Moving artificial intelligence computation into orbit
From the same company that industrialized rocket reuse, SpaceX has now turned its gaze toward the sky not merely as a highway but as a home for computation itself. Elon Musk has announced plans to build Gigasat, an 11-million-square-foot satellite manufacturing facility, with the goal of placing AI data centers into orbit and delivering one gigawatt of computing capacity per year by late 2027. The ambition is large, but the company frames it deliberately as an engineering and manufacturing challenge rather than a scientific leap — a distinction that speaks to how SpaceX understands its own deepest advantage. In doing so, it positions itself to reshape not just how humanity connects, but where intelligence itself is housed.
- SpaceX has announced plans to move AI computation into orbit, targeting operational data center satellites by late 2027 — a timeline that compresses what most would consider a decade of work into roughly two years.
- The Gigasat factory, at 11 million square feet, represents a capital commitment that signals this is not a prototype experiment but a bet on sustained market demand for space-based computing infrastructure.
- The project's credibility rests on a deliberate claim: that no fundamental breakthroughs are needed, only the manufacturing scale and deployment logistics SpaceX has spent two decades mastering through Starlink.
- Orbital data centers would disrupt the geography of cloud computing, offering globally distributed, latency-differentiated infrastructure accessible anywhere on Earth — a structural advantage for AI inference, financial systems, and distributed machine learning.
- The announcement arrives ahead of a potential SpaceX IPO, lending the project a dual purpose: a genuine technological roadmap and a compelling revenue narrative for public markets evaluating the company's future.
Elon Musk has announced that SpaceX will build a facility called Gigasat — 11 million square feet dedicated to manufacturing satellites designed not for communication alone, but as orbital data centers. The target is one gigawatt per year of AI computing capacity delivered from space by late 2027, with initial orbital tests beginning as early as the end of 2026.
What distinguishes this plan from speculative moonshots is the company's insistence that the technology is already in hand. The satellites will draw on systems proven through Starlink — existing power architectures, thermal management, and launch infrastructure. SpaceX is not asking the laws of physics for new favors. It is asking its own factories to perform at a scale and consistency that traditional aerospace has rarely achieved.
The strategic logic runs deeper than novelty. Cloud computing has always been anchored to the ground — to specific cities, specific buildings, specific cables running beneath oceans. Orbital data centers are distributed by their very nature. They are accessible from anywhere, and they offer latency and redundancy profiles that terrestrial systems cannot replicate. For AI inference, real-time analytics, or applications demanding global reach without geographic bottlenecks, that difference could justify a significant cost premium.
SpaceX enters this space not as a newcomer but as an operator already running thousands of satellites and the logistics network to support them. Gigasat would layer a new product onto an existing platform — a compounding of capability rather than a fresh start.
The announcement also arrives with notable timing, preceding a potential public offering. Markets reward legible paths to revenue, and a space-based AI compute business offers exactly that: a defined product, identifiable customers in the cloud and AI sectors, and a credible manufacturing roadmap. Whether orbital compute can be delivered at prices customers will actually pay — and whether Gigasat can meet its production targets — remains the open question at the center of an otherwise carefully constructed story.
Elon Musk has laid out a plan to move artificial intelligence computation into orbit. SpaceX will build an 11-million-square-foot factory called Gigasat, dedicated to manufacturing satellites designed as data centers. The company aims to have these orbital machines operational and delivering computing power by late 2027, with a target capacity of one gigawatt per year—a measure of sustained computational throughput that would make the constellation a meaningful player in the global AI infrastructure market.
The timeline is aggressive but grounded in existing engineering. According to sources familiar with the project, SpaceX intends to begin testing orbital AI computing by the end of 2026. Musk has emphasized that the satellites will rely mostly on technology already proven in space, rather than requiring fundamental breakthroughs. This matters because it suggests the company believes the bottleneck is not innovation but manufacturing scale and deployment logistics—problems SpaceX has spent two decades learning to solve.
The Gigasat factory represents a significant capital commitment. Building an 11-million-square-foot facility is not a modest undertaking. It signals that SpaceX sees sustained demand for this product, not a one-off experiment. The factory will be purpose-built for satellite production at scale, which is a different engineering challenge than building individual spacecraft. The company will need to solve problems of standardization, quality control, and supply chain management that traditional aerospace has struggled with for decades.
What makes this move strategically interesting is the convergence it represents. SpaceX already operates Starlink, a constellation of thousands of satellites providing broadband internet. These new data center satellites would share launch infrastructure, ground stations, and operational expertise with that existing network. The company is not starting from zero. It is leveraging a platform it has already built and continues to expand.
The implications ripple outward. Cloud computing has historically been bound to terrestrial data centers—massive facilities in specific geographic locations, connected by undersea cables and terrestrial fiber. Orbital data centers would be distributed by definition. They would offer latency characteristics different from ground-based systems. They would be accessible from anywhere on Earth with a satellite connection. For certain applications—financial trading, real-time AI inference, distributed machine learning—these properties could be valuable enough to justify the cost premium of space-based infrastructure.
Musk's framing emphasizes that the technology is not exotic. Existing satellite platforms, existing power systems, existing thermal management approaches. The novelty is in the application and the scale of manufacturing. This is a deliberate choice in how to present the project: not as a moonshot requiring breakthroughs, but as an engineering execution problem. Whether that framing holds up will depend on whether the Gigasat factory can actually produce satellites at the rate and cost the company is projecting.
The announcement comes ahead of a potential SpaceX IPO, a detail that colors the timing and the messaging. Public markets reward clarity about paths to revenue and profitability. A space-based AI data center business offers both: a clear product, identifiable customers in the AI and cloud computing sectors, and a timeline to meaningful scale. Whether the business model proves durable—whether customers will actually pay for orbital compute at prices that make the venture profitable—remains an open question. But SpaceX has now committed publicly to finding out.
Notable Quotes
The satellites will rely mostly on technology already proven in space, rather than requiring fundamental breakthroughs.— Elon Musk, SpaceX
The Hearth Conversation Another angle on the story
Why would anyone want a data center in space when we have perfectly good ones on the ground?
Latency and distribution. A satellite orbiting overhead can serve requests from anywhere on Earth simultaneously. For certain applications—real-time AI inference, financial systems, distributed machine learning—that matters more than the extra cost.
But doesn't space introduce new problems? Heat, radiation, reliability?
Yes. But SpaceX says they're using proven technology, not inventing new solutions. The hard part isn't the physics—it's manufacturing at scale and keeping costs down.
One gigawatt by 2027 sounds like a lot. Is that realistic?
It's ambitious. But SpaceX has already built a factory-scale operation for Starlink. They understand satellite production in ways most companies don't. Whether they can hit that specific number on that specific timeline is another question.
Who would actually buy this? What's the customer?
AI companies needing distributed inference. Cloud providers wanting a new service tier. Anyone who benefits from low-latency computation near the edge. The market exists—whether orbital compute is the right answer is what they're testing.
Does this change how we think about the internet?
Potentially. It blurs the line between satellite internet and cloud infrastructure. If it works, you're not just getting connectivity from space—you're getting computation. That's a different thing entirely.