Nvidia Vera CPU Enters Full Production with Anthropic, OpenAI, and SpaceX as First Customers

Nvidia CEO Jensen Huang announced at Computex 2026 in Taipei that Anthropic, OpenAI, and SpaceX are among the first major customers for Vera, Nvidia’s first standalone data center CPU. The chip enters full production in Q3 2026, according to Bloomberg.

Vera has been in production since March 2026. Nvidia VP Ian Buck hand-delivered the first systems to Anthropic, OpenAI, SpaceX, and Oracle Cloud Infrastructure, according to multiple reports.

Direct Competition with Intel and AMD

Vera is Nvidia’s first CPU that goes head-to-head with Intel’s Xeon, AMD’s Epyc, and Amazon’s in-house Graviton processors. Huang claimed Vera is 1.8 times faster than x86-based chips on “essential AI-related workloads,” marking the first time Nvidia has made direct performance comparisons against the current data center CPU standard, according to Bloomberg via The Hindu BusinessLine.

The move addresses a shifting dynamic in AI infrastructure. As workloads transition from model training to running inference and agent services, general-purpose CPUs play a larger role alongside GPU accelerators. Nvidia is positioning Vera to capture that adjacent spend rather than cede it to Intel, AMD, or custom silicon programs at hyperscalers.

Broader Computex Announcements

Nvidia also announced updates to its open-source DSX software for data center management, claiming it enables operators to use up to 40% more accelerator chips within the same power budget. The company introduced DGX Station for Windows, high-end workstations for developing and deploying AI software on Microsoft’s platform. Dell and other PC makers will begin selling the machines in Q4 2026, according to Bloomberg via The Hindu BusinessLine.

Silicon Diversification for Agent Workloads

The fact that both Anthropic and OpenAI are early Vera adopters signals that major model providers are diversifying their compute dependencies. Running production agent systems requires reliable throughput on both the GPU (for inference) and CPU (for orchestration, memory management, and I/O). Vera gives Nvidia a product for the CPU side of that equation, reducing the surface area where customers might turn to competing silicon.

Huang continues to argue that Nvidia is the only company producing all components data center operators need, integrated in a way that minimizes deployment complexity, according to Bloomberg.