CoreWeave

A look inside the Day 1 keynote at #FullyConnected26. Brian Venturo opened with the reason this conference exists: "We didn't retrofit a cloud for something else. We built the AI cloud from first principles, from the ground up, for AI." Then Chen Goldberg put the whole week on one slide: the AI loop. "Run, observe, curate, improve, evaluate, and run again." She launched the thing that runs it, CoreWeave Forge, a development layer for teams building and improving models and agents, open across the models, frameworks, and clouds you already use. Then she ran the loop live on stage, and ran the handoff to Corey too. Literally 🏃♀️ Corey Sanders, also known as the human in the loop, took it from there with three more launches: CoreWeave Agent Lens, so you can see how your agents actually behave in production. CoreWeave Model Distillation and RL Rollouts for teams doing reinforcement learning. Then he let CoreWeave ARIA, our AI Research and Iteration Agent, work a real problem live. Brandon Hootman of Caterpillar Inc. sat down with Chen to talk about autonomy on the job site, where "only just three or four years ago" bringing autonomous operations to a construction site "almost felt unrealistic." Ian Buck of NVIDIA pulled the cloth off the NVIDIA Vera CPU tray with Chen. The first CPU built for AI agents is coming soon to CoreWeave. Six generations of NVIDIA GPUs on CoreWeave, and in Ian's words, "they've never, ever been turned off." Maulin Patel of Capital One showed what it takes to evaluate agentic AI at enterprise scale, inside a bank, in production. On the GPU side, NVIDIA Vera Rubin NVL72 is ready for production, and Silas Alberti announced Cognition as the first customer in the world running on it. His line on what that means for Devin their AI software engineer: "not only does the model get smarter, but it also gets cheaper." Michael Intrator closed it out with the CoreWeave origin story, including the rule he's run the company on: "make every decision to drive quality and then make every decision to drive speed, because they are, to an extent, the same thing." Brian came back to send everyone to lunch: "Try it, see what's real, make the next one better." Thank you to Caterpillar Inc., Capital One, NVIDIA, and Cognition for being on that stage with us.

We've run these integrations in production for years. From the first wave of open-source LLMs to the latest in AI agents, CoreWeave and our Weights & Biases Models product have kept production integrations running with, VAST Data, IBM and others. The CoreWeave Partner Network shows exactly which tools are proven and ready to run on CoreWeave Cloud, and it gives us a repeatable way to go deeper with the leaders AI teams already depend on. The bar for getting in hasn't moved. As Seann Gardiner, our VP of Business Development, puts it, "An integration that passes on a spec sheet can still fail under real traffic." Every partner in the network has been tested on our infrastructure under real production conditions, and partners are here because customers are already running them or asking for them. See who's in the network: https://crwv.co/utcBq

Big news from #FullyConnected26: CoreWeave Forge is here 🎉 A production trace that never reaches the next training run is a signal you paid for and threw away. Most teams do it every day, because the tool that catches the trace and the tool that runs the training came from different vendors and were never built to talk. Forge closes that gap by unifying Weights & Biases, post-training from OpenPipe, and marimo notebooks in one connected environment with CoreWeave Training, Inference, Sandboxes, and Registry. Run, observe, curate, improve, evaluate. The traces you flag in production become the datasets you train on and the evaluations you gate with. Open across any model, framework, or cloud. MasterClass and Canva are already building on it. Free, Pro, and Enterprise available today. Closed loop. Open stack. See the announcement: https://crwv.co/utcwI

Our CEO Michael Intrator joins CNBC live from Fully Connected to discuss the world's first Vera Rubin deployment, the Vera CPU, and more. Watch now.

Cognition engineers ran the first customer-executed benchmark on NVIDIA Vera Rubin NVL72, on CoreWeave Cloud. They measured it against their own NVIDIA GB200 NVL72 baseline. Up to 4.8x the total token throughput for SWE-2 inference. 3.8x the output token throughput for reinforcement learning. Agentic coding is an unforgiving workload. Long contexts. High concurrency. Token volumes where cost per token decides what you can ship. For Cognition, those numbers mean more concurrent Devin sessions per GPU, drastically accelerated research loops, and lower cost per session, with no loss in generation speed. The whole stack runs on CoreWeave: training, reinforcement learning, and the inference that serves users. All of it on one platform. The benchmark is theirs. So is the workload. https://crwv.co/utcq6

Lights. Camera. Action: Fully Connected starts NOW. See you at Moscone South for our day 1 keynotes, featuring our executive team and leaders from NVIDIA, Capital One, Cognition and Caterpillar Inc.

Spotted: all around San Francisco! That’s right: we’re in town for #FullyConnected26 and officially on Instagram, too. Follow along for all conference highlights, and if you’re here, tag us! 👉 @CoreWeave_ on Instagram 👈

You've checked in, got your badge, and settled in. Now let's kick off #FullyConnected26... 5:00pm: Pioneer's Night opens in the Pioneer Pavilion. Demos, arcade games, live music and the people actually building, ready to talk. 5:40pm: CTO and Co-Founder Peter Salanki takes the stage for Ask Me (Almost) Anything: CTO Stories Behind the Scenes. His hot takes, plus answers to your questions about running AI in production. See you tonight 🎉

Fully Connected starts before the keynotes do. And we're kicking it off in style with Pioneer's Night. 👾 90s arcade games 🤝 Networking 🎵 Live Music ...and a few surprises Plus, an AMA with our CTO and Co-Founder Peter Salanki. Bring the question you'd normally save for the hallway. And stay for the DJ. See you tomorrow ➡️ Pioneer's Night. Moscone South. 5:00-7:00pm