How SpaceX’s AI Problem Became A $26B Google and Anthropic Deal

Three weeks ago, xAI looked like one of SpaceX’s biggest financial drags. The AI segment posted a $2.47 billion operating loss in Q1 2026, while AI capex reached $12.7 billion in 2025 and another $7.7 billion in Q1 2026. Grok was also facing sharper competition from Claude, Gemini, and ChatGPT. The math was blunt: Starlink was printing money, and xAI was consuming most of it.

Then two things happened. On May 20, SpaceX's S-1 filing disclosed that Anthropic had agreed to pay $1.25 billion a month to rent the entire output of xAI's Colossus 1 data center till May 2029. On June 5, a separate SEC filing revealed that Google had signed a deal with SpaceX at $920 million a month till June 2029. Combined, these two agreements represent approximately $26 billion in annualized compute revenue, from two of the best-funded AI companies in the world.

Let's break down what these deals actually cover, why xAI had this much capacity sitting idle in the first place, how the new revenue math reshapes the valuation story, and what should still give investors genuine pause.

SpaceX xAI Compute Deals With Anthropic and Google: Key Terms

The Anthropic deal, disclosed in SpaceX's S-1 filing with the SEC, gives Anthropic exclusive access to Colossus 1, the Memphis, Tennessee data center xAI originally built to train its Grok models. The first two months carry a discounted rate while the infrastructure ramps up; the full $1.25 billion monthly rate takes effect after that. Annualized at full rate, this one customer relationship alone is worth approximately $15 billion per year.

The Google deal, filed separately on June 5, 2026, pays SpaceX $920 million a month starting in October, after capacity ramps through September at a reduced fee, with the full arrangement running through June 2029. A Google Cloud spokesperson described it as a "short-term, timely agreement to ensure we have bridge capacity" for its Gemini Enterprise agent platform. That choice of words, "bridge capacity," carries real investor significance, and we'll come back to it.

Both contracts include 90-day termination clauses after December 31, 2026. Neither side is permanently locked in.

A third agreement, disclosed in April 2026, involves AI coding startup Cursor using xAI data center capacity as part of a $10 billion collaboration deal. SpaceX also secured the right to acquire Cursor for $60 billion later this year, or pay $10 billion for ongoing collaboration on AI coding and knowledge work. SpaceX additionally disclosed in its filing that it "expects to enter into additional similar services contracts for compute capacity with third parties," signaling this is not viewed internally as a one-off arrangement.

Why xAI Had Idle Colossus 1 Compute Capacity

Here is the detail that makes these deals more complicated than a straightforward revenue upgrade.

The reason SpaceX offered Colossus 1 to Anthropic is that xAI was barely using it. xAI had moved its training functions to Colossus 2 after finding that the mixed H100, H200, and GB200 GPU architecture at Colossus 1 made efficient large-scale model training difficult. Before the Anthropic deal closed, the effective utilization rate at Colossus 1 was approximately 11%. SpaceX's own S-1 described the Anthropic arrangement as a mechanism that "allows us to monetize unused compute capacity in our infrastructure." That is honest disclosure, and it deserves to sit at the center of how investors read these deals.

Think of it like an IT park developer in Hyderabad who built a sprawling campus expecting their own software company to fill every floor. The company never grew into the space. So rather than let the building sit empty, they started leasing floors to Infosys and TCS at premium rates. The building now earns good rent. But the software company on the ground floor is still dealing with the same product challenges that left the upper floors empty in the first place.

That is the honest structure of the Colossus 1 situation. xAI built more than $30 billion worth of GPU infrastructure to power Grok. Grok's usage dropped as competitors like Claude and Gemini gained ground. The hardware sat idle. And rather than carry that idle infrastructure as pure cost, SpaceX rented it to two of the companies most directly competing with Grok in the market.

The revenue is real. But the reason it exists tells investors something important: these contracts are not evidence of xAI's AI model strength. They are evidence of its infrastructure surplus, which is a different thing entirely.

Elon Musk has publicly stated that SpaceX is actively seeking more AI compute customers following the Anthropic deal. If additional hyperscalers follow, the revenue line could grow further. If Grok improves and xAI wants that capacity back for its own training, things get complicated.

How the Compute Deals Change SpaceX’s Revenue Run Rate

Taking these contracts at stated terms, the forward revenue picture for SpaceX changes materially. The estimates below use annualized run rates and are not official company guidance.

Note: These are rough projections. Actual numbers will depend on whether Google's capacity is fully delivered by the September 30, 2026 deadline, Starlink's subscriber growth trajectory, whether additional compute customers sign on, and whether the Anthropic and Google deals hold through 2027.

The headline though is that against SpaceX's 2025 revenue of $18.7 billion, this forward picture represents a significant step-up in revenue scale, driven almost entirely by the AI compute contracts.

What the New Estimated Revenue Multiple Means for SpaceX IPO Valuation

A forward multiple of roughly 38-43x is still expensive. Nvidia, at peak AI enthusiasm in 2023-2024, traded at approximately 20-35x revenue. Microsoft typically sits at 12-14x. But 38-43x is a different conversation from 95x. The original valuation required investors to underwrite an AI segment with almost no external revenue visibility. The new picture has named customers on monthly contracts.

The caveat that matters here is that AI segment capital expenditure ran at $7.7 billion in Q1 2026 alone, suggesting an annualized capex run rate of approximately $30.8 billion. Revenue growth that is offset by continued capex of that scale may not translate into meaningful margin improvement at the segment level in the near term. The revenue line is moving. The cost line may be moving just as fast.

Why the Timing of SpaceX’s Compute Deals Matters Before IPO

The Anthropic deal was disclosed in the S-1 on May 20, 2026. The Google deal was filed on June 5. SPCX was priced at $135 per share on June 3, 2026, with first-day trading on June 12. Both deals became public within three weeks of pricing. 

Whether that sequence reflects standard negotiation timelines, deliberate pre-IPO preparation, or something in between is not something that can be confirmed from the outside. What is observable: both filings transformed the AI segment's narrative from "loss-making with no named customers" to "contracted infrastructure provider with $2.17 billion in monthly committed revenue" at the most consequential public moment in the company's history. Google holds approximately 5% of SpaceX, which means it has a direct financial stake in the IPO pricing as well. A deal that simultaneously gives Google needed compute capacity and strengthens SpaceX's IPO revenue narrative is not a neutral transaction from a governance standpoint.

Whether you call it coincidence or strategy probably depends on how you read Elon Musk's history with pre-announcement timing.

Key Risks to the SpaceX xAI Compute Revenue Thesis

The core thesis risk is this: both contracts are short-cycle bridge agreements. They convert idle infrastructure into revenue, which is a smart use of an expensive asset. But they are not 10-year infrastructure deals. If Google and Anthropic execute on their own capacity buildouts, the entire $26 billion in annualized compute revenue could compress significantly within 18 to 24 months.

The deals work as long as the AI infrastructure shortage holds. The AI infrastructure shortage is exactly what every large tech company is spending hundreds of billions to fix.

What the xAI Compute Deals Mean for SpaceX Investors

The compute deals change SpaceX's near-term revenue math in a real way. What was a segment earning $0.82 billion per quarter while losing $2.47 billion per quarter now has approximately $2.17 billion in contracted monthly compute revenue from two creditworthy customers. That is not a cosmetic change to the story.

At the same time, these deals exist because xAI's AI models didn't fill the infrastructure that was built for them. Investors need to hold both of these things simultaneously. Buying SpaceX on the thesis that xAI will become an AI model leader is a different bet than buying SpaceX because it accidentally assembled one of the largest GPU rental businesses in the world. Both bets might work. But they are not the same bet.

The valuation has become less indefensible. The forward revenue math looks meaningfully different at 38-43x than at 95x. But the governance concentration, the $29.1 billion in long-term debt, the cancellation clauses in these contracts, and Starlink's ARPU compression have not changed. A stronger near-term revenue picture is not the same as a resolved risk profile.

For investors who were waiting for post-listing price discovery before making an evaluation, the compute contracts add real credibility to the bull case. But they also do not remove the reasons to wait.