The AI Compute Crisis

The demand for AI compute capacity is exploding, and the infrastructure to support it is hitting physical limits on Earth. Luke Lango, presenting the XPanse thesis at InvestorPlace, cites the scale of the problem: 1,500 new data centers under construction in America, with lawmakers in at least 14 states introducing legislation to ban new construction. Meta’s Hyperion facility requires 3,650 acres, 7.5 gigawatts of power, and 23 million gallons of water per day.

The problem is threefold: data centers need enormous amounts of land, power, and water. As AI models grow more powerful and more companies build AI infrastructure, these constraints are becoming binding. You cannot build a data center if you cannot get the land, the power, or the water permits. And the NIMBY opposition is growing, with 14 states actively considering bans.

Lango’s solution is what he calls “Elon’s genius solution to the AI power bottleneck”: orbital data centers. As we explain in our full review of the INI XPanse presentation, this is the second pillar of the XPanse thesis. For the overall framework, see our article on XPanse.

The AI1 Satellite: World’s First Orbital Data Center

Lango introduces the AI1 satellite as “the world’s first orbital data center.” The concept is elegant in its simplicity:

Power. Solar arrays harvest “practically unlimited energy from the Sun,” generating 5 to 7 times more power than Earth-based solar panels. There is no nighttime in orbit, no clouds, and no weather to reduce output. The sun shines 24/7.

Cooling. Space is -454 degrees Fahrenheit. “The heat waste radiates directly into the vacuum of space.” No water cooling is needed. This eliminates one of the biggest constraints on Earth-based data centers, which require millions of gallons of water per day for cooling.

Land. There are no land-use regulations in space. No NIMBY protests. No zoning battles. No environmental impact statements for land use. You can deploy as many orbital data centers as you can launch.

Musk has filed with the FCC to launch up to 1 million orbital data centers like the AI1. The scale is enormous, and the regulatory filing shows serious intent. For more on the space infrastructure angle, see our article on space stocks.

The Economics: Launch Cost

The key question for orbital data centers is economics. Launching hardware into space currently costs about $1,500 per kilogram. For orbital data centers to compete with land-based facilities, costs would need to be closer to $200 per kilogram.

But SpaceX has already achieved a 97 percent reduction in the cost of spaceflight. In 2026, SpaceX conducted 166 orbital launches, compared to 92 for China and 17 for Russia. Over 10,000 Starlink satellites already orbit Earth. And Musk is targeting $10 per kilogram for Starship, which would make orbital data centers what Lango calls “a no-brainer.”

If launch costs reach $10 per kilogram, the economics of orbital data centers transform entirely. At that cost, you could launch server hardware into space for less than it costs to build and permit a data center on Earth. The combination of free solar power, free cooling, and no land costs would make orbital data centers cheaper to operate than terrestrial ones. For more on the space economy, see our article on Rocket Lab on Nasdaq.

Big Tech Validation

The orbital data center concept is not just Lango’s speculation. Major technology companies are already pursuing it:

  • Google has “entered deal talks with him” for orbital data centers (reported by WSJ, May 12, 2026)
  • Anthropic has “agreed to pay him $15 billion a year” and expressed interest in orbital AI compute
  • Blue Origin (Jeff Bezos) has asked the government for permission to launch more than 50,000 orbital data centers
  • Starcloud, a startup, has already demonstrated orbital computing: “a satellite with an Nvidia chip that’s currently running Google Gemini model up there in space”

Sam Altman is quoted saying “putting data centers in space is ridiculous,” but Lango refutes this by pointing to the Starcloud demonstration. A satellite with an Nvidia chip running Google Gemini in space is not a theoretical concept. It is a working prototype. Google and Anthropic being in talks with SpaceX suggests they take the concept seriously despite Altman’s skepticism.

The Terafab Connection

Orbital data centers are not standalone infrastructure. They connect to the Terafab, the 100-million-square-foot chip facility in Texas that Lango says will “double American chip production.” The Terafab is “trained by xAI’s Grok, powered by SpaceX’s AI1 satellites, and ultimately staffed by Tesla’s Optimus robots.”

The connection is that the Terafab needs enormous compute power for AI-driven chip design and manufacturing optimization. Orbital data centers can provide that compute power without the land, water, and power constraints that limit Earth-based facilities. The AI1 satellite concept, combined with the Terafab, creates a vertically integrated AI infrastructure that does not exist anywhere else. For more on the Terafab, see our article on Elon Musk’s Terafab.

The Investment Angle

Lango identifies two investment angles related to AI compute capacity:

The $15 space stock (paid pick) makes the solar arrays that SpaceX uses for its orbital data centers. Its hardware was aboard the Orion spacecraft for NASA’s Artemis II mission. If SpaceX scales to 1 million orbital data centers, the demand for solar arrays scales with it. This is a picks-and-shovels play on the orbital data center buildout.

The NASA ETF (free ticker) provides diversified exposure to the space economy, including direct SpaceX holdings. Lango positions this as the conservative option: “I believe the biggest gains will come from the stocks I’ve selected in my special reports.” For more on the investment framework, see our articles on Innovation Investor and growth potential.

Considerations

The orbital data center concept is genuinely innovative and addresses a real problem. The data center crisis on Earth is real: land, power, and water constraints are binding, and opposition is growing. The AI1 satellite concept, with free solar power and free radiative cooling, is an elegant solution. The Starcloud demonstration proves the concept is technically feasible, and the interest from Google, Anthropic, and Blue Origin validates the market demand.

What to consider: The Anthropic “$15 billion a year” figure is presented as a confirmed deal but may be a projection or negotiation position. The FCC filing for up to 1 million orbital data centers is a regulatory aspiration, not a deployment plan. Launch costs need to drop from $1,500/kg to closer to $200/kg for orbital data centers to be economically viable, and while Musk is targeting $10/kg for Starship, that target has not been achieved yet. And the AI compute demand projections, while directionally correct, involve uncertainty about timelines and magnitudes.

What makes the orbital data center thesis compelling is that it addresses a real problem with a technically feasible solution, validated by real demonstrations (Starcloud) and real corporate interest (Google, Anthropic, Blue Origin). The direction is clear: if launch costs continue to decline, orbital data centers become economically viable, and the companies supplying the components (like the $15 solar array stock) are positioned to benefit.

Ready to learn more? Click here to access Luke Lango’s full research.

This is not financial advice. Always do your own research before investing.