SpaceX's plan to launch 1 million AI data center satellites starting in 2028 could cost $2 trillion, nearly matching the company's entire projected IPO valuation, and leading space scholars warn the economics do not work.
Back
SpaceX's $2 trillion AI satellite plan faces reality check from scholars
Elon Musk's plan to begin launching one million AI data center satellites into orbit in 2028 could cost roughly $2 trillion — nearly the entire projected valuation of SpaceX ahead of its initial public offering — and leading space scholars warn the economics do not work.
"Launching a million satellite orbital data center constellation is fantasy," Robert Zubrin, a rocket designer who worked on an early prototype of NASA's Space Launch System and founder of the Mars Society, said in an interview. Zubrin, who has known Musk for 25 years and served as a de facto mentor, said the scheme could drain SpaceX's coffers and trigger a financial crash.
SpaceX has already lofted 10,000 Starlink satellites at roughly $2 million each to build and launch, Zubrin said. Scaling that to one million units would cost about $2 trillion, matching the upper end of SpaceX's IPO valuation target. The company's S-1 filing, submitted ahead of a public listing as early as June, values SpaceX at $1.75 trillion and leans heavily on AI revenue projections — 93 percent of its $28 trillion total addressable market comes from AI models, apps, and services, according to the prospectus.
The core problem is cost per kilowatt. Each Starlink satellite generates 20 kilowatts of power from its solar panels at a cost of $100,000 per kilowatt, Zubrin said. Terrestrial solar arrays produce electricity at roughly $3,000 per kilowatt, while gas-fired generation costs about $1,000 per kilowatt. Even commercial nuclear power, at $5,000 to $10,000 per kilowatt, is an order of magnitude cheaper than space-based solar. "Space solar power is vastly more expensive than solar power on Earth," Zubrin said.
Google's alternative timeline
A study published by nine Google scholars under Project Suncatcher, titled "Towards a future space-based, highly scalable AI infrastructure system design," concluded that orbital AI data centers may not become economically competitive with terrestrial facilities until the mid-2030s. The researchers said launch costs would need to fall to $200 per kilogram — achievable only if SpaceX reaches roughly 180 Starship launches per year, a cadence they project by about 2035.
SpaceX currently averages three launches per week across its Falcon 9 fleet. Musk's plan calls for one Starship launch every hour, or 8,700 flights annually, to deploy the million-satellite constellation by 2028. The Starship has completed 12 flight tests; its most recent test on May 2026 saw the upper stage splash down successfully in the Indian Ocean, but the first-stage booster failed to reignite its 33 engines and crashed into the Gulf of Mexico.
Brian Hurley, founder of the New Space Economy think tank, said top researchers have reached a consensus that orbital AI data centers will not become economically feasible until the mid-2030s. "Economically meaningful deployment at scale is more likely a decade-plus proposition," Hurley said. "Terrestrial data centers benefit from mature supply chains, ground access, utility-scale power contracts, and decades of operational learning. Orbital infrastructure starts out with much steeper costs."
IPO stakes and the AI narrative
SpaceX's S-1 filing shows the company generated $18.7 billion in revenue in 2025, with Starlink contributing $11.4 billion, or 61 percent of the total. The Space division posted an operating loss of $657 million on $4.08 billion in revenue, alongside $3 billion in research and development spending. The company's AI division, formed after SpaceX acquired xAI in February, controls just 3.4 percent of the AI market, according to the S-1. Its largest revenue stream comes from renting graphics processing units in its Colossus data centers to rival Anthropic, generating $1.25 billion per month, or $15 billion annually.
Musk has denied reports that SpaceX lowered its IPO valuation target from $2 trillion to $1.8 trillion, calling the claims "false" on his social media platform X. The company's confidential S-1 filing positions it for a public listing that could be the largest in market history. Wedbush analyst Dan Ives has placed the probability of a subsequent Tesla-SpaceX merger at 80 percent to 90 percent, with a target completion in the first half of 2027.
Zubrin said Musk may be using the AI satellite narrative to amplify the IPO. "He's calculating that people are looking at this and saying, well, I don't know if this is really going to work, but no one's ever lost money betting on Elon Musk," Zubrin said. The internet boom of the late 1990s saw a similar dynamic, he added, with investors pouring money into search engines before the business models were proven.
SpaceX's AI satellite plan faces a fundamental math problem: the cost of generating power in orbit is 30 to 100 times higher than on the ground, and no foreseeable launch cost reduction closes that gap within Musk's timeline. Google's Suncatcher team plans to launch twin prototype satellites with Planet Labs early next year to test inter-satellite optical links and on-orbit machine learning — a small-scale demo that, if successful, would still be a decade away from commercial viability. For SpaceX, the question is whether the AI narrative can sustain a $1.75 trillion valuation long enough for the technology to catch up with the ambition.
This article is for informational purposes only and does not constitute investment advice.
