Nvidia Vera CPU wins early data center demand at labs now

Nvidia Vera CPU secured early users at Anthropic, OpenAI, and SpaceX, with full production set for the third quarter following Jensen Huang’s announcement at Computex in Taiwan.

Early adopters and production timeline

Jensen Huang said Anthropic PBC, OpenAI, and SpaceX will be among the first to run the new central processing units in their data centers. Nvidia’s upcoming microprocessor, branded Vera, moves into full production in Q3.

The disclosure puts recognizable AI developers on the customer list ahead of volume availability. It also positions Nvidia (NVDA) to broaden its data center stack beyond GPUs.

Positioning for the inference-heavy cycle

Nvidia is signaling it is aware and already ahead of shifting AI data center patterns. As workloads tilt from training to running models and related services, demand can lean harder on general-purpose CPUs.

That dynamic has fueled concern that accelerators might become less crucial for some stages of deployment. By naming first-wave users for Vera, Nvidia is arguing its platform remains central even as compute mix evolves.

Broader stack and competitive context

Vera slots alongside Nvidia’s accelerators and networking, aiming to keep developers within the company’s ecosystem as architectures diversify. The strategy echoes prior moves with Grace CPUs and NVLink interconnects to bind components tightly.

Rivals are active. AMD (AMD) competes with EPYC in CPUs and Instinct accelerators, while Intel (INTC) fields Xeon for inference and training adjacencies; Arm-based designs continue to win design wins across hyperscale.

Signals for customers and investors

Early uptake from Anthropic, OpenAI, and SpaceX gives Nvidia reference deployments that matter in procurement cycles. Labs and enterprises tend to follow established users when evaluating new silicon for production.

Named customers reduce uncertainty around first-generation ramp, even without public performance or pricing details. For investors tracking NVDA’s data center revenue mix, a CPU foothold offers incremental levers as accelerator growth matures.

What the shift could change in data centers

Inference, retrieval, and orchestration layers can consume substantial CPU time depending on model size and latency targets. Vera aims at those roles while leaving dense matrix math to accelerators.

If general-purpose compute carries a larger share of deployed AI cycles, total cost of ownership and energy profiles could look different from training-heavy builds. Integration and software tooling will determine how much value Vera captures.

Next milestones

The next checkpoint is volume production in the third quarter and the pace of deployments at the named labs. Any disclosures on performance-per-watt, memory bandwidth, or socket compatibility will be scrutinized.

Procurement updates from large AI developers and hyperscalers will indicate whether Vera broadens beyond initial users. Watch for software support inside Nvidia’s stack and how it dovetails with existing inference frameworks.

Nvidia’s message is straightforward: as AI compute shifts toward running models at scale, the company intends to supply the CPU as well as the accelerator. The early customer slate at least opens that door.