources include teleoperation and tactile gloves, but MicroAGI focuses on egocentric video. It launched with two products: Shift, which collects and annotates training data, and Atlas, which handles deployment.

Rohmann said he spent days in March and April at a camping table at the Pinakotheken in Munich, handing out head straps to the public in the rain. The footage is then annotated, turned into 3D environments and used for hand-pose tracking before it trains models.

“Instead of marking homework or serving coffee or working in a bakery, you can now wear a headset and train a robot in your spare time,” Ahlfeld said. “These initial data sets can get you started much faster and give the robot some basic intelligence.”

Training the models is a separate challenge. He said a large training run needs 50,000 high-performance graphics processing units (GPUs) working in unison, and a failure can waste $100 million. That is why neoclouds such as CoreWeave were created, and his department aims to help customers train such complex models for physical AI.

He said the US gained an edge in large language models partly because few would risk that much on a training run that might fail. CoreWeave is now building data centers in Europe.

Ahlfeld said anyone can now connect a roughly €300 Hugging Face robot to a frontier AI model and be amazed by what it can do, but turning such experiments into reliable production systems will still take years of validation.

Cognition, a San Francisco-based artificial intelligence (AI) lab, is expanding in London to serve UK enterprises adopting AI software engineering.

Its UK team has grown by more than 200% in six months, and the London hub will support operations across Europe, the Middle East and Africa (EMEA), the company said.

“The UK is one of the world’s most important technology and financial services markets and we’re seeing significant demand from organizations ready to put AI software engineering to work,” said Russell Kaplan, co-founder and president at Cognition. “This is a long-term investment for Cognition. We want London to become a major hub for our growth across the UK and EMEA.”

Founded in 2023, Cognition developed Devin, an autonomous AI agent that plans, writes, tests and deploys code.

Narry Singh has joined as Senior Managing Director. He most recently served as global co-head of digital at AlixPartners and previously as global head of growth and strategy at Accenture Digital. He is also a former non-executive director of the London Stock Exchange.

Cognition said Singh will work with chief executives, finance chiefs, and boards to build the business case for AI software engineering, put it to use responsibly across enterprises, and manage the workforce and operating-model changes that follow.

Aman Thind has joined as field chief technology officer (CTO) for financial services. He was previously managing director and global head of technology strategy and architecture at JPMorgan Chase, and chief architect at State Street, where he led the firm’s AI, cloud and digital asset strategy.

He also co-founded BestX, a fintech State Street acquired in 2018.

Thind will partner with chief information officers, CTOs and engineering leaders to bring AI agents into software development and support Cognition’s banking and financial services customers globally, the company said.

In the UK, The Access Group, a business software provider, has deployed Devin to more than 2,000 engineers. Cognition said the rollout contributed to a fivefold increase in merged pull requests and sped up some legacy software migration tasks by up to 480 times.

Across Europe, Cognition recently partnered with McKinsey & Company to help enterprises scale AI-powered software development and get more return from their AI spending.

Rival coding platform Cursor announced a London headquarters in June, saying its EMEA revenue had tripled quarter over quarter.

Cognition is widening how enterprises can access its tools. In September, it signed a multi-year collaboration with Amazon Web Services (AWS) to help companies speed up cloud modernization, clear security and engineering backlogs, and run autonomous software engineering inside existing AWS environments.

Backed by a September round that raised more than $2 billion at a $48 billion valuation, Cognition plans to keep growing its London team as it chases enterprise demand.

Cisco, a San Jose-based networking and security company, warns that the rush to deploy AI agents is outpacing companies’ ability to monitor them, and unpredictable bills are now a bigger headache than high ones. The findings come from its Splunk State of Observability 2026 report, released on October 20.

Oxford Economics surveyed 2,050 IT operations (ITOps) and engineering professionals in 10 countries, including the UK, between February and March.

“Most companies will soon have far more agents at work than people,” said Jeetu Patel, Cisco's president and chief product officer. “Managing agents at scale requires reimagining both security and observability.”

The report found that 55% of respondents could not accurately control their AI costs.

An additional 60% said unexpected billing spikes caused more operational strain than high baseline costs, while 57% said telemetry costs had overtaken the infrastructure they were monitoring.

“For today’s enterprises, the excitement of AI is clashing with the reality of the bill,” said Kamal Hathi, senior vice president and general manager of Splunk at Cisco. “We believe observability is the essential bridge to restoring predictability and confidence to the agentic AI era.”

Concern is widespread, with 83% of respondents worried about AI agents' unintended consequences.

While 87% said monitoring AI was a priority, 54% had no clear business outcomes tied to it. More than half (56%) could not measure metrics such as large language model (LLM) response quality, and while 85% said visibility into hallucinations was important, only 58% said they tracked them effectively.

Monitoring AI has also pulled IT and security teams closer, with 81% saying it had increased collaboration. Joint troubleshooting rose to 77% from 62% in 2025.

“This report shows that observability and security data are converging, and that teams need to collaborate more than ever,” said Patel. “Common data platforms that help us detect and respond to issues in real-time are the key to deploying agents we can trust at scale.”

Some 32% of UK respondents said they used AI agents all the time to diagnose root causes, compared with 20% globally. At the same time, 55% said they were strongly worried about agents' unintended consequences.

Elsewhere, 51% of Indian firms said their monitoring of AI model quality was very effective, compared with 35% globally. In Saudi Arabia, 98% said AI let them focus more on innovation, while 74% of German respondents rated visibility into AI privacy as very important.

The report identified a group of leaders who closely monitor model quality, cost and security. Of these, 91% could accurately attribute AI costs to specific agents and models, compared with 69% of other firms.

First Light Fusion, an Oxford-based fusion developer, will fire projectiles into graphite at over 3 km/s to help UK defense scientists study extreme impacts.

The tests will study crater formation, energy dissipation and transitions in material strength.

“This program demonstrates the growing demand for First Light Fusion’s capabilities beyond fusion energy,” said Paul Holligan, vice president of commercial and partnerships at First Light Fusion.

Founded in 2011, the company is developing FLARE, a two-stage fusion fuel system.

In March 2025, it dropped plans to build its own fusion power plant, turning to partnerships with other fusion developers and to materials testing for space and defense. Six months later, it published a paper setting out FLARE as a reactor-compatible route to high-gain inertial fusion.

The campaign for the Defence Science and Technology Laboratory (Dstl), the research arm of the UK Ministry of Defence, will reproduce a 2017 experiment by Hébert and colleagues that fired steel projectiles into porous graphite.

Replicating the study allows direct comparison with published results and tests the accuracy of First Light’s methods and measurements.

The work will also probe a change in graphite’s behavior observed at impact velocities above about 3 km/s. Dstl will supply the targets, which are configured for recovery after each shot so researchers can inspect the damage in detail.

The laboratory said the program would deepen its understanding of how materials behave under hypervelocity impact. It said First Light’s facilities and expertise would let it examine these effects across a carefully controlled range of conditions, supporting the development and validation of improved material models for defense applications.

First Light’s gas gun and pulsed-power platforms, built for its fusion work, can recreate the extreme temperatures and pressures relevant to aerospace, defense and space testing. They include the UK’s largest two-stage gas gun.

In June, its VIPER amplifiers pushed projectiles past 12 km/s, comparable to orbital debris, at a Texas A&M test station.

First Light expects demand for its fusion-built facilities, amplification technology and simulation tools to keep growing among government and commercial organizations in defense, aerospace and advanced materials research.