SpaceX's IPO filing reveals three AI lab customers worth about $27.8 billion in annualized contract value. One is named: Anthropic. The other two aren't. That concentration is the prospectus's load bearing risk.
SpaceX filed to go public as a rocket company. The arithmetic of the offering, however, no longer adds up that way.
According to the company's S-1 and S-1/A registration statements on file with the SEC, SpaceX's integrated AI compute business, built around the assets it absorbed from xAI, has three paying customers generating roughly $27.8 billion a year in contracted revenue (S-1, SEC EDGAR, S-1/A#2, SEC EDGAR; figures as summarized by Motley Fool and CryptoBriefing). Only one of those three counterparties is disclosed by name in the filings.
That single named customer is Anthropic. The AI lab signed on for more than 300 megawatts of compute capacity at a price of about $1.25 billion a month, or roughly $15 billion a year, per separate reporting from TechCrunch, Wired and Business Insider. The two other AI compute customers are not identified in the prospectus, an asymmetry consistent with the customer-concentration disclosure framework that SEC registration statements use (S-1, SEC EDGAR).
Two things distinguish the $27.8 billion figure from actual booked revenue. First, it is annualized contract value, the run rate implied by multi-year purchase commitments, not revenue recognized under GAAP. Second, it is read off the same S-1 disclosures that devote substantial space to the distinction, the company's own way of telling prospective investors that the headline number is not the same line that will appear on the income statement (Yahoo Finance roundup of the S-1 disclosure). For investors used to reading software and infrastructure prospectuses, that is a familiar construct. For general investors meeting SpaceX at IPO for the first time, it is the framing the prospectus itself builds around the dollar figure.
The concentration is the load-bearing risk. Three customers, one named, one revenue line that did not exist before the IPO filing, and a counterparty set drawn from a handful of AI labs whose own funding trajectories are still being written. The S-1 itself treats customer concentration as a material risk factor, the prospectus version of the line every infrastructure pitch eventually meets: the operator's revenue is only as durable as the buyer's balance sheet (S-1/A#2, SEC EDGAR).
That dependency runs through xAI. Before the merger, xAI was the largest single drag on SpaceX's 2025 bottom line; the AI compute segment the prospectus now describes is the integrated successor to that operation, and the 2027 inflection the filings point to is when contracted compute revenue begins to ramp materially (S-1, SEC EDGAR). The same business that was burning cash eighteen months ago is now the unit underwriting SpaceX's public-market valuation.
What to watch from here. The next S-1 amendment will reveal whether management narrows the gap between annualized contract value and recognized revenue, the most direct test of whether the $27.8 billion figure is a contract artifact or a near-term earnings line. The two unnamed customers will eventually become legible through either amended filings, peer AI lab disclosures, or downstream capacity-offtake announcements. And the concentration metric itself, how much of projected revenue depends on each of the three counterparties, will tell public-market investors whether the offering prices as a rocket company or as a single-segment credit exposure dressed in rocket branding.
For now, the public filing gives readers two numbers worth separating: a $27.8 billion annualized contract figure, and the GAAP revenue recognition timeline that determines when the rockets-company story becomes the AI-utility story on the income statement.