Supermicro launches rack with 256 NVIDIA Vera CPUs for agentic AI at scale

Supermicro announced on October 22, 2026, the NVIDIA Vera CPU Rack, a rack-scale system with 256 NVIDIA Vera CPUs aimed at agentic AI, reinforcement learning and data processing. According to the company, a single rack delivers 22,528 cores and 45,056 threads and supports up to 22,500 concurrent agent sandboxes — the isolated environments where agents execute code, use tools and evaluate results.

The product's logic lies in how work is split inside the AI factory: GPUs remain the engine that generates tokens, code and queries, while CPU-based sandboxes execute actions, evaluate results, run tools and orchestrate workflows. At the scale of AI factories with hundreds or thousands of GPUs, Supermicro argues, dedicated CPU compute is needed to host thousands — or millions — of parallel execution environments for agents.

Liquid cooling and security for agents

The rack is 100% liquid-cooled, using Supermicro's DLC-2 direct liquid cooling stack, and integrates NVIDIA BlueField-4 DPUs for networking and in-silicon security. On the software side, it natively supports the NVIDIA OpenShell Reference Architecture, which applies least-privilege, policy-based controls to agents at runtime, and NVIDIA Sentry, which runs on the BlueField-4 DPUs as an out-of-band monitor with attested telemetry and the ability to quarantine an agent that moves outside its permitted boundaries within milliseconds.

The foundation is the third-generation NVIDIA MGX design, creating common infrastructure with Vera Rubin NVL72: the rack can run standalone or alongside GPU racks, with a defined network topology over NVIDIA Spectrum-X Ethernet when combined. For power, customers can choose a traditional AC feed or an 800 VDC sidecar power rack with battery backup units (BBU) instead of a conventional UPS — a trade-off that, according to the company, frees data center floor space and can improve end-to-end power conversion efficiency. As a standalone platform, it also accelerates agent orchestration, databases, cluster managers such as Slurm and Kubernetes, data analytics and large-scale software build pipelines.

"As agentic AI transforms data center requirements, Supermicro is extending its portfolio with fully integrated infrastructure," said Charles Liang, president and CEO of Supermicro. Chris Marriott, vice president of enterprise platforms at NVIDIA, said Vera CPUs and BlueField-4 DPUs, integrated with Supermicro's liquid-cooled DCBBS architecture, give customers "the dedicated compute and hardware-based security to scale the execution of agentic workloads across their AI factories."

The rack is available for quotation now, but Supermicro has not disclosed a shipping date. The announcement follows the start of shipments of Supermicro's own Vera Rubin NVL72 in September and an upward revision of its fiscal first-quarter revenue forecast to $18.5 billion to $19.5 billion — a backdrop that helps explain why the company is betting on dedicated agent infrastructure, not just GPUs.

Image credit: Supermicro, via the official press release (PR Newswire).

Fontes

Transparency: this article was produced with the help of artificial intelligence, cross-checked against public posts on X and internet sources. Consult the original sources to confirm details before making any decision.

Muse