SpaceX is building a constellation of AI satellites designed to process artificial intelligence workloads in low Earth orbit, with the first commercial services targeted for 2028. The project, called Starmind, represents a radical shift in how companies might solve one of AI's biggest challenges: the enormous power consumption and cooling demands of training and running large language models (LLMs). Instead of constructing massive data centers on Earth that consume gigawatts of electricity and require intensive water cooling, SpaceX aims to move the compute itself into space, where solar power is near-constant and the vacuum provides passive thermal dissipation.

What Exactly Is Starmind?

Starmind is not a separate company or publicly traded stock. It operates as an internal division of SpaceX under the SpaceX AI brand, which was created after SpaceX acquired xAI in early 2026 and integrated it with existing Colossus data center operations. The project was officially announced on July 9, 2026, when Elon Musk posted a single word on X: "Starmind." Nvidia was confirmed as the exclusive chip partner on August 4, 2026.

The core concept is straightforward but ambitious: instead of building ever-larger data centers on Earth, SpaceX wants to launch a megaconstellation of up to 1 million orbital AI satellites. Each satellite would carry the same computing hardware found in ground-based data center racks, but operating in an environment with unique advantages. Musk described it as "the lowest-cost location for AI compute within two to three years" in a June 2026 presentation.

Musk

How Do the Orbital AI Satellites Actually Work?

• Compute Payload: Each first-generation AI1 satellite carries Nvidia Rubin GPUs (graphics processing units) and Vera CPUs (central processing units), configured as an optimized Nvidia NVL72 computer. This is the same rack-scale system that hyperscalers like Microsoft and Google are installing in ground-based data centers.
• Power System: The satellites feature a 70-meter solar wingspan (roughly two-thirds the length of a Boeing 747-8 fuselage) that generates approximately 210 kilowatts of power. They maintain an average compute payload of 120 kilowatts during orbital operation, with peak capacity reaching 150 kilowatts when battery-assisted.
• Thermal Management: Instead of water cooling, each satellite deploys a 110-square-meter liquid radiator that dissipates heat directly into the vacuum of space. This eliminates the need for water-intensive cooling systems required by ground-based data centers.
• Data Connectivity: Satellites communicate with each other using petabit-class inter-satellite laser links and transmit data back to Earth via the existing Starlink constellation, which serves as the backhaul network.
• Orbital Position: The satellites operate in sun-synchronous orbit at approximately 600 kilometers altitude, ensuring near-constant solar illumination and stable thermal conditions.

How Powerful Is a Single Starmind Satellite?

The computing power of a single AI1 satellite is substantial. SpaceX describes each one as equivalent to one full Nvidia NVL72 rack. To put this in perspective, a single Starmind satellite delivers approximately 25 times the AI compute power of an Nvidia H100 GPU, which has been the industry standard for AI training and inference. This means that even a modest initial constellation of 1,000 AI1 satellites would represent 25,000 H100-equivalent units of compute, roughly comparable to a mid-scale AI training cluster.

Each satellite weighs approximately 2.8 metric tons and achieves an efficiency ratio of 70 kilowatts per metric ton. SpaceX intends to launch 30 to 50 Starmind satellites per Starship flight, with thousands targeted annually starting in late 2027.

When Will Starmind Actually Launch?

The timeline is aggressive but realistic given SpaceX's track record with Starship development. Two AI1 prototype satellites are targeted for early 2027. Volume production will begin at the Gigasat factory in Bastrop County, Texas, in late 2027. Commercial service is targeted for 2028, though the full constellation of up to 1 million satellites would take years to deploy.

The FCC filing for the project was submitted on January 30, 2026, requesting authorization for up to 1 million satellites. This regulatory approval is a critical milestone that has already been achieved, clearing the path for development and testing.

Why Does This Matter for AI Infrastructure?

The AI industry faces a fundamental constraint: data centers consume enormous amounts of electricity. Training large language models can cost hundreds of millions of dollars in compute resources alone. Ground-based data centers also face challenges with power grid capacity, water availability for cooling, and land use restrictions. By moving compute into orbit, SpaceX sidesteps all three constraints simultaneously. Solar power is abundant and near-constant in sun-synchronous orbit, the vacuum provides free thermal dissipation, and there are no land use or power grid constraints.

This approach could fundamentally reshape how hyperscalers like Microsoft, Google, and Meta plan their AI infrastructure investments. Rather than competing for limited power grid capacity and water resources on Earth, companies could potentially lease compute capacity from Starmind satellites, paying for processing power delivered via the Starlink network.

How Can Investors Get Exposure to Starmind?

There is no separate Starmind stock ticker. The only publicly traded paths to Starmind exposure are SpaceX (NASDAQ: SPCX), which owns and is building the constellation, and Nvidia (NASDAQ: NVDA), which is designing the compute payload for each satellite. When Starmind begins generating revenue in 2028, that revenue will flow into SpaceX's AI segment, which already appears in quarterly financial statements alongside Connectivity (Starlink) and Space (launch) divisions.

It is important to note that multiple STARMIND-branded tokens trade on decentralized exchanges on the BNB Chain. These are community-created cryptocurrencies with no affiliation to SpaceX or Nvidia and carry no exposure to the actual satellite program. The only legitimate public market exposure is through SPCX or NVDA on the Nasdaq.

Starmind represents one of the most ambitious infrastructure projects in SpaceX's history. If successful, it could reshape how the AI industry approaches compute infrastructure, moving beyond the constraints of Earth-based power grids and water systems. The first prototypes will arrive in orbit within months, and the commercial implications could be profound.