Dragonfly Lens · Live Tracker

The AI IPO Wave

The biggest cluster of mega-cap listings in history just formed. How they actually stack up — with candlesticks, the cycle overlay, and the historically-best way to get an entry.

Figures as of June 2026, with the SpaceX, OpenAI, Anthropic and Cerebras rows corrected 21 September 2026 (see note below). Lockup countdowns + charts update live in your browser. Hover any dotted termLike this — hover any dotted term for a plain-English definition and why it matters. for a plain-English explanation.

How they stack up

Corrections, 21 September 2026. An outside review found four stale or mislabelled cells in the table above, all now fixed. (1) The revenue column was headed ~~“Revenue (2025)”~~ while Anthropic's cell held ~$45B run-rate — a May 2026 figure in a 2025 column, with no caveat. Anthropic's 2025 exit run-rate was about $9B; Sacra put the run-rate near $65B by July 2026. The ~20× multiple was computed off the hybrid. (2) SpaceX read ~~“Debuts Jun 12 2026”~~ → it listed that day and has traded for three months. (3) OpenAI read ~~“Q4 target”~~; Sam Altman told Fortune on 12 September 2026 that a 2026 listing would be “ill-advised” and said “I would say not 2026,” with 2027 now the target. (4) Cerebras showed a ~~~$23B~~ valuation, which was the pre-IPO target; it priced at $185 on 14 May 2026, raised $5.5B, and closed day one at $311 for about $66B. All four are the same failure: a real number kept past the date it described. Corrections log →

Company	Revenue (basis stated per row)	Profitable?	Valuation	P/RevPrice-to-revenue: the company's value divided by a year of sales — how many years of revenue you're paying up front. Higher = pricier / more speculative.	Status
SpaceX SPCX	~$18.7B (+33%) + ~$26B/yr AI-compute signed (Anthropic $1.25B/mo, Google $920M/mo — ramps late 2026)	No (only Starlink profitable)	~$1.8T at listing	~100×	Listed 12 Jun 2026 · $152.05 on 21 Sep
OpenAI	~$25–30B run-rateThe latest month's sales × 12 — an optimistic annual snapshot that can overstate a fast-moving figure.	No — loses ~$1.22 per $1	$850B–$1T	~38×	Filed May 22 · not 2026 — Altman, 12 Sep 2026; now targeting 2027
Anthropic	~$9B 2025 exit run-rate; the ~$45–47B figure we previously showed here was a May 2026 run-rate, and Sacra put it near $65B by July 2026	No (70% inference margin)	~$965B	~107× on 2025; ~15× on the mid-2026 run-rate	Filed Jun 1 · ~Oct target
CoreWeave CRWV	$5.13B (+170%)	No — $1.17B loss	public	~10×	IPO'd Mar 2025
Cerebras CBRS	$510M (+76%)	No — $76M loss*	~$66B day-one close; ~$44B at the 21 Sep close of $208.71. ~$23B was the pre-IPO target.	~129× at day one; ~87× now	Priced 14 May 2026 at $185, raised $5.5B
Quantinuum QNT	$30.9M (+34%)	No — $193M loss	~$13B	~400×	IPO'd Jun 4 2026

* Cerebras shows a GAAP profit, but it's a one-time accounting gain on a contract with its largest customer (G42). Strip it out and it lost ~$76M — on a customer base that is ~86% one region (UAE).

The throughline: the revenue spread is ~1,000× — and not one is profitable all-in. That's the single biggest predictor of how an IPO ages. The nuance that changed in June: SpaceX is now the only one with a dated, contracted path out — ~$26B/yr of signed compute revenue (Anthropic + Google) ramping from late 2026. Signed isn't earned: serving that compute costs billions in GPUs and power, both deals carry 90-day outs, and xAI still burns ~$6.4B/yr. But if the ramp holds, SpaceX could turn all-in profitable in 2027 — which would also start the S&P 500 eligibility clock (the index requires GAAP profitability). Watch the first public earnings, not the press releases.

See it: the candlesticks

Pick any name and watch the actual pattern — the pop, the lockup cliff, the fade. (Already-public names have full history; the just-listed and not-yet-priced ones show little or none.)

Chart:

What history says — the receipts

The actual record of the high-flying IPOs most like this wave:

IPO	IPO'd	Pop → peak	Fell from peak	To bottom
Average	—	—	≈ −74%	≈ 10 months

The average: these gave back ~74% from their peak and took ~10 months to bottom — so the historically better entry was ~10–14 months later, after the lockupA 90–180-day ban on insiders selling after an IPO. When it lifts, a flood of shares can hit the market and push the price down. flood and the first couple of earnings resets. But the legends came back: Amazon survived −96%, Nvidia −90%, Meta −54% — and became giants. The lesson isn't "these are bad," it's that the entry is everything.

The cycle overlay — when, not just what

The biggest disasters didn't crash because of the companies — they crashed because of when they listed. Snowflake, Coinbase & Rivian all peaked at the late-2021 liquidity top, then cratered through 2022's Fed tightening (also a midterm year). Where we sit now rhymes — and the lockups land into a historically friendlier 2027.

2020–21
Liquidity flood & froth peak

2022
Fed tightening · crash · midterm

2023–25
Recovery

2026
Late cycle · midterm · the wave

2027
Pre-election · historically strong

Midterm years average an ~18% intra-year drawdown — the weakest of the 4-year cycle. The pre-election year (2027) is historically the strongest (S&P ~+15% in the year after a midterm) — and that's exactly when these lockups expire. Supply (lockups) vs. demand (cycle + index inclusion) plays out over a year. Cycle patterns are tendencies on a small sample, not laws.

How the shares are split — the overhang

These debuts floatThe slice of shares actually available to trade publicly. Insiders hold the rest — and the bigger that block, the bigger the wave of supply waiting at lockup. only a sliver while insiders keep the rest. The bigger the retained stake, the bigger the supply wall when the ~180-day lockup lifts.

Name	Public float	Insiders can sell	Who holds the rest
SpaceX	~5%	~Dec 2026	Musk ~84% of the vote (10× super-shares); ~95% retained
OpenAI	sliver (TBD)	~180d post-pricing	Microsoft 27% · Foundation 26% · employees 25% · VCs 13%
Anthropic	sliver (TBD)	~180d post-pricing	Amazon 21% · Google 15% · founders/staff/VCs ~64%
Cerebras	~8%	~Nov 2026	founders + VCs + G42-linked
Quantinuum	~18%	~Dec 1 2026	Honeywell + Cambridge Quantum ~82% (Honeywell ~49% vote)

The cap table is the tell: Coinbase came public as a direct listing (no lockup) and fell −92%; Google's fair Dutch-auction price barely dipped. The more concentrated the ownership and the frothier the debut, the harder the round-trip.

Our read — where these could land

A scenario model, not a forecast and not advice. It applies the historical drawdown pattern + each company's multiple, profitability, and share structure. Reality will differ — and we'll track this against what actually happens.

Name	At	The structural tell	Scenario fall
SpaceX	~100×	Starlink profitable + scaling; ~$26B/yr compute signed (Anthropic + Google); Musk premium	mildest ~30–50%
Anthropic	~20×	Best margins + revenue leader; cheapest multiple	~30–55%
OpenAI	~38×	Loses $1.22/$1, being leapfrogged, big overhang	~50–70%
Cerebras	~45×	86% of revenue from one region (G42/UAE)	~50–75%
Quantinuum	~400×	Pre-revenue option; Honeywell ~82% = supply wall	deepest ~60–85%

Lockup-expiry watch

The supply cliff. ~180 days after listing, insider shares unlock — historically a better entry than day one.

The ideal way to play it

Don't buy the day-one pop. You're buying from flippers at peak euphoria.
Circle the lockup date (above). That supply flush is historically a better entry than the prospectus.
Demand profitability — or a credible path. The single biggest predictor of long-run returns.
Let it report 2–3 quarters first. The first earnings prints reset the hype.
If you must own it early, size it like a call option — small. A few become Amazon; most fade.

Every call we make is timestamped and cryptographically signed — so you can verify our record instead of trusting it. This page is entry #1 in that record: as these unlock, we'll mark our scenario read against what actually happened. The model leaderboard changes weekly. The math of an IPO doesn't.

Educational research, not personalized investment advice. Dragonfly Lens is not a registered investment advisor. Past performance does not guarantee future results. Tickers named are examples illustrating a framework, not recommendations to buy or sell. Charts via TradingView. Sources: company SEC filings & earnings; Jay R. Ritter (Univ. of Florida) IPO statistics; a16z; press reporting (June 2026).