Crusoe

We had a blast hosting NVIDIA Staff Research Scientist Dhruv Nathawani in our San Francisco office last week. Dhruv works at the intersection of synthetic data and foundation models for NVIDIA Nemotron, a family of open models with open weights, training data, and recipes designed for efficient real-world deployment at scale. In his talk, Dhruv covered: 🔧 Synthetic data as a lever for model performance, evaluation, and scalability 📈 Improving model generalization across tasks with synthetic datasets Crusoe’s ongoing guest speaker series brings in leading builders and researchers like Dhruv to foster continuous learning across the team. Want to join us? Check out open roles: 🔗 https://crusoe.ai/careers/

Engineers building AI products, we have a question for you... And our leaders have a few thoughts of their own ➡️ https://lnkd.in/gGtndt86

$ 𝘪𝘯𝘪𝘵 𝘩𝘢𝘤𝘬𝘢𝘵𝘩𝘰𝘯 --month 𝗺𝗮𝘆 --year 𝟮𝟬𝟮𝟲 We're gearing up for Hack.Crusoe, our upcoming internal hackathon built entirely around Crusoe Managed Inference! From May 5–11, teams across Crusoe are going heads-down to build on our own inference platform and ship something real. One week. One product. No limits on what we can create. We can't wait to share what our engineers come up with. Stay tuned to see Managed Inference in action! #hackcrusoe26 #crusoe #inference #hackathon #developers

Most AI takes come from the outside looking in. We went a different direction. 'From electrons to tokens' features five Crusoe leaders sharing what they're actually seeing in AI infrastructure right now: ➡️ The AI factory concept ➡️ Energy as the defining constraint ➡️ Why speed is the only choice ➡️ The evolving role of the engineer ...and more. Not analyst research. Not polished corporate takes. Real answers from the people building it. An authentic view from inside the machine: https://lnkd.in/gGtndt86 Featuring: Cully Cavness Chris Dolan Erwan Menard Tamanna Sait Kyle Sosnowski

Director of Portfolio Planning Malavika (Malu) Kannuswamy joined us in San Francisco for a fireside chat, “Beyond Networking: How Real Connections Build Extraordinary Careers.” Malu shared candid stories and practical advice with Crusoe’s SF and Sunnyvale teams, offering a clear look at how intentional relationships, collaboration, and consistency can shape a long-term engineering career. Grateful to everyone who joined yesterday with thoughtful and insightful questions — and special thanks to Senior Engineering Manager Rithu John for organizing such a meaningful session! 📸 Moments like these reflect Crusoe’s commitment to building a culture of learning, curiosity, and community. Through our networking speaker series, we’re excited to bring in leaders like Malu who are investing in the next generation of builders. Want to join our team? We’re hiring: 🔗 https://crusoe.ai/careers/

The people building the next generation of AI infra tend to find each other at events like this one. We're co-hosting an AI infra and open-source models meetup in San Francisco on April 30 with dstack, during AMD Dev Day. Seven lightning talks (training, inference, open models) from practitioners who are actually in the weeds: AI experts, OSS contributors, and ML engineers doing the hard work. If that's you, or if you want to be in the room, come through. Starts at 5 pm. 👉 luma.com/ykt5b294

Jensen Huang on the Dwarkesh Podcast! An hour-plus covering NVIDIA's supply chain bets, whether TPUs can break their hold on AI compute, the China chip question, and why NVIDIA won't become a hyperscaler. We sponsored this episode because the infrastructure questions Jensen wrestles with are the same ones Crusoe is building toward — starting with what actually determines real-world inference performance when everyone's running the same GPUs. Our answer: MemoryAlloy technology. Cluster-wide KV caching that delivers up to 9.9x faster TTFT and 5x better throughput than vLLM. Worth understanding before you pick your next cloud. 🎙️ Full episode: https://lnkd.in/gV5MH86J

An important note for AI builders in 2026… ➡️ “If you’re supply-constrained at a chip level, a data center level, and an energy level, it’s very advisable to work with people who understand all three layers and optimize them together. That translates into more cost-efficient infrastructure and a faster time to market,” said Erwan Menard, Crusoe’s SVP of Product during a webinar we hosted last month. Together, Erwan and Kyle Sosnowski, Crusoe’s VP of Cloud Engineering, broke down insights from our 2026 AI infrastructure trends report, which found that 98% of AI decision-makers rated complete control over infrastructure as critical to their success. Kyle and Erwan’s key advice: ⚡ Optimize for inference, not just training 🧠 Invest early in memory-aware inference architectures 🔌 Plan now for the energy wall 🏗️ Choose partners delivering purpose-built AI infrastructure Read more on our blog: https://lnkd.in/g975fMZ8

Data center power optimization innovator Neuralwatt needed to prove its software to investors by running a data-driven demo on NVIDIA HGX H100 GPUs, but access was the biggest roadblock. Crusoe offered unrestricted access to bare-metal telemetry on the NVIDIA HGX H100 GPU clusters running in our renewable-powered Iceland data center. 🇮🇸 The results: 📈 33% increase in AI inference throughput 🔋 40% reduction in idle GPU power draw 🧠 33% improvement in density Neuralwatt is one of five real-world stories we’ve gathered from the frontier of AI development, featuring builders who worked with us when they were facing critical roadblocks in cost, latency, and control. Download ‘5 examples of AI infrastructure done right’: 🔗 https://lnkd.in/grJ4JbjU

That's a wrap on HumanX 2026 in San Francisco. 6,500+ leaders, builders, and investors. Three days of real conversations about where AI infrastructure is headed, and what it actually takes to get AI into production. We had a great time connecting with so many of you at the booth. If you stopped by to demo Crusoe Cloud and Crusoe Managed Inference — thank you. Those conversations were the highlight of the week. Huge shoutout to our booth team for representing Crusoe all week long: Ian Kistner, Daniel Sobajian, Fabian Oseguera, Nathan Castro, James Barr, Henry Targett, Parker Lawlor, Joseph K., Abraham S., Young Jeong, and Rohit Kalmankar. 📸 A few moments from the floor. #HumanX2026 #AIInfrastructure