NVIDIA AI Infrastructure Push: Vera Rubin Racks, Meta/Google/Cisco Partnerships
NVIDIA opened its Vera Rubin rack designs at the OCP Global Summit and announced major AI networking partnerships with Meta, Google, and Cisco, targeting 12.8Tb/s bandwidth and clusters beyond 100,000 GPUs.
What we know
NVIDIA contributed Vera Rubin rack and compute tray designs through its MGX 1.2 architecture, supporting racks up to 250kW, per an account on X. With Cisco, NVIDIA announced USFP optics targeting 12.8Tb/s — 4x bandwidth density improvement — and Meta is adopting Spectrum-6 switches to scale AI clusters beyond 100,000 GPUs, per another account on X. The original announcement drew 114 likes and 19.7K views via another account on X. Goldman simultaneously argued hyperscalers deserve their premium valuations because 'powered compute is forever' per another account on X.
By the Global Eye 24 desk, with AI assistance. Updated as the story develops. Spot an error? Write to corrections@globaleye24.com.