AI IPO wave: SpaceX, OpenAI, Anthropic go public

The most revealing detail in SpaceX's prospectus is buried in a single phrase: the Anthropic arrangement 'allows us to monetize unused compute capacity.' ^[1]That unused capacity was not marginal. After xAI shifted Grok training to its newer Colossus 2 facility, the original Colossus 1 cluster — 220,000-plus Nvidia GPUs drawing more than 300 megawatts — had fallen to roughly 11% effective utilization. ^[2]A supercomputer running at one-ninth of capacity is, on a balance sheet, a stranded asset. The compute deals turned it into contracted income. Anthropic agreed to pay $1.25 billion per month — about $15 billion a year — for access to Colossus 1 through May 2029 ^[3], and Google committed $920 million per month for roughly 110,000 GPUs from October 2026 through June 2029. ^[4]Together that is more than $2 billion flowing monthly into SpaceX's datacenter operation, roughly $26 billion annualized. ^[5]The mechanism is clean: idle silicon that was depreciating in the dark suddenly underwrites an 'AI segment' revenue line just as the roadshow needs one. The skeptical read, voiced by analysts tracking the deals, is that the contracts reflect xAI's infrastructure surplus rather than any model strength — SpaceX found buyers for space it could not fill itself. ^[6]

The structural problem critics keep returning to is that the buyers and the owners are the same people. Google pays SpaceX $920 million a month for bridge compute to run Gemini Enterprise — while simultaneously holding a roughly 7% equity stake worth more than $100 billion after the IPO. ^[4][7]Dollars paid as rent partly return as equity appreciation in an asset Google already owns. Commentators have mapped the wider loop: Nvidia funds OpenAI and Anthropic, who pay Google for cloud, who pays SpaceX and xAI for compute, who buy Nvidia chips — a circuit one observer called possibly elegant and another called 'the most elaborate way to launder a balance sheet ever attempted.' ^[7]Whether the demand underneath is real is genuinely contested. IDC's Tim Law rejects the skepticism outright — 'I thoroughly believe the demand is there and will grow' ^[8]— and the numbers behind the renters are not fiction: Anthropic scaled from roughly $1 billion to $35 billion ARR in twelve months, and Google needed bridge capacity because grid connections take 5-10 years and new GPUs up to a year. ^[7]Renting an existing cluster was the only fast option. The reflexivity does not require the demand to be fake; it only requires the same dollars to be counted as revenue, valuation, and equity simultaneously.

Strip away the deal-making and the gap between what these companies earn and what they are priced at is stark. Morningstar's bear analyst pegs SpaceX's fair value at $780 billion — roughly 48% below its private valuation — noting that only Starlink is actually profitable while xAI is projected to burn around $10 billion in 2026. ^[9]OpenAI, last valued near $852 billion, is projecting a $14 billion loss in 2026 and does not expect profitability until around 2029, even with 900 million-plus weekly ChatGPT users. ^[10]Anthropic carries a ~$965 billion valuation after its $65 billion Series H, against a ~$47 billion revenue run-rate as of May 2026 — itself up from about $10 billion a year earlier. ^[11]The Reddit reaction crystallized the disconnect into a single comparison that dominated discussion: Walmart earns roughly 15 times Anthropic's revenue at a similar valuation. Community sentiment skewed skeptical-to-bearish, with a recurring read that three simultaneous filings at trillion-dollar marks look like 'selling the top' before audited margins are forced into daylight. Notre Dame law professor Patrick Corrigan framed the same tension more soberly, comparing the moment to the early internet and asking 'whether the price investors are going to end up paying is going to match up to the substance and fundamentals of what AI is really going to do.' ^[8]

The timing was not coincidental. Anthropic's compute contract surfaced 22 days before SpaceX priced, and the Google contract landed just 6 days before — giving SpaceX roughly $75 billion in contracted future revenue from two marquee customers to underpin its AI-segment valuation right as bankers set the price. ^[5][6]The filings themselves were tightly sequenced: Anthropic on June 1, SpaceX's pricing on June 3, the Google disclosure on June 5, OpenAI's confidential S-1 on June 8, SpaceX's Nasdaq debut on June 12. ^[12]Wedbush's Dan Ives read it as 'an opening of the floodgates for the IPO market, which has been relatively dormant for a few years.' ^[8]But the floodgates open onto a concentrated pool. Bank of America warns the combined fundraising — estimated at $150-200 billion — amounts to a large-scale transfer of accumulated risk from early private investors to public markets, and that the listings push tech's S&P 500 weight past the 48% threshold, surpassing the Roaring Twenties, the Nifty Fifty, 1980s Japan, and the 1990s TMT bubble peaks. ^[13]Adding fragility, both compute contracts carry 90-day termination clauses ^[6], meaning the contracted-revenue base that anchors the SpaceX story can be unwound in a quarter. The question public investors inherit is whether they are buying durable infrastructure or the top of a self-referential cycle.