The deep collaboration between SpaceX and NVIDIA is advancing to a new stage.

NVIDIA CEO Jensen Huang recently stated publicly that he enjoys working with Musk and revealed that the engineering teams of both companies are jointly developing a new server architecture designed specifically for the space environment while also being suitable for ground deployment.

Huang said:

I am delighted to collaborate with Elon. He is using our technology to train and deploy his own AI models, and our two engineering teams are working very closely together to build a new type of server, an entirely new architecture that could perform well both in space and on Earth.

Musk then reposted and responded on X, saying:

The Star Mind computer jointly designed by SpaceX and NVIDIA far exceeds almost everyone's expectations and will enter mass production next year. Working with the exceptionally talented NVIDIA (NASDAQ: NVDA) engineering team is an absolute pleasure.

This statement quickly pushed market attention on the project higher and further reinforced NVIDIA's core position in the AI infrastructure sector.

Huang: Musk Has Unique Advantages in AI and Computing Power

When discussing this collaboration, Huang emphasized Musk's comprehensive advantages in AI training, model deployment, and infrastructure construction.

He pointed out that Musk not only uses NVIDIA's technology to train and deploy AI models but also rents out computing power externally through Starlink, while demand for xAI's Grok Bot is "rising rapidly."

Huang said:

His combination across these three things is quite unique. He loves the intersection of computing and physical infrastructure construction, and no one understands energy better than he does.

Huang also compared Musk with Google, suggesting similarities in their closed-loop capabilities spanning self-built computing power, self-developed models, and open commercial use.

Space Data Centers: Targeting the Two Major Pain Points of Energy and Heat Dissipation

The Star Mind project adopts a combined architecture of NVIDIA's next-generation Vera CPU and Rubin GPU.

Each Star Mind AI1 satellite carries a computing structure equivalent to a complete NVIDIA NVL72 rack server, containing 36 Vera CPUs and 72 Rubin GPUs.

NVIDIA stated that the Space-1 Module's AI computing capability is approximately 25 times that of the previous-generation H100 GPU, a significant performance improvement. The mass production base is expected to be located at SpaceX's Gigasat manufacturing facility in Bastrop, Texas, and Musk confirmed that mass production will begin in 2027.

The core application scenario targeted by this collaboration is deploying data centers into low Earth orbit.

The space environment offers conditions in energy supply and thermal management that ground-based data centers find difficult to match: orbiting satellites can obtain ample solar energy through large deployable solar wings while also releasing heat directly in the vacuum environment through radiative heat sinks, thereby escaping the dual constraints of the ground power grid and cooling water resources.

In terms of data transmission, processed information will be transmitted directly to the ground via Starlink inter-satellite laser links, forming a complete closed loop from computing power production to data delivery.

Huang positioned this as "a brand-new market we will explore together," indicating that the partnership between the two sides has extended from pure hardware supply toward jointly defining a new type of computing infrastructure.