The large-scale expansion of SpaceXAI into the field of artificial intelligence is gaining a new technological dimension. The company is officially expanding its collaboration with NVIDIA, integrating Vera processors into its infrastructure for developing agentic systems and the Grok model. This is not just about modernization—it is a strategic move that involves scaling the Vera Rubin platform to gigawatt-level capacities and, most ambitiously, moving computation beyond Earth.
Why CPUs Have Become Critical for AI Agents
The key point here is a paradigm shift. While previously the main load fell on GPUs responsible for model inference, NVIDIA is now betting on processors as the "conductors" of agentic systems. The Vera CPU, equipped with 88 proprietary Olympus cores and LPDDR5X memory with bandwidth up to 1.2 TB/s, is designed to accelerate code execution, data processing, and agent orchestration. In my assessment, this is a logical response to the bottleneck of modern AI: while the GPU is busy with computations, the CPU must rapidly coordinate actions, launch tools, and pass tasks back to the accelerators. NVIDIA claims that in a number of agentic AI and reinforcement learning scenarios, Vera delivers 1.8 times faster processing compared to x86 counterparts—a serious argument for infrastructure giants.
Gigawatts on Earth and Satellites in Orbit
SpaceXAI is not limited to terrestrial data centers. The company has announced the first Starmind AI satellite, whose computing system will be built on an optimized version of NVIDIA Vera Rubin NVL72. This is a bold experiment: adapting the platform for space requires solving non-trivial challenges related to power consumption, cooling, and reliability. Meanwhile, on Earth, the company intends to end 2026 with more than 2 GW of computing power, and approach 10 GW by the end of 2027. Elon Musk, commenting on the choice of architecture, emphasized its superiority: "Our design is significantly simpler, cheaper, denser, and lighter than a traditional rack."
In essence, we are witnessing the formation of a unified technological chain: the Vera CPU handles agentic logic, Vera Rubin handles training and inference for Grok, and the NVL72 will become the foundation for orbital computing. The launch of the first satellite is scheduled for 2027, making this project one of the most ambitious in the industry.
My analysis: This is not just an expansion of the partnership, but a signal that the future of AI lies in hybrid architectures, where CPUs and GPUs work in tandem, and infrastructure extends beyond traditional data centers. SpaceXAI is becoming not merely a client for NVIDIA, but a strategic partner testing new horizons—all the way to space. The only question is how quickly other market players can follow this example, since the bet on agentic AI is already obvious.