SpaceX Holds the Cost Advantage in Orbital Data Centers
Former Meta CTO Mike Schroepfer told Bloomberg Technology that orbital data centers are plausible but likely economic only for SpaceX, whose vertical integration and launch costs give it a hardware advantage others cannot match. Schroepfer, now a founding partner at Gigascale Capital, argued that ocean-based data centers currently have stronger cost logic because mass can be deployed there about 100 times more cheaply than in orbit, while land-based solar and batteries remain a faster near-term route to new compute capacity.
Space is viable, but Schroepfer says the economics belong to SpaceX
Mike Schroepfer does not treat orbital data centers and ocean data centers as mutually exclusive futures. His view is “a little bit of all of the above”: space, ocean, and land-based solar-and-battery infrastructure may all have roles if demand for compute keeps outrunning available capacity. But he draws a hard line around who can make the space version work.
Ed Ludlow framed the question around SpaceX’s stated orbital data-center ambitions, referring in his prompt to the prospect of a SpaceX IPO and to a “prospectus” that he said was clear about the plan. He then asked whether a private-market investor should bother with another startup pitching the same idea, or whether SpaceX has effectively covered the category.
For orbital data centers, Schroepfer’s core claim is that the business depends less on the abstract feasibility of putting compute in orbit than on the cost structure of the organization attempting it. Space is “very hard and expensive to get there,” he said, and “nobody but SpaceX can do it” because of its vertical integration and wholesale launch costs. Without that cost base, “the economics are just tough.”
Bloomberg showed SpaceX-attributed satellite animation with an on-screen chyron reading, “SpaceX has hardware moat for orbital data centers.” Schroepfer’s explanation for that moat was economic: SpaceX has vertical integration, wholesale launch costs, and an existing operation already launching large volumes of Starlink satellites.
I can put a ton of mass in the ocean a hundred times cheaper than I can put it in orbit.
That comparison is the center of Schroepfer’s investment skepticism. He said he is willing to examine any pitch — “I’m in the business of looking at everything,” and will “update my priors” if new information changes the math — but when he has modeled space-based data centers, he has found them “doable” and “extraordinarily expensive.”
The expense compounds across several categories: launch, radiation hardening, and radiators large enough to dissipate server heat. Servers produce substantial heat; in space, that heat must be radiated away into a vacuum, requiring hardware designed for that purpose. After adding those costs, Schroepfer said he does not see orbital data centers becoming cost-competitive with current land-based or ocean-based alternatives unless the cost of building on land rises dramatically.
The exception is SpaceX. If SpaceX gets launch, integration, and related infrastructure “at wholesale costs,” its economics differ from everyone else’s. Schroepfer called space-based data centers “a great business” for SpaceX, while remaining unclear on “how anyone else can compete on costs.”
The ocean case rests on mass, cooling, and untapped power
Schroepfer’s alternative is not only that the ocean is cheaper to reach. The ocean also offers useful physical properties for compute: mass can be deployed there far more cheaply, and cold seawater provides “almost unlimited free cooling.” In contrast, space cooling requires purpose-built radiators and adds cost.
Bloomberg’s on-screen graphic listed companies in the Gigascale Capital portfolio: Heron, Radiant, Fractile, Panthalassa, and Arch. Schroepfer specifically named Panthalassa when describing ocean-based data center possibilities, while making clear that his thesis extends beyond any single bet.
He pointed to the Southern Ocean as a major untapped energy resource, saying there are “10 terawatts of power” there that remain unused. Ocean data centers are one possible way to use large energy resources that are not currently part of the computing supply chain.
The ocean pitch is not an argument that all compute should move offshore. It is a cost and resource argument: if the bottleneck is the ability to generate tokens cheaply and reliably, then ocean infrastructure may have an advantage before orbit does. The ocean offers cheaper deployment of mass and direct access to cold water for cooling; orbit carries much higher deployment and thermal-management costs for anyone without SpaceX’s cost structure.
When Ludlow asked how submarine data centers compare with space-based data centers on a “dollar per token” basis, Schroepfer returned to assumptions. The comparison depends on the full model. But on “some basics,” he said, the ocean currently remains about 100 times cheaper than space for placing equivalent mass.
Land is not exhausted, and solar can move faster than gas or orbit
Mike Schroepfer also resisted the premise that the next compute buildout must leave land. He said he helped build “tens of millions of square feet” of cloud computing capacity while serving as Meta’s CTO, and he does not think the industry is close to finished using land-based solar resources.
His argument here is practical and time-based. With solar and batteries, he said, a new power plant can be put together in 12 to 24 months. He described that pace as faster than building gas generation or launching infrastructure into orbit. For a compute-constrained market, time to available power matters as much as the infrastructure design.
| Option | Schroepfer's stated advantage | Schroepfer's stated constraint |
|---|---|---|
| Space-based data centers | Potentially viable for SpaceX because of wholesale launch costs and existing large-volume Starlink launches | Launch, radiation hardening, and heat rejection make it extraordinarily expensive for others; no one has done it yet |
| Ocean data centers | Roughly 100 times cheaper than orbit for placing mass, with cold seawater for cooling | No one has done it yet; Schroepfer said both ocean and orbital systems still need to prove they can generate tokens reliably |
| Land-based solar and batteries | New power plants can be assembled in 12 to 24 months | Schroepfer did not say land alone solves all compute demand |
This is why his position is not anti-space. Space can work where SpaceX internalizes the expensive parts. Ocean can work if companies prove they can generate tokens reliably from offshore infrastructure. Land remains attractive because solar and batteries are deployable quickly and, in Schroepfer’s view, still underused.
The market will care when the tokens are real
The limiting factor for both ocean and orbital data centers, Mike Schroepfer said, is proof. Ludlow asked whether Meta, given its large AI infrastructure spending, would be open to submarine or orbital data centers as alternatives. Schroepfer’s answer was that providers will pay attention when the systems produce tokens in front of them.
His imagined demonstration was concrete: a terminal showing a chat experience, paired with a live webcam of “a little buoy bobbing in the Southern Ocean,” or a live satellite feed, with the operator saying that this is where the tokens are being generated. At that point, he said, everyone will “perk up” and give the technology more attention.
Until then, both ocean and orbital data centers remain theoretical. “No one’s done either of these things,” he said. The requirements from customers are straightforward: reliability, availability, and compute now. If a provider can deliver usable compute, buyers will consider it.
At the end of the day, people want reliability, they want availability, and they want tokens now.
That test cuts across the whole argument. SpaceX may have the orbital cost structure. Ocean data centers may have the cheaper mass deployment and cooling profile. Land-based solar and batteries may be faster to bring online. But for customers, Schroepfer suggested, the decisive proof is not the infrastructure thesis. It is whether the system can generate tokens reliably when demand exists.