NVIDIA has started shipping Vera at greater scale, a CPU designed for the tasks that support artificial intelligence agents. AWS received one of the first servers after earlier deliveries to companies and AI laboratories.
Why do agents need a specialized CPU?
Agents do more than run models on a GPU. They also search files, query databases, use tools, execute code, and coordinate several tasks. Many of these operations depend mainly on the CPU.
When many agents work at the same time, processor speed can affect how long the system takes to complete each action. NVIDIA presents Vera as a CPU designed to keep several of these tasks running simultaneously.
Vera’s main specifications
Vera includes 88 Olympus cores designed by NVIDIA and compatible with the Armv9.2 architecture.
The CPU also provides up to 1.2 TB/s of memory bandwidth. This capacity is intended to keep data available while many processes use tools, retrieve information, and coordinate operations.
A CPU that works alongside GPUs
Vera is not designed to replace GPUs. The CPU manages general and coordination tasks, while GPUs continue to perform much of the intensive computation required to run artificial intelligence models.
The processor can be used in standalone servers and will also be part of complete platforms such as Vera Rubin, which combines CPUs, GPUs, and other data center components.
Early deliveries and availability
NVIDIA reported on August 27, 2026, that AWS had received its first Vera server. The company had previously delivered systems to Oracle Cloud Infrastructure and laboratories including Anthropic, OpenAI, and SpaceXAI.
The arrival of the first server at AWS does not mean that Vera instances are already available to every customer. NVIDIA and AWS announced plans to bring this infrastructure to cloud services, but they have not yet provided a general availability date or public pricing.
