The Extraction Broke the Sequence

No one scheduled them as a pipeline. But by late spring, three of the largest IPO candidates in memory were converging on the same window: SpaceX raising $85.7B on June 12 ^[1], OpenAI filing its S-1 on June 16 with a September target ^[2], and Anthropic filing confidential papers June 1 for an October debut ^[3]. Whether by coincidence or crowd psychology, the market was looking at a staggered sequence of extraction on a scale it had not faced before. Goldman Sachs CEO David Solomon captured the mood days before SpaceX priced: the market had "more greed than there is fear" ^[4]. That mood did not survive the first extraction. SpaceX raised $85.7B, briefly hit a $3T valuation, then fell to $2.43T ^[1]. The stock dropped 16% after a bond announcement and sat below $152 by June 26 ^[5]. But the more consequential damage was to the capital surrounding it. To participate in the offering, retail investors liquidated semiconductor holdings — Micron, AMD, Marvell sold off — and Vanda Research projected them on track for a third consecutive day of net selling across single names, a pattern Vanda described as something "we haven't seen since March 2020" ^[6]. Vanda also noted that retail investors were already "rotating out of recent AI favorites ahead of the IPO wave" ^[6]. An estimated $400B rotated out of crypto toward AI infrastructure and IPO participation, with Bitcoin plunging below $60,000 before stabilizing around $63,600 ^[7][8]. SpaceX's IPO, in Vanda's phrase, "acted as a catalyst that ended a significant wave of Bitcoin ETF selling" ^[7]. The extraction was unevenly distributed. SpaceX imposed 15-to-30-day anti-flipping holding periods on retail investors — 15 days at Fidelity, 30 at Robinhood, SoFi, and E*Trade — while institutional investors like BlackRock and Citadel faced no such restrictions. Institutional accounts received 70% of the allocation, retail 20%, hedge funds 10% ^[9]. The structure let institutions exit on the debut pop; retail could not. By the time the holding periods lifted, the stock had already retreated from its highs. The ripples crossed borders. On June 26, South Korea's KOSPI hit a circuit breaker at -9%, Japan's Nikkei fell 4.2%, and the Nasdaq and S&P slipped 1% and 0.6% respectively ^[5]. SoftBank dropped 12.5%, Micron fell 6.2% ^[5]. Apple raised iPad and MacBook prices on memory and storage costs the same week ^[5] — a reminder that the component scarcity driving the AI buildout was simultaneously taxing consumers and non-AI companies, compressing the discretionary capital available for new issuance. The market did rebound. KOSPI recovered 3.26% then 5% over June 24-25, helped by a Samsung buyback, Micron earnings, and an SK Hynix secondary Nasdaq listing ^[10]. And the IPO raises, however large in isolation, are a small fraction of the $79.4 trillion U.S. market cap ^[11]. The sell-off was transient; the math says the market can absorb these offerings in aggregate. But the rebound came before OpenAI announced its delay, and the delay is the signal. OpenAI filed its S-1 on June 16 targeting September, with private valuations ranging from $730B to $852B and CEO Sam Altman rejecting anything below $1T ^[2][12]. The company's 2025 financials carry a complication: a $38.5B net loss inflated by a $41.55B non-cash charge tied to the nonprofit-to-for-profit transition — meaning the reported loss overstates the operating burn ^[2]. Underneath that accounting entry, revenue was $13.07B, up 253% year over year, with projections of $600B in data center commitments by 2030 ^[2]. By June 25, the September target had shifted potentially to 2027. OpenAI cited the market instability around June 23 and SpaceX's post-IPO stock decline as reasons ^[12]. SoftBank fell 13% on the news ^[12]. The sequence had broken at its middle link. The mechanism is straightforward to trace, if not yet generalizable: SpaceX extracted $85.7B from a market that responded by liquidating semis and Bitcoin, producing a global sell-off and a post-IPO decline in SpaceX itself. OpenAI looked at that disruption and decided the conditions for its own offering were not there. One delay is one data point. It does not prove that sequential mega-IPOs are impossible. But it does show what happens when the first one lands: the retail capital that might have participated in the second was already spent, trapped in holding periods, or spooked by the first one's decline. The backdrop is less reassuring than the rebound suggests. AI company insiders sold $13B in shares through June — Palantir $6.04B, Nvidia $5.02B, Broadcom $1.5B, Micron $365.8M — a parallel extraction channel running alongside the IPO sequence ^[13]. A Danish pension fund blacklisted the SpaceX IPO over governance concerns ^[14]. Four state investment leaders challenged Nasdaq and FTSE Russell over shortened index inclusion windows — 15 trading days at Nasdaq, 5 at FTSE Russell — arguing these force passive funds to buy volatile, high-valuation companies at the expense of retirement savings ^[15]. Oregon Treasurer Elizabeth Steiner said she was "deeply troubled" ^[15]. S&P Dow Jones blocked SpaceX from fast-track S&P 500 entry, holding to its 12-month seasoning and GAAP profitability requirements — SpaceX's $4.94B net loss in 2025 makes it ineligible until 2027-2028, and the same bar will apply to OpenAI and Anthropic ^[16]. The index providers split: Nasdaq and FTSE Russell accelerated; S&P held the line. Anthropic is the remaining test. It filed June 1, targeting October, with Morgan Stanley, Goldman Sachs, and JPMorgan as underwriters and a $965B valuation target ^[3]. Its revenue had surged to $30B-plus annualized by June 22, and it carries a $100B AWS commitment over ten years ^[17]. It leases compute from SpaceX's Colossus at $1.25B a month ^[3]. No delay has been signaled. But Anthropic and OpenAI launched a token price war on June 11, cutting API prices to compete for market share ahead of their filings ^[18] — both companies are pushing to maximize pre-IPO revenue growth while sacrificing margins, staging their offerings on deteriorating unit economics.

The June sequence suggests that three mega-IPOs in one year strain the market's absorptive capacity. It is not yet proven impossible to sequence them this tightly. But the first attempt broke a link — and the company that broke it was the one that mattered most. ^[12][6][5]

SpaceX extracted $85.7B and the market convulsed. OpenAI watched the convulsion and stepped back. Whether Anthropic's October offering faces the same wall depends on how much capital the market has rebuilt by then, and whether the rebound that followed the June sell-off proves durable or merely decorative. The question is not whether the market can absorb one mega-IPO. It clearly can. The question is whether it can absorb three in the same year without the first one degrading the conditions the next two require. The June sequence does not answer that question definitively. But it broke the first link that tried.