NVIDIA Announces Vera CPU With 88 cores, Rubin Platform follows
NVIDIA has announced the Vera CPU alongside its Vera Rubin platform at GTC 2026, combining new CPU, GPU, networking and storage hardware into a rack-scale design aimed at agentic AI, reinforcement learning and inference. The company describes Vera as its first CPU built specifically for these workloads, while Vera Rubin packages seven chips and multiple rack types into one modular platform for AI factories.
Source: NVIDIA
Vera features 88 custom NVIDIA Olympus cores, with each core able to run two tasks through what NVIDIA calls Spatial Multithreading. The chip uses LPDDR5X memory and delivers up to 1.2 TB/s of memory bandwidth. NVIDIA says Vera is 50% faster and twice as efficient as traditional rack-scale CPUs, and when paired with Rubin GPUs over NVLink-C2C it can reach 1.8 TB/s of coherent bandwidth, far above PCIe Gen 6.
Source: NVIDIA
NVIDIA is also introducing a Vera CPU rack built around 256 liquid-cooled CPUs. According to the company, a single rack can support more than 22,500 concurrent CPU environments at full performance, targeting orchestration, agentic inference, reinforcement learning, data processing, storage management and HPC workloads. Vera-based servers will also integrate ConnectX SuperNICs and BlueField-4 DPUs, with both single- and dual-socket systems planned by partners.
Source: NVIDIA
The broader Vera Rubin platform adds Rubin NVL72 GPU racks, BlueField-4 STX storage racks, Spectrum-6 SPX Ethernet racks and a new LPX inference rack. NVIDIA says the NVL72 design combines 72 Rubin GPUs and 36 Vera CPUs, and can train mixture-of-experts models with one-fourth the number of GPUs required by Blackwell, while also delivering up to 10x higher inference throughput per watt and one-tenth the cost per token. The company also introduced DSX Max-Q and DSX Flex software, which it says can improve power allocation across AI factories and expand deployment options in power-constrained data centers.
Source: NVIDIA
Support for the platform spans cloud providers, AI labs and server makers. NVIDIA named Amazon Web Services, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, Alibaba, ByteDance, CoreWeave, Lambda, Nebius, OpenAI, Anthropic, Meta and Mistral AI among the companies working with Vera or Vera Rubin systems. On the hardware side, Dell, HPE, Lenovo, Supermicro, ASUS, Foxconn, GIGABYTE, QCT, Wistron and Wiwynn are among the partners set to deliver systems, with NVIDIA saying Vera is already in full production and partner availability will begin in the second half of 2026.
Source: NVIDIA PR
