SpaceX IPO filing reveals $45B Anthropic compute deal and OpenAI's parallel IPO race

Read the S-1 closely and the headline number gets thinner. The deal is structured as $1.25 billion per month through May 2029 — roughly $15 billion a year, roughly $45 billion in total — but either Anthropic or SpaceX can walk on 90 days' notice ^[1]. That clause turns 'forty-five billion dollars' from a contractual floor into a marketing ceiling: any quarter Anthropic chooses, the run-rate collapses to whatever notice period remains. The number that the prospectus, Bloomberg ^[2], and TechCrunch ^[3]are leading with is, structurally, a one-quarter commitment dressed in three-year clothing.

The pricing itself is louder than the term. Quants in the AI-finance corner of Reddit pointed out that at Nebius retail rates Colossus 1's 300MW should clear something closer to $6 billion a year, not $15 billion. Anthropic is paying roughly a $9 billion annual premium for guaranteed access, no auction, and capacity Musk's own xAI had reportedly run at about 11% utilization before the deal ^[4]. Whether that premium is rational (locking in inference headroom before Q2's projected $10.9B revenue print) or simply convenient (a marquee enterprise contract landing in the same week as the S-1) is the question every prospective SPCX buyer should be asking before pricing.

Three months before signing the contract, Musk publicly called Anthropic 'evil,' 'misanthropic,' and accused it of hating 'Western civilization' ^[5]. The S-1 makes him its landlord. Asked about the reversal, Musk reportedly said his 'evil detector' did not trigger — a tell that the framing was rhetorical positioning, not a policy view, and that Musk's capital-allocation discipline overrides his X.com posture when there is a $15-billion-a-year cheque attached.

The inverse is also true for Anthropic. Tom Brown's confirmation that the company is 'scaling up on GB200 capacity in Colossus 2 throughout June' tells you Anthropic is not hedging; it is moving Claude's near-term growth onto infrastructure controlled by a rival CEO who attacked it on social media one quarter ago ^[6]. The implied calculation is that compute scarcity is now binding enough that buying from your loudest public critic is the cheapest available option. That, more than the dollar figure, is the real disclosure: at this point in the cycle, the frontier labs are price-takers on power, not capital.

The point of folding xAI into SpaceX before the S-1 is multiple arbitrage. A rocket company trades at aerospace comparables; a fourth hyperscaler trades on AWS-style multiples. By absorbing xAI, signing Anthropic as anchor tenant, and disclosing that roughly 60% of 2025 capex went to AI infrastructure, SpaceX is asking public-market investors to underwrite the SPCX float at hyperscaler multiples on what is, in practice, a Starlink-plus-Colossus blend ^[3]. New Street Research's Antoine Chkaiban estimates the Anthropic contract alone produces $3-4 billion in annual SpaceX revenue and more than $2.5 billion in cash profit — fat margins precisely because the underlying GPUs were sitting idle and the marginal cost is mostly power ^[5].

The costume is not free. The same prospectus discloses an AI-segment operating loss of $6.36 billion in 2025, a $4.27 billion Q1 2026 net loss, and an accumulated deficit of $41.3 billion ^[7]. The Anthropic contract papers over the xAI losses with one anchor revenue line; the question is whether the public market gives credit for a hyperscaler narrative built on a single counterparty whose contract is cancellable in 90 days, or whether buyers price SPCX as a still-money-losing aerospace company with a profitable cloud side project bolted on at the eleventh hour.

Buried under the deal mechanics is the structure the S-1 sets up for everyone else. Musk emerges from the IPO with 85.1% of the voting power via a dual-class structure that grants Class B shares 10 votes each, of which he owns 93.6% ^[8]. That is not a poison pill or a near-majority — it is a controlling stake that makes the public float effectively non-voting on every material decision: capital allocation, M&A, the timing of any merger with Tesla or xAI that the @KobeissiLetter post and other commentators on X are already speculating about. Public shareholders take all of the downside exposure with none of the steering wheel.

Layer the environmental risk on top. The S-1 has to disclose the NAACP/Earthjustice Clean Air Act suit over 27 unpermitted methane gas turbines powering Colossus in Southaven, Mississippi, with theoretical penalties topping $1.19 billion in a single year and the DOJ reportedly considering intervention ^[9][10]. The same power source that makes the Anthropic deal possible is now a contingent liability inside the prospectus. Combine the super-voting structure, the cancellable anchor contract, and an active regulatory front, and SPCX is a security whose downside the controlling shareholder can manage discretionarily and whose upside narrative depends on a single customer staying happy. That is a governance posture investors usually only accept from founders with much shorter track records of unilateral decisions.

The SPCX filing did not happen in a vacuum. Two days before it, an Oakland jury needed less than two hours to throw out Musk's suit against OpenAI on statute-of-limitations grounds ^[11]. By the time SpaceX dropped its S-1, the Wall Street Journal was reporting OpenAI would file confidentially within days for a September 2026 debut at up to a $1 trillion market cap, with Goldman Sachs and Morgan Stanley underwriting ^[12]. The timing is not coincidence: clearing the Musk overhang was the last legal obstacle to OpenAI's S-1, and Altman moved immediately.

That gives late-2026 IPO calendars a problem of supply. Between SpaceX targeting around $1.7T at pricing, OpenAI aiming at up to $1T, and Anthropic itself raising a $30B+ primary round at a >$900B pre-money valuation ^[13][14], the three companies could collectively soak up more capital from the AI IPO window than any prior cohort. WallStreetBets traders are already hunting for leverage through pre-IPO vehicles like DXYZ and ARKVX and EchoStar/SATS options exposure; the more nuanced read on Reddit's singularity quants is that the $9B Anthropic-to-SpaceX premium and the simultaneous OpenAI debut together test whether AI public-market demand is structural or whether late-cycle insiders are racing to print exits before the bid thins.