In February, SpaceX acquired xAI in an all-stock deal valuing the combined business at $1.25 trillion, the largest merger on record. Days earlier it had applied to operate up to 1 million orbital data center satellites. In June it completed the largest IPO in history.
Read together, these moves describe a rocket and satellite company absorbing AI models and the compute to run them into a single stack. The orbital filing would extend that compute layer into space, and the listing funds the expansion.
This complicates the promise of cheap launch. SpaceX estimates Falcon 9 cut the historical per-kilogram cost of reaching orbit by roughly 85%, and that collapse was widely read as opening access to space. Investment research cast launch as the toll road to orbit and the services above it as an application layer for others to build on.
SpaceX is the company best placed to be that toll road.
The stack rises from launch at the base to satellites and the orbital network, the connectivity sold across it, compute and at the top AI models and applications. To climb is to use a defensible base to reach for the layers above.
The hardest-to-replicate bottlenecks sit low in that stack. Much of the room for differentiation and value capture sits higher. SpaceX’s own revenue split makes the point. In 2025, launch services generated about $2.6 billion against $11.4 billion from connectivity.
Yet SpaceX’s strongest structural positions sit lower in the stack. It says that, since 2023, it has launched more than 80% of the world’s mass to orbit each year, and that Starlink now operates roughly three quarters of all active maneuverable satellites.
Cheaper launch did open the layers above to new entrants, and SpaceX could have stayed their supplier. It has gone further, moving up through every layer itself, from vehicle to satellite to network to terminal to connectivity sold directly to consumers, enterprises and governments. The climb does not stop at connectivity.
The next rung is compute. On the ground, the IPO filing discloses an Anthropic compute contract worth $1.25 billion a month through 2029, and a comparable Google arrangement followed within weeks. These arrangements can be terminated on relatively short notice, and SpaceX says the structure lets it reallocate capacity to its own initiatives. Nor is the ambition confined to Earth: The orbital data center filing would carry that same compute layer into space.
And it is reaching past compute into the models and applications above. Having already absorbed xAI’s models, it has agreed to acquire Anysphere, the parent of Cursor, for an implied $60 billion, buying into AI applications rather than only the capacity beneath them.
Integration is not, by itself, the concern. Apple designs its own silicon and still lets millions of developers ship on its platforms; AWS builds its own chips and models yet hosts rivals’ models beside them.
Neither is fully open, and both set terms others resent. But a competitor can still build a real business on top, move to a rival’s platform, or build rival infrastructure elsewhere. The question is which layers an owner keeps contestable — and on what terms.
Enclosure here runs through constrained access as much as control, and it varies by layer. At the base, launch access is tightening: outside customers face multi-year waits. Higher up, connectivity stays available, but on SpaceX’s terms. The question is why that grip should hold, when dominant infrastructure elsewhere can eventually be contested.
The obvious answer is capital, but capital alone does not make a position durable. Cloud computing is the cautionary case: It is concentrated, capital-intensive and vertically integrated, and AI is raising the value of its top layer sharply. Yet its capacity remains replicable.
A well-financed rival can build in parallel, as Microsoft Azure, Google Cloud and Oracle did years after AWS. What makes space different is that its foundations cannot be replicated by capital alone.
The reason is that space runs on inputs whose usable capacity is constrained and mediated by regulators. Spectrum is finite, governed through national licensing and international coordination through the ITU. Orbital capacity is constrained by collision risk, debris rules, interference and the coordination burden that dense constellations impose.
To stop operators warehousing scarce spectrum resources they do not use, regulators impose deployment milestones, and the milestones reward speed. The ITU, for example, applies phased requirements to covered filings: an operator must deploy 10% of a constellation within two years after the bringing-into-use period, half within five and all within seven, or have the number of satellites covered by the filing reduced accordingly.
National regulators add their own deadlines. Meeting them depends on launch capacity that most operators cannot supply themselves. In the United States, the FCC required Amazon to have 1,616 satellites deployed and operational by July 30, 2026. Amazon, spreading its constellation across five launch vehicles, still fell far behind, and had to add launches on Falcon 9, the rocket of the very rival it is racing. The FCC granted a waiver but temporarily downgraded the processing-round priority of satellites launched after the deadline.
Increasingly, that launch capacity is being pulled inward. Reuters reported that Starlink now accounts for about 79% of Falcon 9 missions, up from 54% in 2020, while at least seven outside companies have been told the rocket is booked until 2028 or 2029.
That is the tell. Launch is open, yet even a rival that deliberately diversified across launch providers still had to turn to the company it is racing.
So a milestone regime meant to prevent hoarding ends up rewarding whoever controls launch.
China is not the counterexample it appears to be. Guowang and Qianfan plan roughly 28,000 satellites combined. But a second bloc-scale alternative is not a return to openness. A Western operator that cannot rely on SpaceX is unlikely to find a Chinese constellation a practical alternative. Enclosure by blocs is still enclosure.
Nor do the Western alternatives yet match Starlink on coverage, scale and integrated launch. OneWeb operates around 650 satellites, Amazon Leo is still deploying, and Europe’s IRIS² will not begin launching until 2029. OneWeb, too, has leaned on Falcon 9 to fill out its constellation. Alternatives exist; an equivalent substitute does not.
If the market cannot supply a substitute, the natural recourse is regulation. But antitrust solves only part of the problem. Antitrust authorities can seek structural separation, force asset sales or require network access. That changes who controls the infrastructure. It does not create more spectrum or remove the physical and coordination constraints on usable orbital capacity.
Intervention is further complicated because the government is not only a potential enforcer but a dependent customer.
Reuters reported that, during the Iran campaign, SpaceX sought to raise the monthly charge for terminals used by U.S. strike drones from roughly $5,000 to $25,000, and that the Pentagon resisted before agreeing for want of an alternative. Both the Pentagon and Elon Musk disputed the account. Even contested, the episode illustrates the risk.
Orbital compute may not succeed. Bain estimates orbital data centers turn cost-competitive only if launch prices fall to roughly $50 to $100 a kilogram, and Starship has yet to demonstrate full reuse. But the enclosure argument does not depend on it: If compute stays on the ground, what shrinks is the size of the prize, not the difficulty of contesting it.
Either way, the exposure is real. For companies in fields such as telecommunications, defense, satellite manufacturing, Earth observation, in-space services and AI infrastructure, simply building on someone else’s platform may be too passive a strategy. Dependence on another firm’s infrastructure is normal. The risk is dependence on an owner that can enter your layer and cannot easily be replaced.
The telecom layer offers an early preview. T-Mobile partnered with Starlink to fill gaps in its terrestrial coverage. SpaceX now says it intends to build a full mobile service and win customers from T-Mobile, AT&T and Verizon. A complement is signalling its intention to become a competitor.
SpaceX is the clearest case, but the exposure runs through the space economy, from launch to in-space services. Firms facing it have four broad responses. Build reduces dependence through ownership; the other three manage it in different ways.
Build means owning enough of the stack to reduce dependence on a potentially competing provider, feasible only for a well-capitalized few able to secure scarce inputs early.
Bloc means aligning with a state- or consortium-backed alternative that can provide a credible second infrastructure base when no individual firm can build one alone, using pooled investment and committed demand to make that alternative viable.
Shelter means concentrating on activities the infrastructure owner has limited incentive or ability to absorb, where regulation, certification, sovereign requirements or deeply specialised expertise make entry into that layer more difficult.
Hedge means continuing to use incumbent infrastructure while preserving strategic optionality, by retaining customer relationships and control of critical data, supporting interoperable interfaces, qualifying multiple providers and maintaining the ability to switch or multi-home.
The point is not to forecast who wins, but to read which game is being played. The strategic question is whether the layer you depend on is one its owner can close around you. In space, that structure is taking shape faster than many firms are adapting to it.
Konstantinos Trantopoulos is a Senior Advisor and Fellow at IMD Business School. He advises executives, boards and investors on strategy, investment and value creation, with a particular focus on how emerging technologies and AI reshape competitive advantage. His work has been published in Harvard Business Review and MIT Sloan Management Review among other leading outlets.
SpaceNews is committed to publishing our community’s diverse perspectives. Whether you’re an academic, executive, engineer or even just a concerned citizen of the cosmos, send your arguments and viewpoints to opinion (at) spacenews.com to be considered for publication online or in our next magazine. If you have something to submit, read some of our recent opinion articles and our submission guidelines to get a sense of what we’re looking for. The perspectives shared in these opinion articles are solely those of the authors and do not necessarily represent their employers or professional affiliations.