SpaceXAI Adopts NVIDIA Vera for Agentic AI
SpaceXAI will deploy NVIDIA Vera CPUs to accelerate its next-generation agentic AI applications. NVIDIA announced the deployment on August 24, 2026.
The move adds purpose-built CPU infrastructure to SpaceXAI’s expanding AI stack. Meanwhile, the company plans to scale its Grok infrastructure on the NVIDIA Vera Rubin platform. It also plans to take an optimized Vera Rubin system into orbit through its Starmind AI satellite programme.
Vera Targets the CPU Work Behind AI Agents
Agentic AI systems require more than GPUs to complete complex tasks. Instead, CPUs handle much of the work between model calls.
For example, agents may orchestrate tools, execute code, process data and run simulations. Therefore, faster CPU execution can reduce delays between individual agent actions. It can also help keep GPUs supplied with work.
NVIDIA designed Vera specifically for these workloads. The processor uses 88 NVIDIA-designed Olympus cores and supports NVIDIA Spatial Multithreading. It also uses high-bandwidth LPDDR5X memory with up to 1.2TB/s of bandwidth.
According to NVIDIA, Vera can deliver up to 1.8 times faster task completion than x86 CPUs across agentic AI, reinforcement learning, and data-processing workloads. However, that figure comes from NVIDIA’s own testing and should not be treated as an independent benchmark.
SpaceXAI already received Vera CPU systems earlier this year. NVIDIA delivered the first systems to SpaceXAI in May, alongside deliveries to Anthropic, OpenAI and Oracle Cloud Infrastructure.
Consequently, the latest announcement represents a move from initial access toward broader deployment. SpaceXAI can use Vera for the CPU-intensive workloads that surround its AI models.
SpaceXAI Expands Grok Infrastructure
At the larger infrastructure level, SpaceXAI plans to expand its Grok AI systems using NVIDIA Vera Rubin. The company is targeting computing capacity at gigawatt scale.
Vera Rubin combines CPUs, GPUs, networking, and software within an integrated architecture. The platform includes NVIDIA NVLink, Spectrum-X Ethernet, BlueField data processing and NVIDIA software. Together, these components target higher performance and better energy efficiency across AI factories.
Furthermore, SpaceXAI intends to use a common architecture across its expanding AI infrastructure. Vera CPUs will handle orchestration and other CPU-heavy tasks. Meanwhile, accelerated computing systems will support training, reasoning and inference.
This approach reflects a broader change in AI infrastructure design. As models become more capable, AI systems increasingly perform multi-step actions rather than simply generate responses. Therefore, infrastructure must support the execution layer as well as model inference.
NVIDIA has also positioned Vera as part of a wider shift toward agentic computing. Its technical research describes agentic systems as workloads that take more actions, call more tools and interact repeatedly with execution environments.
As a result, CPU performance has become increasingly important to overall AI-factory efficiency. Faster execution can shorten feedback loops and help systems complete more tasks within the same infrastructure footprint.
AI Computing Moves From Earth to Orbit
SpaceXAI’s plans extend beyond terrestrial data centres. The company is developing its first-generation Starmind AI satellite around an optimized NVIDIA Vera Rubin NVL72 system.
The planned system would bring accelerated AI computing into orbit. However, orbital computing introduces constraints that conventional data centres do not face.
Power availability, thermal management, bandwidth and physical integration all become critical considerations. Furthermore, space-based systems require high reliability because maintenance and hardware replacement are far more difficult.
NVIDIA and SpaceXAI therefore plan to adapt the Vera Rubin architecture for orbital conditions. The objective is to retain a common NVIDIA architecture and software ecosystem across terrestrial and space-based systems.
The Starmind project remains a planned deployment rather than an operational orbital AI platform. NVIDIA’s announcement also describes several elements of the programme as plans. Therefore, the timeline and final system configuration could change.
Still, the strategy points toward a broader concept for AI infrastructure. SpaceXAI is seeking to connect agentic computing, large-scale AI factories and orbital systems through a common technology foundation.
For NVIDIA, the partnership also expands the potential market for Vera beyond conventional CPU workloads. For SpaceXAI, meanwhile, the technology provides another layer for scaling Grok and future agentic applications.
Ultimately, the deployment highlights an important shift in AI infrastructure. GPUs remain central to model computation, yet CPUs increasingly determine how efficiently agents execute the work around those models. SpaceXAI’s adoption of Vera puts that architectural shift into a large-scale commercial and experimental setting.
