Nvidia’s RTX Spark marks a notable shift, bringing high-performance AI capabilities to consumer handhelds. The device features a Grace CPU with 20 cores, a Blackwell GPU containing 6,144 CUDA cores, and 128GB of LPDDR5X memory. This consumer release highlights Nvidia’s expanding reach, but the company’s focus on data center infrastructure remains equally critical for the broader AI ecosystem.
The Vera CPU serves as the processor component of the Vera Rubin platform, with the Rubin GPU handling graphics duties. Nvidia claims Vera delivers an average speedup of 1.8x over leading x86 CPUs, though the company stopped short of naming specific competing chips in its initial announcement.
This processor is enormous in scale. Vera packs 32 cores built on NVIDIA’s proprietary ARM-based architecture, with Spatial Multithreading pushing the thread count to 128 per socket. The chip supports up to 1TB of LPDDR5X RAM, capable of hitting 1.2TB/s in bandwidth, a figure that matters enormously for AI inference workloads.
Standalone deployment is one option for Vera. The CPU can run agentic AI tasks, reinforcement learning, data processing, and analytics on its own. Nvidia has also built the Vera CPU Rack, which stuffs 256 processors into one system for a total of 22,528 cores and 45,056 threads.
Vera can also act as a host CPU paired with Rubin GPUs. The NVIDIA Vera Rubin NVL72 demonstrates this approach, combining 18 Vera CPUs with 36 Rubin GPUs. Communication between the processors runs at 1.8TB/s through Nvidia’s NVLink-C2C interconnect.
Several major buyers have indicated interest in the platform. While Anthropic, OpenAI, and SpaceX have not publicly confirmed specific production deployments of Vera CPUs, industry analysts expect hyperscalers including ByteDance, CoreWeave, and Oracle Cloud Infrastructure to adopt the technology. The broader demand for ARM-based server chips continues to grow as companies seek alternatives to traditional x86 designs.
Hardware partners are lining up to build standalone Vera systems. Dell, HP, Lenovo, and Supermicro are involved, alongside Asus, Compal, Foxconn, Gigabyte, Pegatron, Quanta Cloud Technology, Wistron, and Wiwynn. The New York Stock Exchange has also expressed interest in the technology, given its need to handle 1.1 trillion messages daily, though reports of a specific partnership with Redpanda on Vera-based infrastructure remain unconfirmed.
The arrival of Vera and the broader Rubin platform comes as Nvidia continues expanding its presence across computing tiers. This push includes efforts to bring ARM-based designs to new form factors, as seen in coverage of the Nvidia Computex 2026 Keynote Poised to Reveal N1X Arm Chip for Windows Laptops. The industry’s broader ARM ecosystem is also heating up, with the MediaTek’s Dimensity 7500 Debuts With Arm C1 CPU and Speedier NPU showing how ARM designs are gaining traction in mobile devices.
