Super Micro Computer, Inc., an innovator in enterprise-level computing, storage, and networking, has expanded its Data Center Building Block Solutions® (DCBBS) offerings through the development of the NVIDIA Vera CPU rack.
Tailored specifically to handle the compute-intensive needs of agentic AI, reinforcement learning (RL), and big data analysis, the complete and fully integrated, 100% direct liquid-cooled system provides maximum compute density for CPUs that can be used for the simultaneous execution of 22,500 parallel sandbox executions from one rack.
Built to function as either a stand-alone high-density compute cluster or in complementation with the existing NVIDIA Vera Rubin NVL72 system currently shipping, the solution provides the necessary host processing power needed by the new generation of AI factories to run the branched multi-agent environment.
Busting the Bottleneck of Host Compute in the Era of Autonomous AI Agents
As AI becomes more advanced and moves from inference to autonomous multi-step agentic reasoning, there is an evolution of needs with respect to the required infrastructure. Although GPUs are still required to produce tokens and synthetic and code blocks, autonomous agents require host CPU computing power to run external applications, launch isolated execution sandboxing, and analyze intermediate results.
In modern AI factories housing thousands of accelerators, standard compute architectures often struggle to run the millions of concurrent agent environments needed to maintain full pipeline throughput.
Supermicro addresses this compute imbalance by packing 256 NVIDIA Vera CPUs into a single rack-scale framework, supplying 22,528 high-performance cores and 45,056 execution threads. Thermal dissipation across the high-density layout is managed through Supermicro’s proprietary DLC-2 direct liquid cooling stack, ensuring peak sustained performance without thermal throttling. High-throughput data movement and in-silicon isolation are delivered natively via integrated NVIDIA BlueField-4 Data Processing Units (DPUs).
Also Read: Lenovo Enhances Hybrid AI Advantage Portfolio to Transition Enterprise AI Experimentation into Scalable ROI
“As agentic AI transforms data center requirements, Supermicro is extending its DCBBS portfolio with fully integrated infrastructure,” said Charles Liang, president and CEO of Supermicro. “For customers seeking air-cooled deployments, Supermicro’s NVIDIA Vera CPU portfolio includes air-cooled 1U and 2U systems, in addition to rack-scale liquid cooling options. Together with NVIDIA, we help accelerate time to deployment with a common purpose-built architecture for secure, scalable agentic AI.”
“As enterprises scale agentic AI and reinforcement learning, supporting thousands of concurrent agent workflows requires a new class of high-density CPU infrastructure,” said Chris Marriott, vice president of enterprise platforms at NVIDIA. “NVIDIA 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.”
Standardized NVIDIA MGX Modular Engineering and Advanced Power Delivery
Built around the third-generation NVIDIA MGX modular reference architecture, the Vera CPU rack aligns with the physical and mechanical footprint of the Vera Rubin NVL72. This shared structural blueprint simplifies infrastructure planning, lowers procurement overhead, and accelerates overall time-to-value for enterprise data centers.
To accommodate varying facility designs, Supermicro provides flexible power delivery options:
Conventional AC Power Configuration: Facilitates straightforward drop-in integration into existing enterprise data center rows.
800 VDC Sidecar Power Rack Integration: Leverages integrated Battery Backup Units (BBUs) directly in place of bulky centralized Uninterruptible Power Supply (UPS) setups. This direct DC distribution architecture reduces facility floor-space consumption, cuts power conversion stages, and lowers operational expenditures (OpEx) while raising overall energy efficiency.
Runtime Governance Enforcement via the NVIDIA OpenShell Architecture
In order to protect enterprise networks against rogue agents, the architecture implements native support for the NVIDIA OpenShell Reference Architecture, creating a security perimeter out-of-the-box:
Least Privilege Policy-Driven Execution: NVIDIA OpenShell operates on the NVIDIA Vera host processors, providing security policies and role-based constraints to active agents at runtime.
Isolated Hardware Auditing: NVIDIA Sentry operates on its own, independent of the BlueField-4 DPUs, providing constant telemetry and out-of-the-box monitoring. In case any agent violates the predefined security constraints, the system can isolate and quarantine the agent in milliseconds without consuming host CPU cycles.