The SpaceX AI Revenue Story Has a 90-Day Expiration Date

Goldman Sachs projects $322 billion in AI revenue by 2030. Goldman is also the lead underwriter. Here is what the S-1 actually says.

“When a bank tells you a stock is worth buying and is also getting paid to sell you that stock, the conflict of interest is not incidental. It is the business model.” — The Charted Course

Twenty-two days before SpaceX priced its IPO, the company filed an amended S-1 disclosing two data center contracts that changed the shape of the offering. Anthropic agreed to pay $1.25 billion per month for access to 325,000 Nvidia GPUs across the Colossus I and II facilities in Memphis, Tennessee. Google signed a separate deal worth $920 million per month for 110,000 GPUs (SpaceX S-1 amendment, June 5, 2026; TechCrunch, June 5, 2026). Combined, those contracts represent approximately $2.2 billion in monthly revenue from AI infrastructure.

Both contracts include 90-day termination clauses. Either party can exit after the initial period with 90 days' notice. Goldman Sachs, projecting SpaceX's AI revenue growing from $3.2 billion in 2025 to $322 billion by 2030, is also the lead underwriter on the IPO. That is the detail the headline numbers do not include. For the broader pricing context, see what the IPO valuation actually requires investors to believe.

What the S-1 Actually Says

The Colossus I and II data centers were built by xAI, Elon Musk's AI company, for its own model training workloads. xAI has since migrated its training operations to Colossus II. Colossus I, now partially vacated by the original tenant, is what Anthropic is currently paying $1.25 billion per month to occupy.

SpaceX's own S-1 language describes the arrangement as a way to “monetize unused compute capacity.” That framing matters. The infrastructure was not purpose-built for external customers. It was built for internal use, and the external contracts emerged because the internal tenant moved out. Anthropic is filling a vacancy, not anchoring a purpose-built AI infrastructure business.

Musk himself posted on X that the arrangement is a “180-day lease with 90-day mutual cancellation notice.” The S-1 references a contract running through May 2029. Those two descriptions do not say the same thing. The ambiguity has not been publicly resolved. Investors pricing SpaceX as an AI infrastructure company are relying on contracts whose duration is disputed by the CEO's own public statements.

The contracts at a glance

The Goldman Problem

Goldman Sachs published projections in advance of the IPO showing SpaceX's AI segment revenue growing from $3.2 billion in 2025 to $322 billion by 2030. That is a 100-fold increase in five years. Goldman is also the lead underwriter on the IPO, meaning the bank is collecting fees directly tied to the success of the offering.

This is a structural conflict of interest that is common in large IPOs and rarely named directly. The underwriter's job is to price and sell the offering. Projections that support a higher valuation serve that job. Projections that complicate the valuation story do not. This does not mean the Goldman projections are wrong. It means the incentives of the person producing the forecast are not aligned with the incentives of the person reading it.

For context on the AI revenue projections: xAI, acquired by SpaceX in February 2026, had approximately $500 million in 2025 revenue. Within SpaceX's combined entity, it is being valued at $250 billion — a 500x revenue multiple. Cloudflare, widely considered a successful high-growth tech IPO, debuted at approximately 30x revenue. The xAI implied multiple is roughly 16 times that.

What the Disclosure Timeline Looks Like

The sequence of events is worth stating plainly. SpaceX filed its original S-1 in early 2026. The Anthropic and Google contracts were disclosed in S-1 amendments filed on May 20 and June 5, 2026 — the latter just six days before the June 11 IPO pricing date. The Goldman projections incorporating those contracts circulated in the weeks before pricing. The retail investor event was held on June 11, the day of pricing.

The contracts significantly change the revenue picture. Combined, the AI data center contracts add approximately $26 billion in annualized contracted revenue — roughly doubling SpaceX's existing Connectivity segment run rate. Starlink, the core recurring business, generated $3.26 billion in Q1 2026 across 10.3 million subscribers in 164 countries (SpaceX S-1 amendment).

The timing — material contracts disclosed weeks before pricing, incorporated into underwriter projections, presented to retail investors on pricing day — is not illegal. It is how large IPOs work. That is precisely the point. The structure is designed to maximize excitement at the moment of maximum commitment. Understanding the structure is not cynicism. It is literacy.

The Mechanism Behind the Certainty

Complexity Paralysis is the behavioral mechanism most relevant here. When financial information is genuinely complex — competing S-1 interpretations, 90-day clause implications, underwriter conflicts of interest — most people resolve the complexity by defaulting to the simpler, more emotionally satisfying narrative. In this case: SpaceX is an AI infrastructure company with $26 billion in annual contracted revenue and a clear path to dominance.

The simpler story is not wrong because it is simple. It is wrong because it omits the cancellation clause, the disclosure timeline, the CEO's contradictory public statement, and the conflict of interest in the projections. Complexity Paralysis is why those omissions survive. Holding all four details simultaneously requires effort. The brain prefers the version that requires less. Authority Bias compounds it — when Goldman Sachs attaches a number to a projection, the institution's reputation substitutes for the analytical work most readers cannot do themselves.

What the Conventional Wisdom Gets Right

The underlying contracts are real. Anthropic and Google are paying for GPU access. The infrastructure exists. The revenue, while it lasts, is genuine. A company that can attract over $2 billion per month in data center commitments from two of the most sophisticated AI organizations in the world has built something that matters.

The conventional wisdom is also right that AI infrastructure is a genuine growth market. Demand for GPU compute is not fabricated. If SpaceX can retain and expand these contracts, the AI segment could become a durable revenue driver alongside Starlink.

The conventional wisdom is wrong about one thing specifically: certainty. The contracts are real. The duration is not guaranteed. The projections are Goldman's best case produced by the bank collecting fees to sell the stock. None of that makes SpaceX a bad investment. It makes the current price a bet on a specific future that has not yet been earned.

The Honest Summary

SpaceX disclosed approximately $26 billion in potential annual AI infrastructure revenue — on infrastructure its original tenant vacated, under contracts either party can exit before Q4 2026, based on projections produced by the bank collecting fees to sell the stock.

The contracts may hold. The projections may prove conservative. SpaceX may become the dominant AI infrastructure provider of the next decade. All of that is possible.

What is not possible is knowing any of that before the 90-day cancellation windows expire — and the IPO priced before those windows even started.

Educational content only. Not financial, investment, legal, or tax advice. Consult a qualified professional before making financial decisions.