Supermicro announced its Vera CPU rack on October 22, 2026, adding a fully liquid-cooled, 256-processor system to its Data Center Building Block Solutions portfolio. The company says the rack supports up to 22,500 concurrent AI agent sandboxes and is available for quotes now, either as a standalone platform or a companion to its NVIDIA Vera Rubin NVL72 systems.
Key Takeaways
- The rack combines 256 NVIDIA Vera CPUs, 22,528 cores and 45,056 threads for agent execution and other CPU workloads.
- Shared rack infrastructure supports deployment alongside Vera Rubin NVL72, but the CPU system can also operate independently.
- Supermicro is accepting quote requests; the announcement does not specify a shipping date for the new rack.
Table of Contents
- Vera CPU rack targets agent execution
- Shared infrastructure, separate deployment options
- Runtime controls and deployment services
- Conclusion
Vera CPU rack targets agent execution
The new hardware supplies dedicated CPU capacity for the work surrounding AI model output. While GPUs generate tokens, code and queries, CPU-based sandboxes execute actions, run tools, assess results and coordinate workflows. Supermicro positions the rack for agentic AI, reinforcement learning and data processing.
Its configuration combines NVIDIA Vera CPUs with BlueField-4 data processing units and Supermicro's DLC-2 direct liquid cooling technology. The company describes the rack as 100% liquid-cooled, with BlueField-4 providing networking and hardware-based security.
The introduction adds a specific rack-scale product to Supermicro's Vera portfolio, which also includes air-cooled 1U and 2U systems.
Shared infrastructure, separate deployment options
The CPU rack uses the third-generation NVIDIA MGX design, giving it a common infrastructure architecture with Supermicro's Vera Rubin NVL72, which the company says is already shipping. That shipping status applies to NVL72; the newly announced CPU rack is available for quotes.
Customers can choose a traditional AC feed or an 800 VDC sidecar power rack. The latter supports battery backup units instead of a conventional uninterruptible power supply. Supermicro says this approach can reduce floor-space requirements and improve power-conversion efficiency.
For paired installations, Supermicro offers an optional design incorporating NVIDIA Spectrum-X Ethernet, hot-aisle containment, redundant in-row coolant distribution units and rear-door heat exchangers.
Runtime controls and deployment services
The rack supports NVIDIA's OpenShell Reference Architecture. OpenShell runs on Vera CPUs to apply least-privilege, policy-based controls during agent execution, while NVIDIA Sentry runs on BlueField-4 DPUs as an out-of-band monitor.
According to the announcement, Sentry enforces policy, supplies attested telemetry and can quarantine an agent that exceeds its permitted boundaries within milliseconds.
Standalone uses extend beyond agent sandboxing to database operations, cluster management, Slurm and Kubernetes scheduling, data analytics and large software builds. Supermicro says one team can handle site surveys, design, integration and testing, on-site deployment and ongoing support for either standalone or paired installations.
Conclusion
The announcement adds a dedicated CPU rack option for customers building agent infrastructure, with standalone and NVL72-paired configurations. Quote availability is confirmed, but a shipping date for the new rack was not provided.
Read More News
Disclaimer
This FinanceVelix article provides general information, not personalized financial, investment, legal, or tax advice. Market prices, interest rates and economic conditions can change; investments carry a risk of loss. Facts reflect the material available when published and may change. Check important information independently and consult a qualified professional when needed. Mentioning an asset, company, or service is not an endorsement.
