Crusoe

We’ll be in San Jose Oct 20-21 for PyTorch Conference North America 2026. Find us at our booth to: ➡️ Take the Prompt and Collect challenge ➡️ Demo Crusoe Managed Inference and AI Cloud Platform with our team Plus, catch Crusoe speakers throughout the conference: 📅 Oct 20, 11:45 AM: Suman Debnath and Janaki Ram Goteti teach an introduction to GPU programming with Triton 📅 Oct 20, 12:45 PM: Suman Debnath and Jiale Huang lead “Fitting Large MoE Fine-Tuning onto Few Nodes: Expert & Context Parallelism in DeepSpeed” 📅 Oct 20, 6:30 PM: join us, dstack and SGLang (LMSYS Org) for AI Infra Night, an official Open Source week event. We’ll bring together AI experts, ML engineers, AI researchers, and AI infra enthusiasts for drinks, food, and plenty of time to talk shop 📅 Oct 21, 10:15 AM: Yiwei Song delivers a keynote, “AI Model Customization Platform” Register for AI Infra Night: 🔗 https://luma.com/6j69xyal

We're in Las Vegas for Yotta 2026, the must-attend event for digital infrastructure leaders. Crusoe is leading conversations at the event. Here's a look at what we're up to: 📅 Yesterday: Our Chief Data Center Officer, Chris Dolan, joined leaders from Generate, Foresight and HDR for "Working through the Ripple Effect of Cancellations and Delays," a session examining the cascading effects of project disruptions. 📅 Today at 4pm PT: Our Sr. Director of Engineering, Jason Hutzler Hutzler, joins "Islanded, Grid-Connected or Inject-and-Withdraw?" alongside the CEO of our partners ON.energy and panelists from Latitude Media, CyrusOne and PDM - Power and Data Management to explore different paths to powering large-scale AI infrastructure. 📅 Tomorrow at 9:35am PT: Crusoe Co-Founder and CEO Chase Lochmiller Lochmiller takes the main stage with Yotta's Matthew Stansberry to share what Crusoe has learned from building at gigawatt scale and what it tells us about where AI infrastructure goes next. You won't want to miss these conversations. We're excited to see you in Vegas!

24 hours. 350 builders, product developers, and engineers. What can you build in a single day? Today, on Day Zero of The AI Conference, Crusoe is sponsoring an all-day hackathon in SF. The challenge behind Hack Day is simple: take an idea from zero to a working product in just one day. The winning team will take the conference's main stage tomorrow to pitch a panel of top venture capitalists. We’ll be here all day. Come meet the Crusoe for Developers team and let’s build. Space is limited. Register: 🔗 https://lnkd.in/ga6Ei9TQ

Crusoe was founded on a simple idea: bring compute where energy already exists, not the other way around. Today, large swaths of wind and solar generation are curtailed — deliberately throttled back — because it's produced far from demand. Our approach solves that: co-locating AI infrastructure directly at generation sites turns stranded power into productive capacity, and gives developers a commercial reason to repower and expand. We call it "across the meter": AI factories that draw power when they need it, and give surplus back to the grid when they don't — an asset to grid reliability, not a burden on it. More on our energy-first approach: 🔗 https://lnkd.in/gX4gYgpZ

Our co-founder and CEO Chase Lochmiller sat down with The Atlantic's Charlie Warzel to talk about how, and why, Crusoe is building AI infrastructure differently. A few things he shared: the water usage narrative doesn't hold up. At our Stargate campus in Abilene, Texas, each building uses about as much water as 10 single-family homes since GPU cooling runs in a closed loop. On the economic side, Crusoe will contribute more than a third of combined city and county tax revenue in Abilene, and in Armstrong County, one of only three Texas counties with zero oil and gas revenue, our investment is more than doubling tax receipts while catalyzing new wind, solar, and battery capacity. Chase also previewed our next move: smaller, modular data centers that ship on a flatbed truck and deploy wherever power already exists, cutting years off timelines and easing construction impact on host communities. Watch the full conversation: 🔗 https://lnkd.in/e2n_txpZ

Spotted on the 101 in the Bay Area 👀

One validation is a result. Two in a row is a pattern. Crusoe is now an NVIDIA Exemplar Cloud for training on NVIDIA HGX B300. Together with our HGX B200 validation earlier this year, that makes Crusoe an Exemplar Cloud on both Blackwell generations. Exemplar doesn't test isolated components. It runs real LLM training workloads end-to-end and measures them against NVIDIA's reference performance. Here's what it took across facility design, orchestration, and networking: https://lnkd.in/gZyM_Gck Credit: Alex Akesson Young Jeong Sanchit Pathak

🗣️ "This was not tens of nodes or hundreds of nodes, but thousands of nodes on extremely short notice… We were able to strike a deal within a couple of days, and we got those nodes up and running within a few hours of signing." Chenyu Zhao , Co-founder, Fireworks Fireworks AI serves 40 trillion tokens a day, and every one runs on a GPU somewhere. Their GPU fleet grew 10x in a year. 📈 How they scale on Crusoe Cloud, powered by NVIDIA accelerated computing: → Crusoe AutoClusters checks node health and swaps in fresh nodes automatically, so no training or serving time is lost → A Crusoe engineer responds within minutes, and the median issue is resolved in under four hours → The two teams share monitoring and automated failure handling, with Fireworks observing 99.99% availability Fastest turnaround from cluster delivery to serving customers: just hours. Read the full story → https://lnkd.in/eFdrKfXi

Every team building on AI faces the same decision: call a closed frontier model, run and customize on open models, or train something proprietary. The right answer depends on where you are in the lifecycle. Join us in San Francisco on October 7 for a night of real talk with founders at Ollama and Trajectory who have answered that question differently. Crusoe’s SVP of Product Management, Erwan Menard, will join Ollama’s co-founder Jeffrey Morgan and Trajectory’s co-founder Michael Elabd for a panel discussion on how their teams are making these decisions, and how cost, control, performance, and speed factor in. Stick around after the talk for an audience Q&A and networking reception. Space is limited. Register: 🔗 https://luma.com/lnnlfa87

🗣️ "We benchmarked the cluster, and we realized that Crusoe matches the state-of-the-art speed in terms of both the GPUs and the storage. We decided to go with Crusoe because it was the best possible choice." - David Novotný, Co-founder and CTO, SpAItial AI SpAItial builds world models that generate physically grounded environments, the kinds of places a robot can learn to move before it ever acts in the real world. Training one means sharding a model across NVIDIA Hopper GPUs on Crusoe. Near-linear scaling depends on the interconnect, and on storage fast enough to keep the GPUs fed. The team runs on a high-leverage principle: the question is always what they don't have to do themselves. Crusoe Managed Kubernetes handles the control plane and security patching. Crusoe manages the private VPC over NVIDIA Quantum InfiniBand networking, so SpAItial's engineers never own the networking layer. In their own benchmarks, the cluster held over 97% scaling efficiency, so almost no compute is lost to interconnect overhead as the cluster grows. Training experiments now iterate roughly 30% faster. Watch the full story → https://lnkd.in/e-rcg6Xd