SpaceX Wants to Launch 1 Million Satellites: The $1.5 Trillion Plan That Will Transform AI

The Project Nobody Thought Possible

I won't sugarcoat it: what SpaceX filed with the FCC on January 31st is arguably the most ambitious infrastructure plan in human history. One million solar-powered satellites operating as orbital data centers. If you'd told me this five years ago, I would have dismissed it as cheap science fiction. Today, with Starship operational and the SpaceX-xAI merger in motion, my verdict is clear: Elon Musk isn't dreaming, he's planning.

The FCC application describes an "Orbital Data Center System" composed of solar-powered satellites designed to "accommodate the explosive growth of data demands driven by AI." The scale is difficult to process: there are currently about 15,000 operational satellites orbiting Earth. SpaceX wants to multiply that figure by nearly 70.

After analyzing the FCC documents, Bloomberg and TechCrunch reports, and Musk's statements at Davos, I've reached a conclusion that might surprise you: this isn't just feasible, it makes brutal economic sense. Let me explain why.

The Numbers That Justify the Madness

SpaceX's proposal is based on a premise Musk has repeated multiple times: "The lowest-cost place to put AI will be space, and that will be true within two years, three at the latest." This statement, which sounds like typical Musk bravado, is actually backed by concrete numbers.

Cost Comparison: Earth vs. Space

What most guides won't tell you is that the biggest operational cost of a data center isn't the hardware: it's electricity and cooling. In space, both problems disappear. Satellites can capture sunlight almost constantly (95% of the time versus 24% on Earth), and the vacuum of space acts as an infinite heat sink.

According to Starcloud's analysis, orbital data centers could achieve 10x CO2 savings and reach carbon neutrality in 3-5 years of operation. The only significant environmental cost is launch emissions.

How These Satellites Would Work

SpaceX's proposal isn't vaporware. It's based on technology that already exists and is operational: Starlink V3 satellites.

Technical Specifications

The trick is that SpaceX wouldn't need to invent anything new. Musk confirmed on X: "Simply scaling up Starlink V3 satellites, which already have high-speed laser links, would work." V3s are already in development, with initial tests planned for early 2026.

The key is the laser mesh communication between satellites. This allows data to be processed in orbit without constantly needing to come down to Earth. Latency between satellites would be virtually zero, although communication with ground stations would add 20-50 milliseconds (comparable to a data center in another city).

The SpaceX-xAI Connection: The Master Plan

If you ask me directly why SpaceX is filing this application now, the answer lies in a Reuters report from January 29th: SpaceX and xAI are in negotiations to merge ahead of SpaceX's IPO.

Merger Timeline

Combined Valuations

The logic is impeccable: xAI needs massive compute capacity to train its models (Grok and future ones). SpaceX can provide that capacity through orbital data centers. Under a single corporate structure ("X Holdings"), the synergies are evident.

The banks involved in the IPO — Bank of America, Goldman Sachs, JPMorgan, and Morgan Stanley — are already working on the deal. If it goes as planned, it would be the largest IPO in history.

The Context: Why Earth Is No Longer Enough

To understand why Musk is betting on space, you need to look at what's happening on Earth. According to the International Energy Agency (IEA), data center energy demand will double from 460 TWh (2022) to over 1,000 TWh by 2026. That's equivalent to Japan's entire energy consumption.

The consequences are already being felt:

• Local officials are blocking permits for new data centers due to power grid stress
• Microsoft, Google, and Amazon are competing fiercely for nuclear energy contracts
• Industrial electricity prices are rising in areas with high data center concentration

The terrestrial power grid simply cannot scale at the pace AI demands. Musk put it at Davos: "More memory factories are needed because AI's needs are very high... and the power grid can't keep up."

Regulatory Obstacles: The FCC Won't Make This Easy

My verdict is clear on this: approval of 1 million satellites is not going to happen quickly, if it happens at all.

SpaceX's History with the FCC

The FCC has a habit of "trimming down" SpaceX requests. Recent example: approved 7,500 Gen2 satellites but held back on the remaining 14,988 due to orbital congestion concerns. A review process for 1 million satellites "could extend for years."

Required Reviews

Main Concerns

1. Space debris and Kessler Syndrome: One million satellites exponentially increases cascade collision risk
2. Light pollution: Astronomers already complain that Starlink ruins observations; 1 million satellites would be catastrophic
3. End-of-life management: How will thousands of satellites per year be deorbited?
4. Radio frequency interference: Starlink satellites already "leak electromagnetic radiation" detectable by radio telescopes

SpaceX's attempts to mitigate light pollution (DarkSat, VisorSat) have been "abandoned for engineering reasons." This doesn't help their case with regulators.

The Competition: SpaceX Isn't Alone

This isn't a single-player game. There's a global race for orbital data centers.

Main Competitors

Starcloud already achieved a historic milestone: they trained AI models (NanoGPT, Google's Gemma) in orbit using the first Nvidia H100 GPU in space. Their next satellite (October 2026) will have 100x more power generation capacity and will use the Nvidia Blackwell platform.

Jeff Bezos predicted in December 2025: "We will be able to beat the cost of terrestrial data centers in space within the next couple of decades." Musk says 2-3 years. The difference in timelines reveals the difference in aggressiveness between them.

China is following a more cautious but methodical strategy: satellite testing 2025-2027, cost parity 2028-2030, large-scale facilities 2031-2035. Their Beijing project targets 16 gigawatts of capacity.

The Skeptics Have Valid Points

Not everyone is convinced. Morgan Stanley published an analysis with the following concerns:

• Cosmic radiation: Can damage computing chips
• In-orbit maintenance: Practically impossible with current technology
• Space debris hazards: A single impact can destroy millions of dollars of infrastructure
• Hardware upgrades: Very expensive (each upgrade requires a launch)
• Data governance: What jurisdiction applies to data processed in orbit?

The Verge was more direct: "Approval of 1 million satellites is not likely to happen quickly. The plan might be more of a bargaining chip."

This last point is interesting. SpaceX could be asking for 1 million to negotiate down and get, say, 100,000. Which would still be 10 times more than all satellites currently in orbit.

The Impact on Astronomy: Acceptable Collateral Damage?

Astronomers are on the verge of a nervous breakdown. And with good reason.

Documented Problems

• The Vera C. Rubin Observatory will photograph the entire sky every 3 nights, capturing thousands of satellites in each image
• Detection of near-Earth asteroids is compromised: "Hazardous objects often first appear in our skies during twilight hours... when satellites are most likely to interfere"
• NASA research shows that even space-based observatories could be affected by light pollution from megaconstellations

One million additional satellites would turn this from a manageable problem into a catastrophe for science. The uncomfortable question is: does humanity prefer cheaper AI or astronomical science? Musk seems to have chosen.

What This Means for You

If you ask me directly how this affects you as a user or investor:

As an AI User

• Long-term (2028-2030): Potentially access to cheaper AI computing
• Short-term: No immediate impact, terrestrial data centers will continue to dominate
• Latency: Additional 20-50ms if your data is processed in orbit (noticeable for real-time applications)

As an Investor

• SpaceX IPO in June 2026: Possibly the largest in history
• Terrestrial data center stocks: Could face pressure if the orbital concept takes off
• Defense/space companies: Beneficiaries of the race for orbital infrastructure

As a Citizen

• More visible satellites in the night sky (and not romantic shooting stars, but chains of light dots)
• Regulatory debates about who controls orbital space
• Existential questions about whether AI should take priority over astronomy

Conclusion: Kardashev II Civilization or Marketing Smoke?

SpaceX's FCC document includes a phrase that reveals the true ambition: "Launching a constellation of one million satellites that operate as orbital data centers is a first step towards becoming a Kardashev II-level civilization — one that can harness the Sun's full power."

It's grandiose. It's typical Musk. And, honestly, it's technically correct. If you can deploy large-scale solar infrastructure in space, you're on the path to a civilization that captures stellar energy.

My verdict is clear: this project won't be completed as described. The FCC will trim the application, astronomers will protest, timelines will extend, and costs will be higher than projected. But something of this will get built. Starcloud already has hardware in orbit. Google and China are on their way. The question isn't whether there will be space data centers, but how many and whose.

SpaceX, with Starship operational, the xAI merger in progress, and a $1.5 trillion IPO on the horizon, has more resources than anyone to try. And if Musk is right about the economics — that space will be the cheapest place for AI in 2-3 years — then what seems crazy today will be obvious tomorrow.

See you in orbit.

Frequently Asked Questions

How many satellites does SpaceX want to launch?

SpaceX filed with the FCC for permission to launch up to 1 million satellites as part of its "Orbital Data Center System." There are currently approximately 15,000 operational satellites in orbit, so this proposal would multiply the global satellite population by nearly 70.

When would these space data centers be operational?

According to SpaceX's plans and analyst projections: initial Starlink V3 deployment in 2026-2027, pilot programs in 2027-2028, and full commercial service between 2028-2030. However, Musk has a track record of promising ambitious timelines that are later delayed.

Why would space be cheaper than Earth for AI?

Two main factors: near-constant solar energy (95% capacity factor vs 24% on Earth) and free cooling via radiation to the vacuum of space. Terrestrial data centers spend over $20 million annually on electricity and cooling water.

What is the SpaceX-xAI merger?

Reuters reported on January 29, 2026 that SpaceX and xAI (Musk's AI company) are negotiating a merger ahead of SpaceX's IPO. The goal would be to create a combined entity valued at approximately $1.5 trillion, with xAI benefiting from SpaceX's orbital computing capacity.

What problems would this cause for astronomy?

One million additional satellites would severely worsen the light pollution already affecting observatories. The Vera C. Rubin Observatory, near-Earth asteroid detection, and potentially even space-based observatories would be compromised by the number of visible satellites.