Qarnot’s Post

Scaleway has acquired French HPC specialist Qarnot to strengthen its AI cloud platform with high-performance computing and energy-efficient infrastructure. Key developments: • Scaleway has acquired French HPC company Qarnot. • The deal expands AI and high-performance computing capabilities across Scaleway's cloud platform. • Qarnot brings patented direct liquid-cooling and waste heat recovery technology. • The acquisition enhances infrastructure for AI model training, engineering simulations, and scientific research. • Qarnot operates more than 50,000 computing cores across multiple sites in France and Europe. • The partnership reinforces Europe's sovereign cloud ecosystem with energy-efficient AI infrastructure. Read More: https://lnkd.in/dfGuFdqK #Scaleway #Qarnot #ArtificialIntelligence #CloudComputing #HighPerformanceComputing #HPC #SovereignCloud #Sustainability #France #Konsulteer Sandeep Sharma Harsh Sharma

"Telecommunications is entering the AI era — the radio access network is the next AI infrastructure" - Jensen Huang, Founder and CEO of NVIDIA AI-RAN evolves legacy Telco infra to Accelerated Computing. It changes the Telco economics with increased spectrum and energy efficiencies, automated operations and new opportunities for monetization with AI. NVIDIA Nokia

The AI conversation is evolving—from software and models to energy, compute, and global infrastructure, as scaling challenges become increasingly tied to real-world infrastructure constraints. Explore the full analysis 👉 https://lnkd.in/gJKUWJjz

It's great to see how customers and partners are expanding their AI Factory footprints using NVIDIA DSX: Aolani and Rafay deliver a product-ready AI platform using DSX OS "enables organizations to provision Kubernetes clusters, virtual machines, AI workspaces, and inference environments through a secure self-service experience while maintaining centralized governance, policy enforcement, and operational visibility." https://lnkd.in/gJ_4S4av SK Group will use DSX and VeraRubin to create the first AI factory planned to come online in 2027: https://lnkd.in/gsvuCxMJ NAVER Corp continues its path toward gigawatt-scale sovereign AI infrastructure using DSX https://lnkd.in/gyXdfZZh #AIFactory

Friday reflections within the Global AI Infrastructure world. I spoke to operators this week sitting on clusters worth more than most companies will earn in a decade, quietly worried about one thing that never makes the press release: who is going to actually run this when it goes live. A GPU cluster without an ops team is not an asset. It is stranded capital with a power bill. Three things stuck with me this week: - Landlord and operator are not the same job, and confusing them is expensive. The team that commissions the building is not the team that keeps 20,000 GPUs healthy at 3am. Both are hard. They are not interchangeable. - Speed to power is the number one site-selection criterion right now, and everyone is optimising for megawatts. But power without people is just a very expensive space heater. Speed to power starts with speed to people. - The best infrastructure leaders are almost never looking. They are heads-down, building, loyal, and quietly overloaded. You do not find them by posting a job. You find them by knowing the market well enough to know they exist. This work is a privilege because it sits right at the seam where physical infrastructure meets digital systems, at the exact moment that seam is being stress-tested harder than it ever has been. Every week I learn something from people who are genuinely building the backbone of what comes next. Grateful for the operators who take the call, the candidates who trust me with their next move, and a market that never lets you get comfortable. Have a good weekend, everyone.

Most infrastructure investors are pouring capital into the same two bets: GPU clusters and foundation models. Both are real. Neither is where the scarcity actually lives. Over the next 36 months, the constraint isn't compute. It's the layer that makes distributed compute usable. Private 5G networks, LEO satellite backhaul, and edge nodes are proliferating fast. But they're still largely isolated from each other. No unified control plane. No single addressable surface. Just a collection of expensive infrastructure islands. That orchestration gap is where the real value gets built. A few companies are working on exactly this, each attacking a different piece of the same problem: • Celona is building enterprise private 5G with a software-defined management layer. The hardware is almost beside the point. The orchestration is the product. • Aalyria (spun out of Google X) is tackling multi-domain network orchestration across satellite, terrestrial, and aerial links. The complexity they're solving makes most AI infrastructure problems look simple. • Rakuten Symphony is assembling an open RAN stack that could become the connective tissue between carrier networks and enterprise edge deployments. • Athonet, now inside HPE, is one of the few teams that has actually deployed private LTE and 5G at scale with real enterprise integration. The reason these companies aren't getting GPU-level valuations is straightforward. The story is harder to tell. "We train models" closes rounds. "We orchestrate multi-domain connectivity at the edge" requires a whiteboard. But the companies that win the AI era won't just be the ones with the best models. They'll be the ones whose models can actually reach the physical world reliably, at the edge, at scale. That's the infrastructure bet most people are still sleeping on.

Every major shift in computing creates a new layer of infrastructure that becomes critical to scaling what's next. As AI systems continue to scale, high-speed connectivity between compute and memory is becoming one of the most important challenges in modern AI infrastructure. Seligman Ventures is excited to lead Eliyan Corporation's $145M Series C and partner with repeat founders Ramin Farjadrad, Patrick Soheili, and Syrus Ziai as they build the next-generation interconnect technologies for the AI era. We invested in Eliyan because we believe the company is well positioned to address one of the industry’s most important architectural challenges. The team combines deep semiconductor expertise with a proven track record of execution and the confidence of an outstanding syndicate of strategic investors spanning semiconductors, memory, networking, and optical connectivity. We're also proud to share that our Managing Partner, Umesh Padval, will join Eliyan's Board of Directors as the company enters its next phase of growth. Eliyan is the latest addition to our growing AI infrastructure portfolio and reflects the continued focus of Umesh Padval, Ashish Kakran, and Eddie Ackerman on backing foundational technologies shaping the future of computing. Congratulations to the Eilyan team and fellow investors on this milestone! https://lnkd.in/eVcM7EBb

Congrats! Stathera, Inc. closed a $55 million Series B financing round, led by Maverick Silicon with participation from Celesta Capital, BDC Capital, MediaTek Innovation Fund, TXC Corporation, and Ultratech Capital Partners. The funding arrives as data center operators seek components to manage higher density, power, and synchronization requirements driven by AI server growth. IDC’s analysis of AI infrastructure spending projects strong double-digit growth through the late 2020s, as operators confront increased thermal and power loads from GPU clusters. While timing components rarely command the spotlight, they provide critical foundational infrastructure. The company’s focus on silicon oscillators directly addresses these emerging density and synchronization gaps. Precision timing has become a strategic constraint for hyperscale designs. According to the CEO, AI data center performance is increasingly limited not by raw compute, but by data movement and synchronization across tens of thousands of processors. Quartz components, the industry default for decades, are struggling to scale at the pace modern AI systems require. This limitation is pushing hyperscale customers toward standard semiconductor alternatives that offer narrower size and power envelopes. https://lnkd.in/gBqA2Dep

AI is no longer a future trend. It’s becoming a key driver of infrastructure investment around the world, and as demand for AI, cloud computing, and digital infrastructure grows, so does the need for innovative solutions that support the next generation of data centers. Recently, Flavio Crovador Marques (Account Manager) represented Nelson Global Brazil and GT Silex at the ABDC - Brazilian Data Centers Association Strategic Meeting in São Paulo, joining industry leaders to discuss opportunities and challenges shaping this rapidly evolving market. GT Silex delivers integrated thermal, NVH, acoustic, and aftertreatment solutions for large industrial power applications worldwide, supporting the critical infrastructure behind rapidly expanding digital technologies. By staying connected with industry leaders, we continue to identify opportunities, anticipate market trends, and support our customers’ long-term success. Thank you to ABDC for bringing together organizations committed to advancing the future of this important industry and fostering collaboration across the market.

Effective AI infrastructure performance hinges on optimizing data movement, not solely on raw computational power. AI workloads present unique challenges to traditional enterprise architectures, making efficient data delivery at the right time absolutely critical. Initiatives like StorageAI and SDXI are designed to enhance interoperability across storage, memory, networking, and accelerators, ensuring data is moved with maximum efficiency. In a recent article, Cyril Bandolo shares reflections from AI Infrastructure Field Day 5 discussions with SNIA and Tech Field Day delegates, exploring how open standards could significantly shape the future of AI infrastructure. #AIIFD5 #AIInfrastructure #SDXI #SNIA #StorageAI Learn more about this vital shift: