Nvidia Has Shipped New Vera CPU Server Racks
The processor is designed to handle agentic AI workloads and is now being deployed by early enterprise customers.
Updated on Oct. 12, 2026 in Semiconductors
Live Poll
Do you trust that the current shift to specialized AI processors benefits the broader tech ecosystem?
Nvidia has officially begun shipping its Vera CPU racks, which are built to support agentic AI tasks. These units are now in production and are currently being deployed by early partners including Meta and CoreWeave.
Why it matters
The Vera CPU was developed to resolve specific orchestration management limitations within existing NVL72 platforms. It provides the necessary bandwidth and single-threaded performance required for complex agentic tool calling and code execution.
Each Vera CPU rack features over 20,000 cores built on the Arm v9.2-A architecture. It provides a 2-fold performance improvement for agentic workloads and 1.8-fold gains on Terminal Bench compared to the Grace generation.
The players
Nvidia
A designer of graphics processing units, AI accelerators, and high-performance server architectures used in massive data center deployments.
Meta
A hyperscale cloud operator and AI research organization that develops open-source large language models and deploys extensive GPU-based compute clusters.
CoreWeave
A specialized cloud infrastructure provider that focuses on offering GPU-accelerated compute resources for AI, rendering, and simulation workloads.
Amazon Web Services
The world's largest cloud computing platform which provides infrastructure, databases, and AI-as-a-service offerings to global enterprises.
The details
Vera utilizes Nvidia's MGX modular reference architecture, which allows for flexible system configurations within data centers. These CPU racks are designed to function as a side-cabinet alongside an NVL72, an interconnected system of GPUs, to offload the specialized orchestration tasks that bottleneck current hardware. The design focuses on high single-threaded performance to accelerate agentic tool calling, a process where AI agents execute code or interact with external software tools to complete tasks.
Timeline
Spring 2026: Vera entered full production.
October 2026: Nvidia confirmed that Vera CPU racks are shipping.
Second half of 2026: The Vera Rubin Superchip is scheduled for release.
The Tech Race
The introduction of Vera positions Nvidia to better compete for the specialized CPU market within agentic AI infrastructure. This launch follows the Grace generation and aims to optimize the bottlenecked performance of the NVL72 platform as competitors increase focus on AI-centric processing.
Enterprise users and cloud developers gain access to higher performance for agentic workloads through partners like CoreWeave. Amazon Web Services also plans to incorporate the Vera CPU into its cloud offerings, though specific service availability and pricing remain unannounced.
The takeaway
Nvidia is shifting its CPU strategy toward optimizing high-bandwidth, single-threaded performance specifically for agentic AI architectures. Watch for the Vera Rubin Superchip release in late 2026 for further benchmarks on how this architecture performs in combined CPU-GPU clusters.
What happens next
Nvidia is scheduled to release the Vera Rubin Superchip during the second half of 2026.
Further reading
For more background on how hardware advancements drive AI efficiency, see our Semiconductors coverage.
Live Poll
Do you trust that the current shift to specialized AI processors benefits the broader tech ecosystem?
