On Monday of this past week, Elon Musk-led SpaceX revealed that it had acquired Musk’s AI company xAI, in an all-stock transaction at a reported $250B valuation. With SpaceX being valued at $1T – the world’s most valuable private company – the $1.25T combination is being called the largest merger ever. (xAI will be a wholly owned subsidiary.) The stated rationale for the merger is heavily focused on orbital data centers and their potential to drive down the cost of AI compute. SpaceX’s recent application to the Federal Communications Commission (FCC) for permission to launch up to 1M satellites has been fast-tracked and is already open for public comment. With a SpaceX IPO slated as soon as Jun 2026, it’s unclear whether the xAI acquisition will juice the IPO or serve as an overhang.

As we have said before, the rationale for orbital data centers is fundamentally about cost. A data center in space can be cheaper than a terrestrial one for 4 primary reasons – land, energy, cooling, and regulation. Tech leaders/firms are taking it seriously, including efforts from SpaceX/Tesla/xAI CEO Elon Musk, Google CEO Sundar Pichai, Nvidia, Amazon founder Jeff Bezos, Relativity Space CEO Eric Schmidt, and OpenAI CEO Sam Altman.

As a result, Musk has made orbital data centers “a matter of utmost urgency” since mid-to-late 2025. It was even one of the main reasons for SpaceX to press for an IPO this year (alongside ramping up Starship rocket launches, building a lunar moonbase, and working on missions to Mars). SpaceX reportedly had a key breakthrough in the fall of 2025, in figuring out how to scale data centers in space. Musk seems to be using his coming June birthday and a rare planetary alignment to press for an accelerated IPO timeline, which would get it ahead of other hotly anticipated IPOs (e.g. Anthropic, OpenAI).

Even though energy is just 10-15% of the total cost of ownership of a data center, the bottlenecks in the AI data-center buildout can largely be traced to electricity. This is particularly true right now, when localities are starting to ask AI players to bring their own power infrastructure.

In space, solar isn’t an intermittent source. Solar panels can harvest the sun’s rays directly, efficiently, and continually, without being interrupted by clouds, precipitation, seasons, or nightfall. Solar panels in space are 5x to 8x more productive than on Earth. According to Musk, just avoiding atmospheric dampening allows for 30% more power. Cooling is also extremely efficient due to the coldness of space (3 Kelvin or -454°F in shadow). Land use conflicts would be reduced as well, due to the amount of space in space.

There are myriad technical and operating-model problems that still need to be solved – such as how to build hardware/chips that can withstand the radiation in space, how to disperse the AI chips’ heat in a vacuum, how to achieve data center-scale inter-satellite links, how to control large clustered satellite formations, and how to scale the production of satellites (which Musk calls “just solar panels and a battery”). Musk clearly has an eye towards achieving the vertical integration needed to bring equipment, launch, and other costs down so orbital data centers can reach cost parity.

An orbital data center today would probably be 3x-4x the cost of a terrestrial data center, given what it takes to build them and get them into space. However, with time and a new industrial stack and scale, this could eventually flip to space-based data centers costing up to 90-95% less than terrestrial options. Musk believes that there would be no batteries required and little to no ongoing operational or maintenance needs (non-operating units would be burned up upon re-entry).

SpaceX is particularly advantaged in having the lowest-cost rocket platform in the Falcon 9, which has launched 11K+ satellites into space at an internal cost of $15M each (vs. the 4-5x market price). About 9,600+ satellites are currently operational and serving Starlink’s 9M+ subscribers. (By contrast, Amazon is struggling to put enough satellites into orbit to meet its FCC deadline – it has launched about 150 of the 1,600 satellites needed, and is using SpaceX as well as Bezos’ Blue Origin for its rocket launches.) SpaceX’s new data-center satellites would interconnect via optical links with its existing Starlink broadband satellites, which would transmit the data down to ground stations on Earth.

The data-center strategy is heavily reliant on the availability of SpaceX’s larger reusable Starship, which is under development and gearing up for its 12th test flight. The Starship rocket, which is designed to deliver more than 4x the payload of the Falcon 9, could eventually bring the cost per pound of payload down to under $100. Musk expects the Starship to begin delivering satellites to orbit this year, including the more powerful V3 Starlink satellites and the next-gen direct-to-mobile satellites. His vision is for Starship to get to a point where it is launching every hour with 200 tons per flight, for a potential of millions of tons delivered to orbit every year.

Musk sees a path to putting 100 GW per year of solar-powered AI satellites into orbit – roughly equivalent to a quarter of US annual energy consumption. This would involve 1M tons of satellites launched per year, generating 100 kW of compute power per ton. With Starship, he believes he could deliver 300-500 GW of solar satellites annually to orbit. Musk has speculated on X that even 100 TW (100,000 GW) per year would be possible if there was “a lunar base producing solar-powered AI satellites locally and accelerating them to escape velocity with a mass driver” (like a catapult).

SpaceX’s advantage will eventually erode as rivals such as Rocket Lab, Bezos’ Blue Origin, and copycat Chinese space companies start to catch up – which is likely one of the reasons for Musk’s urgency. While it’s unlikely that SpaceX will be launching 1M satellites anytime soon, as Musk has noted, “the need to launch these satellites will act as a…forcing function to drive Starship improvements and launch rates” – similar to what building out the early Starlink constellation did for the Falcon.

SpaceX needs capital to achieve its ambitions. It is only generating $15B-$16B in revenue ($8B in profit) annually, of which 50-80% comes from Starlink. An IPO could raise as much as $50B (at a targeted $1.5T valuation), which would be a healthy cushion to help it scale its orbital data centers.

For those who want to invest in Elon Musk through a single equity, SpaceX-xAI has obvious allure. xAI, notably, also owns Musk’s social platform X (formerly Twitter), which it acquired in early 2025 in an all-stock $113B transaction that valued X at $33B. Musk-controlled Tesla (which was included in one early merger scenario) has a $2B stake in xAI, based on a recent $20B Series A round that established the $250B valuation for xAI. Given all of the intertwinings among Musk-affiliated entities, it has industry watchers calling the network Musk’s “personal conglomerate.”

xAI probably needs SpaceX more. It is only generating in the low hundreds of millions of dollars in annual revenue, while burning nearly $1B per month. In addition to chips and data-center infrastructure, xAI has some potential money pits in the form of looming lawsuits and liabilities (e.g. alleged deletion of evidence, data use, nonconsensual sexual content) – although the subsidiary structure partially shields SpaceX from xAI’s liabilities. (It also keeps xAI’s debt covenants from being triggered, since it doesn’t count as a change of control.) While the acquisition was not favored by all SpaceX investors – especially those looking forward to cashing out via an uncomplicated IPO – Musk’s control of both entities leaves little room for dispute. (Musk owns 42% of SpaceX with 80% voting control, and owns 51% of xAI.)

It will take time before orbital data centers are real. Musk – who’s been known to put forward ambitious timelines – commented in Nov 2025 that it could be in 4-5 years and recently revised his estimate down to 2-3 years, at which point he believes space-based data centers will be the lowest-cost way to generate AI compute. (Others such as Jeff Bezos, Google, and academic researcher Phil Metzger have come up with more conservative projections.) The pathway there could be streamlined by the FCC’s transition to a “Default to Yes” framework for regulatory approvals. Musk is framing orbital data centers as a necessary step towards humanity becoming a Kardashev Type II civilization (stellar) that can consume a star's energy directly – a vision referenced directly in SpaceX’s FCC application.

In the meantime, SpaceX’s existing Starlink constellation continues to grow, following the FCC’s approval in Jan 2026 of a total of 15K Gen2 satellites (about 6,200 are already operational). SpaceX has been working on direct-to-cell for some time, including working with chipmakers to build satellite connectivity into handsets and acquiring $17B in wireless spectrum licenses from EchoStar. Apple is reportedly in talks with SpaceX to have Starlink provide satellite-based connectivity for the iPhone 18.

Jan 23 2026 (3 Shifts): Data centers in space

Jan 9 2026 (3 Shifts): 2026 IPO filings have begun