ICYMI: the CoreWeave Partner Network is here. A strong ecosystem is important to our customers, and it’s important to us. That’s why we’ve brought together a curated network of technology partners with integrations proven on CoreWeave in real-world production environments. From infrastructure and data to security, observability and more, the network helps you build on the tools you already trust. That way, you can move from experimentation to production faster. More here:

Congratulations to @Cognition for being the first customer live on NVIDIA Vera Rubin NVL72, powered by @CoreWeave. With 4.8x more token throughput than GB200 NVL72 at the same decode speed, this is what next-gen agentic AI looks like. 👏

Congrats @CoreWeave on RL Rollouts! RL post-training involves a lot of back and forth: train the model, generate responses, then train again. Inference workers need to load the updated model weights each time. As models get bigger, that can leave GPUs waiting. CoreWeave’s new service uses ModelExpress and Router in NVIDIA Dynamo to speed up those reloads with minimal downtime. Working with us and @youdotcom, CoreWeave achieved 15× faster model reloads compared with its baseline while post-training Nemotron 3.5 Lightning. Check out their blog below for details

Next on the keynote stage is @dylan522p, who grades AI clouds for a living. (we're ClusterMAX Platinum 3x btw 😉) The @SemiAnalysis_ founder joined @GoldbergChen at #FullyConnected26# on bringing compute, inference, environments, and storage closer together.

Your first agent will never be your best agent. Day 1 of #FullyConnected26# was about that gap. CoreWeave Forge: the whole AI loop in one connected environment. Then Agent Lens, Model Distillation, RL Rollouts, and an NVIDIA Vera CPU unveiling.

Hands down, this hands-on session is one of the coolest I’ve been so far! Today we got to train a robot on Mars. Well, in simulation. The mission: retrieve a fuel cell 200 meters away and bring it back before life support runs out. There was one problem. The quadruped already knew how to walk, but it had been trained for Earth. Earth gravity: 9.81 m/s². Mars: 3.72 m/s². Put the same policy on the Martian surface and the robot basically trips over itself. So we had to teach it how to walk again. We used reinforcement learning, specifically PPO, trained the policy on CoreWeave infrastructure and tested it in simulation. Then you look at what happened, change the training setup and run it again. Sometimes the robot walks. Sometimes it does something completely stupid. That was probably my favorite part. What I liked about the workshop was actually seeing that loop end to end instead of another slide about Physical AI. Led by Anushrav V., @CoreWeave’s Robotics Field Engineering Lead, at Fully Connected.

Introducing CoreWeave Agent Lens: automated observability designed to turn agent traces directly into actionable fixes and continuous model improvements. Close the loop on agentic AI and accelerate production workflows:

When your training run stalls, you don’t need a support queue—you need the engineers who built the platform. Meet CoreWeave Direct-to-Expert: real engineers, fast resolution, right away:

Yesterday we announced the CoreWeave Partner Network! Scaling production AI shouldn't mean navigating complex infrastructure alone. The CoreWeave Partner Network clears the path to production AI with world-class integrations, solutions, and technical expertise. Learn how we're accelerating the path to AI innovation: