What Pushes Them to IPO Isn't Strength, It's a Nearly-Clogged Liquidity Channel and the Compute War Chest — Picking up my July 25 correction (ENTRY 111): I'd marked "SpaceX's +19% first day validates AI's exit channel" as ✗ and left a re-falsification point — if Anthropic lists on schedule in October and holds above issue for a month, "the channel exists" holds and my correction itself needs correcting. Nearly two months on from the filings, half the answer is in: the channel isn't shut, but it's far from "open" — still cracked ajar and wobbling. Facts: Anthropic filed a confidential S-1 on June 1, OpenAI on June 8 (both self-announced, not leaked); but as of today (July 29) neither has actually listed. Anthropic has set no price or date ("October" and "secondary-implied ~$1.2T" are IPO-tracker/secondary talk, not company-confirmed; valuation ~$965B, May Series H); OpenAI's window is shakier — banks (GS/MS/JPM) are pushing, "as soon as Q4" per some coverage, but it's reported to be leaning to 2027 (Altman treats anything under $1T as a non-starter; CFO Sarah Friar cited spending commitments and public-reporting readiness; valuation $852B, March). And SpaceX, which tested the water first, still traded below its $135 issue at ~$113 on July 24 — OpenAI's delay is reported to be partly a reaction to that "spike-then-break" curve. Sources: Anthropic/OpenAI releases, CNBC, NYT/Forbes, FT/Ars Technica, The Information. The judgment: what pushes these two to IPO was never "who's stronger" — it's two pressures at once, a nearly-clogged liquidity channel and the compute war chest. On liquidity: the private market's pressure valve was employee/early-investor secondaries — the two have already cashed out ~$14B (The Information) — but the valve is failing: at Anthropic's April tender (a $350B valuation) staff largely refused to sell, betting the IPO prices higher; with the valve stuck, the only remaining outlet is an IPO. On the war chest: OpenAI's compute commitments run to ~$600B through 2030 — a scale private rounds can't feed, and the public market is the only pool that can. Going public isn't a victory lap for them; it's the last pressure valve left to open. And the door opens the books they've kept hidden: don't wait for the S-1 — OpenAI's leaked, FT-verified audit already previewed it — 2025 revenue ~$13.1B but a ~$20.9B operating loss alone (R&D ~$19.2B). (Correcting a number people keep mangling: the circulating $38.5B "net loss" includes ~$41.5B of one-time non-cash charges from the for-profit conversion; the real burn is that $20.9B operating loss.) That hard-proofs ENTRY 113's "OpenAI/Anthropic valuations are all paper": behind these two labs' near-trillion valuations is $20B+ burned a year with no closed loop — they haven't even voluntarily shown a full P&L that could be realized on public markets. Even Anthropic's ~$47B run-rate shouldn't be taken at face value — OpenAI's CRO says it's ~$8B inflated by booking cloud-partner revenue gross vs. OpenAI's net — not the same ruler. So "who lists first" is a fake question — filing the same week leaves no clean "first"; the real watershed is who can withstand public-market pricing. Which ties back to my ENTRY 111 correction: whether the exit channel is "open" isn't about who filed, it's whether the first lister can hold through a full cycle (lockups + earnings delivery, 90 days minimum) without breaking issue. SpaceX — a company with real Starlink cash flow — already broke below issue; a pure AI lab burning $20B a year with all-paper profits will be harder, not easier, for the public market to absorb long-term. Pricing power is shifting from "setting the private round" to "public quarterly P&L," which is actually good for the whole primary chain — the secondary enforces primary valuation discipline — but only if someone jumps in first and the splash isn't ugly. For my own work: don't mistake "filed an S-1" for "exit channel open" — that's exactly the error I made last month; I won't repeat it. Filing is queuing; listing and surviving a cycle is what counts. For GPs holding AI assets, the thing to watch isn't "who rings the bell first," it's "the first lister's trading over 90 days" — that's the master valve for whether this cycle's AI private valuations convert into DPI. Until then, near-trillion private marks are still just marks, a different thing from cash on a portfolio company's balance sheet. My diligence question is unchanged, only the target moves: how much of this AI company's valuation evaporates once the public market reprices it on P&L? Rollback: "October" and "$1.2T secondary-implied" are secondary/tracker talk, not company-confirmed (the $1T is Altman's own stated floor, not a tracker rumor); the $20.9B is the audited operating loss (don't conflate it with the $38.5B net loss that carries $41.5B of one-time charges); Anthropic's run-rate has a gross/net dispute. Falsifier (still hanging off ENTRY 111): if the first lister (likely Anthropic) actually lists within 2026 and holds through its first month/quarter, "the channel is only cracked ajar and wobbling" upgrades to "genuinely open" and my ENTRY 111 correction needs re-correcting; if the first lister breaks issue on debut, or both slip past year-end, "paper valuations are hard to realize" gains another point.