SpaceX’s AI Data Centers
Coverage from AOL, Data Center Dynamics, and others

SpaceX is building a terrestrial AI-computing business around data centers such as Colossus 2, expanding its GPU capacity alongside the power and cooling needed to operate it.
The company has reported cloud-services agreements with customers including Anthropic and Google, while Microsoft has been discussed as a potential major buyer. The scale of the buildout makes financing, customer demand, utilization, and the price of computing central questions: substantial contracted demand could support returns, but heavy capital requirements and changing prices or commitments could leave capacity underused.
If you read one thing
It gives the clearest current overview of SpaceX’s rapid capacity plans alongside the unresolved demand, power, and investment risks.
Best explainer
It explains the scale of the financing and capital requirements and how uncertain pricing and cancellable contracts could affect returns.
The evidence
It adds concrete reporting on major hosting agreements and planned capacity growth, grounding the customer-demand opportunity in specific deals.
Rapid capacity expansion
SpaceX is increasing terrestrial AI compute capacity, with 1.7 GW reported at the end of Q3 and management targeting much larger capacity in 2027. Delivering those targets depends on building facilities and supporting infrastructure at an unusually fast pace.
Large customer commitments, uncertain durability
Reported agreements with multiple AI customers support a substantial compute-leasing opportunity. However, short or cancellable agreements and spot-market leasing leave future revenue dependent on delivering capacity and sustaining demand rather than on fully locked-in long-term commitments.
Capital intensity and financial exposure
The buildout requires exceptionally large investment: reported spending is rising sharply, has exceeded revenue in recent periods, and is accompanied by substantial external funding estimates and reported chip-financing plans. This expands the financial exposure before the business’s returns are established.
$37.5 billion USD
Nebius remaining performance obligations
“Nebius reported $37.5 billion in remaining performance obligations.”
$10 billion USD
proposed bank loans
“According to a Bloomberg report, Space Exploration Technologies (SpaceX) is seeking $40 billion in financing to purchase AI chips from Nvidia. Apollo Global Management is rumored to be leading the transaction, which would include $10 billion of bank loans and $30 billion of investment-grade debt. SpaceX, Nvidia, and Apollo have not confirmed the deal, which would not be expected to close until 2027.”
$30 billion USD
proposed investment-grade debt
“According to a Bloomberg report, Space Exploration Technologies (SpaceX) is seeking $40 billion in financing to purchase AI chips from Nvidia. Apollo Global Management is rumored to be leading the transaction, which would include $10 billion of bank loans and $30 billion of investment-grade debt. SpaceX, Nvidia, and Apollo have not confirmed the deal, which would not be expected to close until 2027.”
$40 billion USD
proposed financing sought by SpaceX
“According to a Bloomberg report, Space Exploration Technologies (SpaceX) is seeking $40 billion in financing to purchase AI chips from Nvidia. Apollo Global Management is rumored to be leading the transaction, which would include $10 billion of bank loans and $30 billion of investment-grade debt. SpaceX, Nvidia, and Apollo have not confirmed the deal, which would not be expected to close until 2027.”
roughly six months
hosting agreement duration
“Many of the hosting agreements run for roughly six months with early-exit options, so the revenue is not locked in.”
SpaceX raises its near-term compute target while demand remains unsecured
SpaceX reported 1.7 GW of rated compute at Q3 and projected at least 13 GW by the end of next year, alongside sharply higher planned capital spending. The larger buildout remains a target, not secured demand: the company reportedly leases at spot rates and disclosed no chipmaker purchase orders.
Previously
SpaceX is rapidly expanding terrestrial AI compute capacity and positioning data-center hosting as a major business. Reports place operating capacity at about 1.4 gigawatts in mid-2026, with higher near-term targets and plans for several more gigawatts in 2027. Large agreements with AI and cloud customers point to demand, while heavy capital needs, uncertain contract durability, chip and power requirements, and permitting issues complicate the economics.
Reported rated capacity increased to 1.7 GW in Q3, showing the buildout is advancing; new forecasts quantify a much larger near-term target, though financing remains uncertain.
The main new detail is that reports attribute most of SpaceX’s capital spending to AI compute infrastructure. The broader picture of rapid expansion and unresolved commercial and operational risks is unchanged.
