The Companies Placing Bets On Putting Data Centers In Space

Google has launched test AI chips to assess whether they can withstand the vibration, radiation and temperature extremes of space, adding a practical test to a growing set of bets on orbital data centers. The effort, conducted with satellite company Planet, is part of a field that also includes SpaceX, Starcloud, K2 Space, Blue Origin, Axiom Space and Cowboy Space, though its commercial prospects remain uncertain.

The Google hardware consists of Tensor Processing Units, or TPUs, its chips designed for AI computing. The test is intended to reveal how the processors perform under conditions that can damage electronics. The report does not specify the test’s duration, the data Google expects to collect or when it will publish results. Other companies have conducted related tests, and Nvidia has developed a chip design intended for the stresses of orbit.

SpaceX has the most expansive stated plans. Its chief executive, Elon Musk, has said the company aims to start launching orbital data centers in 2027 and has discussed a network of up to one million. The proposed satellites would be about 100 feet high, with a wingspan of roughly 250 feet, solar panels and a large radiator for removing heat. SpaceX currently plans to use Nvidia chips, with a longer-term aim of producing its own chips at a Texas facility called Terafab.

These plans are not the same as demonstrated operating capacity. SpaceX’s prospectus describes its ambitions and risks; it does not establish that the proposed launch schedule, scale or economics have been achieved. The company says its long-term concept could involve launching thousands of data centers annually and consuming as much as 100 gigawatts of power. Those figures are targets, not current operations.

The Economics Depend on Launch Costs

Orbital computing could matter if it supplies power and computing capacity that are difficult to provide on Earth. Supporters point to abundant sunlight in space and the possibility of avoiding some land-use, permitting and local power constraints faced by large terrestrial data centers. The idea is particularly relevant as demand for AI computing grows and communities debate the energy and infrastructure needs of data-center construction.

But those potential advantages depend on expensive systems working together. Data must move between satellites and back to Earth with low enough delay for useful applications. The chips generate heat, while the vacuum of space provides no air to carry it away; large radiators must release that heat. Failures could require robotic repairs, and equipment may need replacement after around five years or sooner, according to the report.

The business case also turns on the price of launching heavy equipment. Lower launch costs are a prerequisite in many of the proposals, not a settled fact. If launches remain costly, orbital facilities may be limited to specialized uses rather than competing with data centers on land.

Tests Precede Commercial Deployment

Companies pursuing orbital data centers include satellite and launch businesses as well as firms developing chips and spacecraft. Their approaches vary, but the shared premise is that future improvements in launch vehicles, computing hardware and satellite systems could make it feasible to place more computing capacity in orbit. The report describes the sector as an early-stage effort, with technical demonstrations and long-range plans rather than a mature alternative to ground-based facilities.

Google’s plan relies in part on launch costs falling. The company’s estimates cited in the report envisage that possibility by 2035 if SpaceX’s Starship flies nearly 200 times a year with full payloads and lower launch prices are passed on to customers. Those conditions are forecasts, not confirmed milestones. Starcloud, in a white paper, argues that scaled orbital systems could eventually have lower operating costs because of energy and cooling advantages. That comparison also depends on launch prices and successful deployment at scale.

SpaceX has argued that it is positioned to pursue the work because it controls rocket and satellite manufacturing. In its IPO prospectus, the company said: “We believe orbital AI compute is an incredibly difficult technical challenge that only we can solve at scale in the near term.” This is the company’s assessment of its competitive position, not an independently established conclusion.

““We believe orbital AI compute is an incredibly difficult technical challenge that only we can solve at scale in the near term.””

— SpaceX, in its IPO prospectus

Key Technical and Cost Questions Remain

Google’s test results are not yet detailed in the source report. It is unclear how the TPUs performed, whether they remained operational throughout the test or what follow-up work is planned. More broadly, there is no confirmed evidence here that any company has demonstrated an orbital data center operating at commercial scale.

Important engineering questions remain, including how reliably satellites can shed heat, maintain high-bandwidth laser links and return data quickly enough for different workloads. The frequency and cost of repairs, replacement parts and chip upgrades are also uncertain. The report says chips may need replacement about every five years or sooner, but does not provide a validated service-life estimate for an operational system.

The financial case is unproven. Projected savings rely on much cheaper launches, high payload utilization and continuing demand for AI computing. The report also notes a counter-risk raised by SpaceX: if AI growth slows, terrestrial data centers might be sufficient. No timetable or cost estimate in the source establishes when orbital facilities could compete with land-based ones.

Watch for Test Results and Launch Progress

The immediate milestone is evidence from Google and Planet’s chip test: how the TPUs handled launch and orbital conditions, and whether the results lead to another test or a larger demonstration. Updates from Starcloud, K2 Space, Blue Origin, Axiom Space, Cowboy Space and Nvidia may also show whether different hardware and satellite designs can address the same problems.

For SpaceX, the stated target is to begin launches in 2027. Progress toward that date, the development and flight cadence of Starship, and any disclosed launch pricing will help clarify whether the company can approach the scale described in its plans. Those milestones would not, by themselves, establish that a profitable orbital data-center business exists. The next phase will need to show both reliable operation in space and a cost case that holds up against terrestrial facilities.

Key Questions

What did Google send into space?

Google launched test Tensor Processing Units, or TPUs, with satellite company Planet to assess how the AI chips handle conditions such as vibration, radiation and extreme temperatures.

Which companies are pursuing orbital data centers?

The report names SpaceX, Starcloud, K2 Space, Blue Origin, Axiom Space and Cowboy Space, alongside Google’s chip-testing effort and Nvidia’s work on chip designs for orbit.

When does SpaceX say it will start launching data centers?

Elon Musk has said SpaceX plans to begin launches in 2027. This is a stated target; the report does not say that commercial orbital data centers are already operating.

What are the main obstacles to data centers in space?

The main hurdles include launch costs, heat removal, radiation, reliable communications and repairs. Data centers in orbit would also need equipment replacement, while delays in sending data to Earth could limit some uses.

Are orbital data centers expected to replace facilities on Earth soon?

The report says a near-term replacement for terrestrial data centers is not expected. Orbital systems may first prove useful in limited applications, and their broader commercial prospects depend on technical progress, lower launch costs and sustained demand.

Source: rss