SpaceXAI, a company specializing in artificial intelligence systems development, has announced an expansion of its use of Nvidia's computing infrastructure. The primary focus of the partnership will be the deployment of Vera processors for developing agentic AI systems and the further advancement of the Grok language model. In the long term, the company plans to scale the Vera Rubin computing platform to gigawatt-level capacity and adapt it for use in outer space.
The Vera CPU was designed specifically to optimize the performance of agentic artificial intelligence systems. Unlike graphics processing units (GPUs), which traditionally handle the direct mathematical computations of neural network models, the Vera central processor is built to accelerate auxiliary tasks during the inference stage. These include executing program code, interacting with external tools, processing data arrays, orchestrating agent actions, and running simulations. The chip's architecture features 88 proprietary Olympus cores and integrated LPDDR5X memory with a bandwidth of up to 1.2 TB/s. According to Nvidia engineers, in specific workloads such as reinforcement learning and data processing, the processor completes tasks 1.8 times faster compared to comparable x86 architecture chips. This specialization allows GPU workloads to be offloaded: while the GPU processes the model itself, the CPU coordinates agent actions, moves data, and generates new tasks for the accelerators, increasing the overall efficiency of AI infrastructure.
Alongside the deployment of new processors, SpaceXAI intends to scale the infrastructure for its Grok model on the Vera Rubin platform. The company has set a goal of growing its computing capacity to gigawatt scale. According to stated plans, by the end of 2026 the total computing capacity of the infrastructure should exceed 2 GW, and by the end of 2027 approach the 10 GW mark. SpaceX founder and CEO Elon Musk previously confirmed that the company will use Nvidia hardware as the exclusive foundation for its AI infrastructure, describing the Vera Rubin architecture as the most optimal solution available on the market. He noted that the developed design is significantly simpler, more economical, denser in layout, and lighter in weight compared to traditional server racks.
The partnership extends beyond ground-based data centers. SpaceXAI is developing the first Starmind AI satellite, whose computing system will be based on an optimized version of the NVIDIA Vera Rubin NVL72 platform. Engineers will need to adapt this hardware for operation in outer space, which requires strict adherence to constraints on power consumption, cooling systems, communication channel bandwidth, and overall hardware reliability. The launch of the first Starmind AI satellite is scheduled for 2027.
As a result of this expanded partnership, SpaceX becomes a major customer for Nvidia's comprehensive AI architecture. Demand is being generated not only for graphics processors, but also for the company's proprietary central processors, networking equipment, and software platform. It is worth noting that in August of this year, the SpaceXAI team unveiled Grok 4.6, which is positioned as a new flagship development focused on executing long-horizon agentic tasks.