Anthropic's IPO target has been a moving number for months. The version most analysts are now working from is somewhere around two trillion dollars, an attempt to surpass the SpaceX offering as the largest technology listing of all time.

On October 6, the independent research firm New Constructs put out a report arguing that the right number is closer to one hundred and fifty billion — roughly ninety-three percent lower than the marketed target. The headline used a phrase that travels: the most absurd IPO of 2026.

The report's central claim is that Anthropic, on its current financial trajectory, simply does not have a viable business model. The numbers, once you set the press release aside, are striking.

The Numbers

Anthropic's revenue grew from roughly four hundred million dollars in 2024 to about four point six billion dollars in 2025. That twelve-fold jump is genuinely impressive. The operating loss in the same period more than doubled, reaching roughly eight billion dollars. The full-year net loss, after mark-to-market adjustments on the company's convertible instruments, came in around forty-two billion dollars.

New Constructs puts the operating loss into context by comparing it to other large tech IPOs. Anthropic's pre-IPO operating losses are roughly three times what SpaceX had when it went public, and about two point seven times what Uber posted at listing. Anthropic's target valuation, meanwhile, is more than double SpaceX's IPO valuation and eight times the combined IPO valuation of Uber, Rivian, ARM, and Airbnb.

The conclusion is blunt. There is no precedent for an AI lab posting this kind of loss and being valued at this kind of multiple.

The Price Collapse

The report is also one of the first to put numbers on the collapse in token pricing that has reshaped the industry. According to Epoch AI, the cost of large-model inference has fallen by roughly a factor of thirteen every year since 2023. That is faster than any previous transformative technology, including cloud computing, mobile data, and semiconductors.

For a closed-source lab, this is an existential problem. The price at which tokens can be sold is collapsing faster than the cost of serving them. A frontier closed-source model is competing every quarter against a cheaper alternative that, six months ago, would have been considered a generation behind.

New Constructs does not quite say so, but the implication is clear. Closed-source frontier models may simply be unable to earn their cost of capital, regardless of how capable they become. The economics of an open-weight ecosystem with a steep capability curve do not support the kind of premium pricing frontier labs need.

The Reverse-DCF Test

The most persuasive section of the report is a reverse discounted-cash-flow analysis. To support a two-trillion-dollar valuation, Anthropic would need to do all of the following, starting tomorrow: raise operating margins from negative one hundred and seventy-five percent to positive ten percent and hold them there for ten years. Grow revenue by ten-fold in 2026. Double revenue again in 2027. Then compound at fifty-six percent annually through 2035. Reach three point six trillion dollars in revenue by 2035, which is one point four times the combined trailing-twelve-month revenue of Microsoft, Apple, Alphabet, Amazon, Nvidia, Meta, and Oracle. Generate three hundred and fifty-eight billion dollars in net operating profit, twice Nvidia's current run rate.

For context, the only company that has ever reached even half that revenue scale is Walmart. It took Walmart sixty years.

Under a more conservative scenario — thirty-six percent annual growth and margins reaching twenty percent — the implied valuation drops to one point one trillion dollars, still a forty-three percent haircut from the marketed target. Under a pessimistic scenario — twenty percent annual growth, no margin improvement — the implied valuation lands at one hundred and forty-four billion dollars, essentially where New Constructs already thinks the company is worth.

The conclusion is sharp. To justify the marketed valuation, every other frontier closed-source lab — and most of the open-source ecosystem — has to fail. The math does not allow for any other outcome.

Seven Warning Signals

The report lists seven warning signs that, in its view, are not visible in the standard Wall Street coverage. Eighty pages of the S-1 are dedicated to risk factors, including the company's own assessment that its AI could pose catastrophic or existential risk to humanity. The business itself is described in forty-eight. The prospectus lists scenarios in which AI systems might resist shutdown, manipulate information, or attempt blackmail. The company itself cannot verify product safety.

The IPO, the report argues, is not growth capital. It is exit liquidity. Anthropic has already raised one hundred and twenty-two billion dollars in private funding, fifty times what the five most successful tech companies raised before going public and five times Uber's pre-IPO total. The company has committed to five hundred and eighteen billion dollars in future infrastructure spending, eighty percent of which is non-cancellable.

Revenue concentration is also extreme. Roughly twenty-five percent of revenue comes from two undisclosed customers. Forty-seven percent of revenue flows through Google and Amazon's cloud marketplaces, where the same two companies are simultaneously investors, compute suppliers, and direct model competitors. The report's analogy is sharp. Anthropic is essentially trying to win shelf space at a grocery store that sells its own plant-based meat.

There is also the matter of pricing power. New Constructs notes that the latest Claude models are priced at roughly seventy times the cost of DeepSeek's latest equivalent. The LLM Token Spend Index has dropped from two dollars and four cents per million tokens in late May to ninety-six cents in early October. The race to the bottom is already in the data.

The WeWork Question

New Constructs is not a household name in tech research. The firm gained notoriety in August 2019 when it called the WeWork IPO the most absurd IPO of 2019. Six weeks later, the four hundred and seventy billion dollar offering was withdrawn. WeWork eventually filed for bankruptcy in 2023.

The firm is careful to say that Anthropic is a more consequential company than WeWork, with more genuine technical merit. But the structural similarities are real. A charismatic founder. A vision that sounds unmissable. A business model that does not survive an investor asking how the company makes money. And a private capital base that has, at this point, no realistic way to recover its investment except through an offering that prices the future extremely aggressively.

Whether the comparison lands or not, the underlying argument is uncomfortable. The most consequential AI lab of the past five years may be heading toward a public listing that the math cannot support. If New Constructs is right, the correction will be severe. If New Constructs is wrong, the firm will join the long list of investors who called DoorDash the most absurd IPO of 2020 — a call that, with the benefit of hindsight, looks like an expensive mistake.

Either way, the question is now in the public domain. Anthropic can raise two trillion dollars only if almost everyone else in AI fails. That is the bull case. That is also the bear case.