At OCP Global Summit, NVIDIA opened Vera Rubin infrastructure designs; Meta adopted Spectrum 6 switches; Visionbay unveiled 800VDC K-1 platform for Vera Rubin; NVIDIA released open source DSX AI Factory tools
Concluded·Day 5·in the edition since 6 Oct 2026·2 sources·updated 10 Oct 2026
CoreWeave's Vera Rubin and Forge at FullyConnected26
The gist
CoreWeave Ships Vera Rubin NVL72, Vera CPU, and Mission Control Tools
The Vera Rubin NVL72 availability gives cloud customers access to NVIDIA's latest GPU generation, with Cognition's benchmarks showing substantial throughput gains for agentic coding workloads. The Vera CPU and Mission Control tooling extend CoreWeave's platform toward supporting the full training-to-agent-deployment loop in one infrastructure layer.
The full picture
CoreWeave has made NVIDIA Vera Rubin NVL72 available on its cloud platform, with Cognition running its Devin AI software engineering agent in production on the hardware and benchmarking up to 4.8x higher total token throughput for inference versus the GB200 NVL72, and 3.8x the output token throughput for reinforcement learning workloads. CoreWeave also announced it will deploy the NVIDIA Vera CPU, which NVIDIA describes as the first CPU built specifically for AI agents, putting 128 CPUs and 11,264 cores in a single rack and supporting more than 11,000 concurrent agent environments alongside GPU training jobs. Testing showed more than 3x faster agent sandbox startup times on Vera CPUs versus prior infrastructure and a 1.7x performance gain across all passing tasks on the Terminal-Bench benchmark. Separately, CoreWeave shipped two Mission Control features in October 2026: a Mission Control Agent in preview that identifies deployment bottlenecks, forecasts throughput improvements, applies approved fixes, and load-tests results; and a Mission Control MCP server that reached general availability as a hosted, read-only server exposing metrics, logs, dashboards, alerts, CoreWeave docs, and cluster state scoped by API token, compatible with Claude Code, Cursor, Codex, and any MCP client.
How it developed
Cognition on October 10 published benchmark data showing Vera Rubin NVL72 delivers 3.8x the output token throughput for reinforcement learning workloads versus its GB200 NVL72 baseline, adding an RL figure to the 4.8x inference throughput result Cognition reported when CoreWeave announced the hardware's availability on its cloud at Fully Connected 2026 on October 6.
First citedCoreWeave Twitter
CoreWeave shipped Mission Control Agent (preview) and Mission Control MCP server (GA)
CoreWeave introduced Agent Lens, an observability tool for detecting and grouping failures across full production agent conversation corpora.
On October 5, CoreWeave reposted the Vera Rubin NVL72 availability announcement, repeating Cognition's benchmark figures of up to 4.8x higher total token throughput versus GB200 NVL72 for SWE-2 inference.
No new facts were added beyond the September 30 announcements.
First citedCoreWeave Twitter
CoreWeave Forge launched at the Fully Connected 2026 conference in San Francisco on October 6, and Cognition ran a live demonstration on Vera Rubin NVL72 hardware onstage at the same event.
CoreWeave had also announced at the conference that Vera Rubin NVL72 was available on its cloud platform, with Cognition benchmarking up to 4.8x higher total token throughput versus GB200 NVL72 for SWE-2 inference workloads.
First citedCoreWeave Twitter
First in the edition · 6 October 2026
CoreWeave reposted availability of Vera Rubin NVL72 with Cognition running Devin in production at up to 4.8x token throughput.
ICYMI post recapped Ian Buck and Chen Goldberg unveiling the Vera CPU live at Fully Connected 2026
CoreWeave announced at FullyConnected26 on September 30 that NVIDIA Vera Rubin NVL72 is available on its cloud, with Cognition benchmarking up to 4.8x higher total token throughput versus GB200 NVL72 for SWE-2 inference, alongside the NVIDIA Vera CPU, which NVIDIA describes as the first CPU built for AI agents, supporting more than 11,000 concurrent agent environments per rack.
CoreWeave also launched Forge, an MLOps platform with MasterClass and Canva as early customers, and a Partner Network of pre-validated ecosystem tools.
CoreWeave announced FullyConnected26 Day 1 keynote featuring CoreWeave, NVIDIA, Capital One, and Cognition executives
Sea Limited announced as first ASEAN enterprise to adopt NVIDIA Vera Rubin platform at AI Day Singapore
Sources
CoreWeave Twitter
- Agentic coding is an unforgiving workload. Long contexts. High concurrency.
- Two things shipped for CoreWeave Mission Control last month:
- Fully Connected fully took over San Francisco.
- ICYMI: @NVIDIA Vera Rubin NVL72 is available on CoreWeave. @Cognition is running @devindevelopers in production on it, a…
- ICYMI: @nvidia's Ian Buck and @GoldbergChen unveil the NVIDIA Vera CPU live at #FullyConnected26. https://t.co/fGqKdbkK3…
- @nvidia's Ian Buck said it best: "it takes CoreWeave to make it real."
- Our CEO Mike Intrator joined @CNBC this morning to discuss the world's first Vera Rubin deployment, the Vera CPU, and mo…
- Live from #FullyConnected26: @NVIDIA Vera CPU is coming to CoreWeave. The first CPU built for AI agents.
- NVIDIA Vera Rubin NVL72 is available on CoreWeave.
Related
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free