Story perspectives

NVIDIA's Vera CPU Enters Mass Production, Powers OpenAI, SpaceX

7/22/2026

Summary

NVIDIA Launches Vera

NVIDIA announced its Vera CPU has entered mass production and is being shipped to OpenAI, Anthropic, and SpaceX.
Vera contains 88 Olympus cores, delivers 1.2 TB/s bandwidth, and claims a 50 % single-core speed gain over x86 CPUs.
SPEC CPU 2026 benchmarks show Vera is 1.8 times faster than AMD EPYC Turin on agentic tasks.
CoreWeave says the Vera Rubin NVL72 provides tenfold performance-per-watt over Grace Blackwell on DeepSeek-R1.
NVIDIA forecasts a $200 billion server CPU market.

Summary

Nvidia Demonstrates 800VDC

Nvidia demonstrated an 800VDC sidecar that converts AC to 800VDC, powering Vera Rubin racks that draw 200 kW and using “cooling” with a heat exchanger effective to 45 °C.
Nvidia says a Vera Rubin rack powers on in 47 minutes after truck arrival, and that assembly installs a compute tray in one minute.
The SuperLab enables rapid hardware testing while Dutch activists released chemical-filled balloons at a nearby Microsoft data centre the next day.

History

Historical Context

NVIDIA’s push to use GPUs for general computing began when Stanford PhD student Ian Buck, supported by DARPA, turned his Brook language into CUDA, which NVIDIA created in 2004 and released in 2007 with architect John Nickolls. CUDA’s success in scientific and AI workloads set the stage for today’s Vera Rubin CPU, targeting the $200 billion server market.

1 of 3

NVIDIA Launches Vera

NVIDIA announced its Vera CPU has entered mass production and is being shipped to OpenAI, Anthropic, and SpaceX.
Vera contains 88 Olympus cores, delivers 1.2 TB/s bandwidth, and claims a 50 % single-core speed gain over x86 CPUs.
SPEC CPU 2026 benchmarks show Vera is 1.8 times faster than AMD EPYC Turin on agentic tasks.
CoreWeave says the Vera Rubin NVL72 provides tenfold performance-per-watt over Grace Blackwell on DeepSeek-R1.
NVIDIA forecasts a $200 billion server CPU market.

1 / 3