SpaceX’s AI division turns profitable, with Anthropic holding the keys

5 hours ago 21

Compute rentals to Anthropic pushed SpaceX's Super Intelligence unit into the black, but the contract lets its biggest customer walk on 90 days' notice

SpaceX’s AI division has posted its first operating profit. The company that built its name on rockets and satellite internet is now making serious money renting out computing power.

One customer did most of the lifting: Anthropic. The deal that put the unit in the black also carries exit clauses that let Anthropic leave with 90 days’ notice once certain dates in early 2027 pass. That is a lot of profit resting on a relatively short leash.

The numbers behind the turnaround

The AI unit, recently rebranded as Super Intelligence, reached operating profitability in the third quarter of 2026. Its engine is a compute rental agreement with Anthropic.

Under the deal, Anthropic gets access to Nvidia GPU capacity at SpaceX’s Colossus I and II data centers in Memphis, Tennessee.

SpaceX reported total Q3 revenue of $14.58 billion. Anthropic accounted for approximately 34.5% of that, or around $5 billion.

That figure makes Anthropic SpaceX’s largest customer, ahead of the US government.

The AI unit itself brought in $7.56 billion in revenue during the quarter. Simple arithmetic shows Anthropic’s roughly $5 billion made up the clear majority of that segment’s take.

The results helped deliver a net profit of $4.4 billion for SpaceX overall. That is the company’s first net profit since it went public in June 2026.

Anthropic’s bigger spending plans

The relationship looks set to get larger, at least on paper. Anthropic’s S-1 filing, dated October 26, 2026, disclosed commitments of up to $120.5 billion through 2029.

Once the arrangement reaches full service capability, which is expected in December 2026, Anthropic’s spending is projected at about $3.1 billion per month. At that pace, a single quarter of spending would comfortably exceed the roughly $5 billion Anthropic contributed in Q3.

The fine print: a 90-day exit

Either party can terminate the agreement with 90 days’ notice once non-cancellable periods end in January and March 2027.

That stands in sharp contrast to Anthropic’s arrangements with other major compute providers, Google and Microsoft. Those deals involve longer-term commitments, according to the research findings.

From launchpads to server racks

In recent quarters, the AI segment has generated more revenue than Starlink and launch services combined.

What this means for SpaceX and its shareholders

When roughly a third of total revenue and most of a segment’s revenue come from one customer, concentration risk becomes the central question.

The 90-day termination window sharpens that risk. After January and March 2027, SpaceX’s single largest revenue source could theoretically shrink or disappear on short notice. Investors who model the projected $3.1 billion monthly run rate forward through 2029 are assuming Anthropic stays, something the contract does not require.

Anthropic’s longer commitments with Google and Microsoft give it alternatives, while the short exit clause with SpaceX preserves leverage at renewal time.

The key dates to watch are January and March 2027, when the non-cancellable periods end. Also worth tracking is whether full service capability arrives on schedule in December 2026, since the projected monthly spending hinges on it.

Anthropic’s commitment of up to $120.5 billion through 2029 signals sustained appetite for compute. For SpaceX, the challenge ahead is turning a lucrative but cancellable relationship into a durable business.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

Read Entire Article