Back to articles

Chips may survive space. Their financing may not.

3 min read

Getting chips into orbit is not just a rocket and engineering challenge. It raises a financing question that terrestrial data-center operators rarely face: if a borrower defaults, can lenders recover the equipment and sell it?

Why collateral matters

Data-center companies commonly borrow to build facilities and repay those loans from future revenue. The chips themselves can serve as collateral, giving lenders a fallback if a borrower cannot pay. Even older accelerators may retain resale value. The source cites Compute Credit Index Research, which says some H100s are selling for about half their 2023 cost basis, while some H200 configurations are selling for roughly their 2024 launch price. The ability to recover and resell hardware helps support these financing arrangements.

Orbit changes that calculation. Lenders cannot readily repossess chips installed in a satellite hundreds of miles above Earth and moving at orbital speed. The Financial Times reported that lenders behind a roughly $40 billion SpaceX chip-financing deal asked that none of the chips bought with the financing be used for orbital data-center plans. The report does not establish whether SpaceX accepted that condition. The central issue is not simply whether chips can function in space, but whether they remain assets a lender can control and liquidate.

Ground-based growth versus orbital ambitions

Any restriction may not immediately slow SpaceX’s computing expansion. The source says the company has found terrestrial deployments at its ground stations lucrative. On its latest earnings call, Elon Musk described assigning rocket-team engineers to accelerate that build-out, including work on casting turbine blades for generators faster than existing suppliers. But ground-based facilities face their own constraints: access to land and power, project delays, local opposition, and uncertainty about whether customers can turn demand for compute into revenue.

Musk began saying in early 2026 that space could be the cheapest place to operate data centers within two to three years. That outlook depends in part on Starship becoming fully reusable. In the latest test described in the source, SpaceX did not recover the booster, although the orbital stage landed back at the company’s Texas launch site. When that stage will fly again, and what refurbishment will cost, remain open questions.

Even with better launch economics, SpaceX and other companies—including StarCloud, Cowboy Space Corporation, Sophia Space and Orbital—still need to buy chips. If lenders cannot rely on recoverable collateral, orbital projects may have to pay more for capital than ground-based competitors. Google’s plan to deploy its own TPUs through Project Suncatcher could give it a different cost structure. SpaceX and Tesla are also working on a US chip foundry, Terafab, but the project has not broken ground and would require years and billions of dollars to reach production at scale.

The business case is more than launch cost

The financing concern highlights a broader test for space data centers: hardware must not only operate in orbit, but also fit a workable investment model. Cheaper launches alone would not resolve the problem if chips cannot be repossessed or resold. The economics will also depend on lending terms, chip sourcing, ground infrastructure and whether computing revenue can justify the upfront investment.

Source: TechCrunch AI

Comments

Checking sign-in status...

Loading comments...

Related articles