CoreWeave

The rumors are true 👀 Proud of the CoreWeave engineering team and our partners (like Dell Technologies!), who are the first to have a fully working NVIDIA Vera Rubin NVL72, achieving yet another major milestone in bringing next-generation AI infrastructure online. These achievements happen through extraordinary execution across hardware, software, networking, and operations. It’s a testament to the CoreWeave teams who make it happen every day. There is so much more to come, make sure you stay tuned.

This is what we mean by partnership 🏁 We brought our Customer Advisory Board to the home of the Miami Grand Prix alongside the Aston Martin F1 Team, and Keith Culley from CrowdStrike said what everyone was thinking: "We find ways to help each other, and find ways to power each other's businesses." Not a bad way to spend a weekend with the people building the most ambitious AI systems in the world. Watch the full film.

We asked, and you answered: The token cost is the wall 🤖 🧱 And profitable agents need new economics. It's what we've been designing with NVIDIA: GB300 NVL72 puts 72 Blackwell Ultra GPUs in one NVLink domain. DeepSeek V4 runs there at 20x lower token cost vs H200. New economics unlocked.

AI agents fail in production. Not because they're poorly built. Because they're trained on data that doesn't reflect the real world. We just fixed that. We just closed the loop between training and inference, so agents improve from real-world experience instead of waiting months in eval cycles that can't cover every scenario. One closed loop. Four pieces working together: ✅ Serverless RL that scales without infrastructure overhead (40% cost reduction, 1.4x faster) ✅ Inference designed for production, not just demos  ✅ Weights & Biases Weave for full observability across multi-agent workflows  ✅ Autonomous improvement via W&B Skills and MCP server Agents that compound in capability over time. Not just more reliable, but getting better as they run. See how. https://hubs.la/Q04jdVL50

If you're building AI at scale, this is the room you want to be in. This is not just another AI conference. Not a vendor showcase. Not another panel discussion about “potential.” This is what production AI actually looks like: Real infrastructure. Real workloads. Real proof from the teams actually shipping. Fully Connected is coming to San Francisco, September 29-October 1st. Accelerate the superintelligence loop with: 3 Days 3 Tracks 2,000+ AI Leaders Live Demos Hands-on Labs Keynotes, special guests, and more. This is the room where it happens. Registration is open. ♾️https://lnkd.in/g7PnjD4X #FullyConnected26

This is what performance looks like. Aston Martin F1 Team Academy driver, Mathilda Paatz, came into last weekend with zero points from the previous race. She left with her first career win and two trophies 🏆🏆 She didn't just compete, she pushed the boundaries of what's been done before. She show up when the pressure was highest. And she made it look inevitable in hindsight. Some people are just built for this. So are their partners. Congratulations to Mathilda and the full Aston Martin team!

"We have to manage the ripples and ride the waves" 🌊 In this latest episode of Who’s Ready for Anything, CoreWeave CMO Jean English sits down with author and technologist Matt Watts to discuss his latest book, The Fourth Wave, and how pioneering companies are turning AI ambition into real-world impact. Don't miss it. 🗓️June 3, 8 am PT LinkedIn Live: https://lnkd.in/gA5p3BuA

ICYMI: CoreWeave Sandboxes is now in public preview 👏 It's the execution layer for RL, agent tool use, and model evaluation, on your own CKS clusters or serverless through Weights & Biases. See it in action. https://hubs.la/Q04gxLZr0

Markets change overnight. AI has to keep up. Dimitris Tsementzis joins the AI Cloud Essentials podcast to discuss why building autonomous trading algorithms is like designing an LLM for an audience that wakes up speaking a completely different language every morning. How do you build systems that function while they're still learning to adapt? Check out the full episode to find out. Link in comments.

Getting agents into production is one thing. Making them profitable is another. The economics are shifting fast. Multi-agent workflows are burning tokens by the multiplier. Hardware roadmaps are getting outpaced by model demands. And most teams are still tracking compute cost when token cost is what's actually moving the P&L. We've been watching teams hit different walls at different scales.