Buried on page 13 of SpaceX's S-1 prospectus is the line item that has done more to justify a reported $2 trillion valuation than any other single disclosure in the filing.
Anthropic, the rival AI company that Elon Musk publicly called "evil" earlier this year, is paying SpaceX $1.25 billion every month for compute capacity inside the Colossus 1 and Colossus 2 data centers in Memphis. The filing describes the payments running through May 2029. Annualized, that amounts to $15 billion a year. Spread across the full term, the deal totals roughly $45 billion.
More Top Stocks Daily: Go behind Wall Street’s hottest headlines with Barchart’s Active Investor newsletter.
Analyst notes have rapidly translated that disclosure into the bull case for SpaceX, which will trade under the ticker symbol SPCX after its planned IPO in June. The company can now tout secured revenue; bona fides for an AI infrastructure pivot; and a floor under the $4.28 billion in quarterly losses that an otherwise-profitable rocket and satellite business is now carrying. One pre-IPO research desk even called it "the strategic masterstroke ahead of the IPO."
Then Elon Musk got on X and clarified the terms.
"SpaceX has not committed to leasing Colossus for years, although it's possible that may be what happens," Musk wrote. "The agreement is a 180-day lease with a mutual 90-day cancellation notice thereafter."
In other words, the $45 billion deal is actually a six-month rental.
What the S-1 Actually Says, and What It Leaves Out
The prospectus describes monthly payments of $1.25 billion through May 2029 and discloses that either party can terminate with 90 days’ notice. It does not disclose that the underlying commitment is a 180-day lease.
Given that, investors reading the document without Musk's X clarification would reasonably model a multi-year contract with cancellation optionality. However, they would be wrong.
The distinction matters because of the way the rest of the SPCX valuation math works.
SpaceX generated $18.7 billion in revenue in 2025. Starlink contributed $11.4 billion of that and posted a $4.4 billion operating profit, making it the only segment consistently profitable on a GAAP basis.
The rocket business (i.e., Falcon, Dragon, Starship) generated $4.1 billion in revenue and lost $657 million, with $3 billion of that flowing into Starship research and development.
The newly-consolidated artificial intelligence (AI) segment, which includes xAI, Grok, and the X advertising platform, generated $3.2 billion in revenue and lost $6.4 billion. In Q1 2026 alone, the AI segment burned $2.5 billion on $818 million of revenue.
Within that segment breakdown, the Anthropic agreement is the swing factor.
Without that $15 billion annual run rate, the AI division's revenue base collapses to roughly $3 billion against losses that the company has guided will continue scaling. With it, the segment looks like a credible third leg of the business… a data center landlord renting capacity at hyperscale prices to one of the most well-capitalized AI companies in the world.
The 180-day framing changes which one of those pictures investors will be buying.
The Contradiction at the Center
Anthropic is no routine customer for SpaceX. The company spent most of 2025 and early 2026 as the explicit competitive target of Musk's AI strategy, and xAI was specifically founded to compete with Anthropic and OpenAI. Musk personally and repeatedly attacked Anthropic on X, including calling the company "evil" in the months immediately preceding the Colossus deal.
Colossus 1 itself was built to train Grok. The 300-megawatt facility went fully operational in December 2024 after what Musk's team treated as a point of pride, a 122-day build. Eighteen months later, the same facility is being rented to the company xAI was created to beat.
The rationale lies in the numbers. Grok's user base is far smaller than competing products from Anthropic and OpenAI. The AI segment's revenue is not large enough to support the capital expenditure required to keep building. SpaceX disclosed $7.7 billion of capital expenditure in the AI segment in Q1 2026 alone, which is more than Space and Connectivity combined. Without external compute revenue, the segment is a capital sink.
Anthropic, meanwhile, is hampered by a compute shortage at exactly the moment the company needs to scale Claude. Renting Colossus solves the immediate problem.
The arrangement makes commercial sense for both parties. What it does not do is provide the kind of multi-year revenue anchor that a $2 trillion valuation requires.
What "Either Party Can Terminate" Actually Means
Standard data center economics depend on decade-long anchor tenants. Hyperscalers like Microsoft (MSFT), Meta (META), and Google (GOOG) (GOOGL) sign 10-to-15-year leases with locked-in capacity commitments precisely because the underlying capital expenditure (billions of dollars in GPUs, networking, and power infrastructure) requires that kind of duration to underwrite.
A 180-day lease with a 90-day cancellation clause is not that. It is a month-to-month rental with a notice period.
For Anthropic, the optionality cuts in its favor. The company can use Colossus to bridge to its own capacity buildout with Amazon (AMZN), Google, and other infrastructure partners. If those alternatives come online faster than expected, Anthropic exits with one quarter of notice.
For SpaceX, the optionality cuts harder. If compute demand from xAI accelerates – say, if Grok user growth picks up, or if Musk's stated ambition to "serve AI at extremely large scale" requires more in-house capacity – the company can claw back the cluster.
Or, to use Musk's exact phrasing on X: "if compute gets super tight, I said we might need it back at some point."
That language is the disclosure equivalent of a yellow flag. It tells public market investors that the $15 billion annual revenue line item they are pricing into the IPO can be revoked by either side, including by SpaceX itself, on a quarterly cadence.
The Valuation Math Without the Anthropic Line
The $1.75 trillion target valuation implies an enterprise value-to-revenue multiple of roughly 70x on 2026 revenue if the Anthropic payments are included at the full run rate. Strip the Anthropic line out and the multiple expands significantly, because the consolidated revenue base shrinks while the AI segment's losses remain.
Tesla (TSLA), the closest public proxy for Musk-led companies, currently trades at a price-to-sales ratio of roughly 11x. Nvidia (NVDA), the AI infrastructure benchmark, trades at roughly 25x sales after this month's blowout earnings. Microsoft, the most relevant cloud hyperscaler comparison, trades at roughly 13x sales.
SPCX at $1.75 trillion on the disclosed financials sits in a category by itself. The argument that the valuation is justified rests on three forward bets: that Starlink V3 satellites deployed in the second half of 2026 will materially expand the connectivity business; that Starship will eventually unlock new markets like orbital data centers and satellite-to-mobile services; and that the AI compute business, anchored by the Anthropic agreement, will scale into a third major revenue engine.
The first two bets are about execution. The third bet depends on a six-month rental.
The Read for IPO Investors
The SPCX listing is expected to price as early as June 11 and begin trading on the Nasdaq as early as June 12. Goldman Sachs is lead-left on a syndicate of 21 underwriters. Retail allocation is reportedly earmarked at 30% of the float, roughly three times the standard for a mega-cap IPO. Index demand will be substantial, and passive funds tracking the S&P 500, Nasdaq-100, and FTSE Russell indices will need to buy SPCX regardless of fundamentals once it qualifies for inclusion.
For investors planning to participate, the relevant due diligence is not whether the rocket business or Starlink can grow. Both have track records that justify large positive value. The question is how much of the $1.75 trillion is being assigned to the AI compute business, and what fraction of that AI value is collateralized by a contract Musk himself has now characterized as a 180-day commitment with 90-day cancellation.
If that fraction is small, the IPO is a Starlink-and-rockets story with optional AI upside. If that fraction is large, the IPO is a Starlink-and-rockets story sold at an AI compute multiple, with the AI compute revenue resting on a rental agreement that could end before the lockup period expires.
The S-1 does not answer that question explicitly. Musk's X post answered the part the prospectus left out. Both documents are now in the public record, and the market gets to decide which one matters more.
On the date of publication, Caleb Naysmith did not have (either directly or indirectly) positions in any of the securities mentioned in this article. All information and data in this article is solely for informational purposes. For more information please view the Barchart Disclosure Policy here.
More news from Barchart
Applied Materials Stock Just Hit New Record Highs. Don’t Count the Rally Out Just Yet. The 180-Day Lease Holding Up SpaceX's $2 Trillion IPO TDOC Stock Pops as Teladoc Teams Up With Walmart A $6 Billion Reason Snowflake Stock Is Up Today