NVIDIA Opens Vera Rubin Architecture to OCP, Unites Hyperscalers on AI Factory Standards
NVIDIA Blog · rss · 2026-10-12
At the OCP Global Summit, NVIDIA contributed its Vera Rubin rack and compute tray designs to the Open Compute Project via the open MGX architecture, giving system builders a common foundation for next-gen AI factory infrastructure.
Key announcements:
- With AMD, Google and Meta, NVIDIA published the Open Data Center Specification v0.7, a unified blueprint for AI factory power, cooling and physical layout; a Grid Resilience Spec (with Google/Meta/Microsoft) requires restoring to near-full load within 2 seconds of grid disturbances.
- MGX 1.2 adds rack-scale blind-mate liquid cooling, a liquid-cooled 5,000A busbar, up to 250kW rack deployments, plus a Vera CPU rack for the agentic AI era (tool calling, orchestration, sandboxed execution).
- NVIDIA open-sourced new DSX AI Factory components: Config Manager (automated switch config), Cluster Readiness Engine (stress-testing GPU clusters with real workloads), Rack Management Service, and a Cluster API Provider that turns raw hardware into production clusters in hours.
- With Cisco: the USFP 400Gb/s module standard; with Google: Falcon Transport on future ConnectX-10 SuperNICs; Meta advanced Spectrum-X/6 Ethernet networking; Visionbay unveiled its 800VDC K-1 AI factory platform.
The throughline: NVIDIA's extreme codesign plus open standards to accelerate deployment and improve tokens-per-watt performance.
More from Infra
- Intel details GPU D-series: 480GB memory, 1.3 PFLOPS FP8 for AI inference economics — AIFlow_ML · 2026-10-13
- Spain moves to tighten data center rules over massive water and power use — Polymarket · 2026-10-13
- OpenBMB open-sources MiniCPM-V 4.7: 35B multimodal models running on phones — Aiden_Tech_Ai · 2026-10-13
- Open-source desktop agent dodaios runs local Qwen models via ollama — and shows where 16GB models fail — InternationalGap3698 · 2026-10-13
- Sipeed teases MaixStation powered by AXEROA AX8860: 256TOPS, 128GB LPDDR5X, aiming at DGX Spark — Marcuss2 · 2026-10-13
- Learned optimizers without 4,000 TPU months: new work claims general-purpose trained optimizers are finally practical — burkov · 2026-10-13