Public markets have priced loss-making growth companies before. They have never been asked to price two of them at once when each carries a valuation flirting with the trillion-dollar line, each is burning capital at a frontier pace, and neither has an obvious comparable on the exchange. That is the situation taking shape as Anthropic and OpenAI move, in parallel, toward the public markets — and it is why this is not the usual “IPO window reopens” story.
The mechanics of the listings will be familiar to anyone who has watched a roadshow. The questions underneath them are not. How do you value a company whose product improves unpredictably, whose costs scale with ambition rather than revenue, and whose competitive moat is measured in months? Founders, operators and investors — in the US and in India — should be watching closely, because the answers will set reference points far beyond these two names.
Two labs, one window
According to reporting compiled by U.S. News and Renaissance Capital in mid-2026, Anthropic reportedly filed confidentially for an IPO after reaching a valuation in the region of $965 billion, having raised roughly $65 billion across its funding history — a figure that would make it the most valuable AI startup to approach the public markets. OpenAI, per the same reporting, is preparing its own listing. Prediction markets cited in that coverage put the odds of an Anthropic IPO by September 2026 at around 46%, which is itself a signal: even sophisticated speculators are split on timing, let alone outcome.
The twin-listing setup is what makes the moment unprecedented. Hot sectors have produced clusters of IPOs before, but rarely have two category-defining leaders of comparable scale queued up simultaneously while still posting enormous losses. Adding to the degree of difficulty, the labs would be listing into post-selloff volatility — a market that has already repriced AI exposure at least once and remains jumpy about the gap between narrative and cash flow. A confident bull market forgives a lot; a nervous one interrogates everything.
That timing cuts both ways. A receptive window rewards first movers with premium pricing and a deep order book. A choppy one punishes the second issuer for the first one’s mistakes — and with two mega-deals in flight, sequencing becomes strategy. Whoever lists first effectively runs the experiment the other will learn from.
The pricing problem
Start with the part that looks easy. Both labs are growing revenue at rates that would make any SaaS founder weep with envy, driven by enterprise API consumption, subscription tiers and a land grab across coding, customer support and knowledge work. On a top-line growth basis, these are among the fastest-scaling companies in technology history.
Now the hard part. That growth is paired with burn on a scale the software industry has not seen. Frontier model training is a capital-expenditure business dressed in software-company clothing: the cost is not the marginal API call, it is the multi-billion-dollar compute commitments required to train the next generation of model and to serve inference at scale. These capex and compute obligations are often locked in through multi-year arrangements with cloud and chip suppliers, which means the spending is contractually forward-leaning even when revenue is not yet there to match it. The result is a profit-and-loss statement that pairs spectacular revenue lines with spectacular losses — and a balance sheet whose future is shaped by purchase commitments more than by today’s cash position.
This is where the valuation exercise breaks. Public-market investors price companies against comparables and against a credible path to free cash flow. A frontier AI lab offers neither cleanly:
- No clean comparable. It is not a pure software business (gross margins are dragged by inference compute), not a cloud provider (it rents far more capacity than it owns), and not a chip company. Pick the wrong analog and the multiple is meaningless.
- Costs that scale with ambition, not demand. Traditional businesses cut spending when revenue softens. A lab that stops training the next model to protect margins may simply stop being a frontier lab.
- A moat measured in months. Model leadership is real but perishable. Investors underwriting a decade of dominance are underwriting something the technology’s own history says is fragile.
None of this makes the companies unworthy of high valuations. It makes those valuations unusually dependent on belief — about future model capability, future unit economics, and future pricing power — rather than on the audited backward-looking math that public markets are built to scrutinise. Pricing a frontier lab is, in the honest telling, an act of underwriting a thesis about the next five years of AI.
What the demand test means
Even if you can price these companies, the market still has to buy them. And here the data offers a cautionary note. Renaissance Capital and PwC figures indicate US IPOs raised roughly $34.2 billion through 31 May 2026 — up about 164% year-on-year, a genuinely healthy reopening. But the same analysts question whether that demand can absorb mega-issuers carrying elevated private valuations. A booming IPO market built on mid-cap deals is not the same thing as a market ready to digest two of the largest technology listings ever attempted, back to back.
The absorption question has three moving parts. The first is sheer size: the dollar value of equity these deals would need to place could rival a meaningful share of all US IPO proceeds for the year. The second is supply dynamics around the lock-up. IPOs typically float a small slice of the company; the much larger overhang of insider and early-investor shares becomes sellable when lock-ups expire, usually months after listing. With private valuations this stretched, a wave of post-lock-up supply meeting tired demand is a recognisable recipe for downward pressure — and early backers sitting on enormous paper gains have every incentive to take some chips off the table.
The third part is signalling. A strong debut for either lab would be read as validation that public markets accept the frontier-AI thesis at scale, unlocking capital across the sector. A weak debut — a deal that prices below range, breaks issue, or trades down through the lock-up — would signal something larger than one company’s mispricing. It would tell every late-stage AI investor that the private marks they have been carrying are not corroborated by the deepest pool of capital in the world. That repricing would ripple backward through venture portfolios, down-rounds and all. In that sense, these are not just two IPOs; they are a public referendum on private AI valuations.
The India read
For an India-first audience, the temptation is to treat this as a distant American spectacle. It is not. The pricing and demand tests playing out on US exchanges will shape the climate for India’s own AI and IPO ambitions in concrete ways.
The first lesson is about valuation discipline. India’s public markets have shown they will pay up for growth but punish unprofitable stories that cannot articulate a path to cash flow — recent new-age listings learned this the hard way. Indian AI founders watching the frontier labs should internalise that the same investors who celebrate revenue growth will eventually demand unit economics. Build the narrative, but build the margin story alongside it.
The second is about capital flows. Mega-listings of this size pull global risk capital toward a small number of names. If the twin AI IPOs go well and soak up tens of billions in institutional appetite, that capital is, at the margin, not flowing into emerging-market growth equity. If they go badly and trigger a broad AI repricing, the chill reaches Indian AI startups raising private rounds priced off US comparables. Either way, the cost and availability of capital for India’s AI ecosystem is partly being set on the Nasdaq.
The third is strategic and arguably the most useful. India is unlikely to fund a sub-trillion-dollar frontier lab in the near term, and that is not the game worth playing. The more relevant opportunity sits in the application and infrastructure layers — vertical AI products, tooling, data and the services economy built on top of these models — where capital efficiency is achievable and where Indian companies can list domestically on the strength of actual profits. The frontier labs are demonstrating, in public and at enormous scale, exactly how hard it is to monetise the base layer. Indian founders and investors should read that not as discouragement but as a map of where the defensible, fundable businesses actually are.
The twin window will open, on someone’s timeline. When it does, the prices set on these two labs will be among the most scrutinised numbers in finance — not because the world needs to know what an AI lab is worth, but because, for the first time, the public will get to vote on it.
