An analyst who called WeWork right values Anthropic at $150 billion against the $2 trillion it seeks in its IPO

By Zendoric · October 8, 2026.

New Constructs, an independent financial research firm, on Tuesday labeled Anthropic's stock market debut "the most ridiculous IPO of 2026". Its own valuation of the company: $150 billion, far short of the $2 trillion valuation Anthropic is seeking in its Nasdaq debut, according to reports gathered by CNBC. To justify that figure, the firm calculates, Anthropic would need to generate twice the net profit that Nvidia — the world's most valuable chipmaker — has accumulated over the last four quarters: some $190 billion.

The underlying figures come, according to Reuters, from a leaked copy of Anthropic's own IPO prospectus (the S-1): $4.6 billion in revenue in 2025 and net losses of $42 billion. New Constructs sums it up as "an unprecedented test of investor credulity" and adds a further warning: Anthropic's own statements about the "catastrophic or existential" risk of AI should, in its view, deter anyone considering buying shares. Anthropic did not respond to CNBC's request for comment.

The firm is run by David Trainer, a veteran Wall Street bear with a mixed track record. He was dead right about WeWork, which he pinned with the same label in 2019: the company, then valued at $47 billion, pulled its public offering six weeks after the report and ended up in bankruptcy in 2023. He was also skeptical of Allbirds in 2021 — a brand that has ended up being sold this year for barely $39 million after reaching $4.1 billion on the day of its debut. But in 2020 he applied the same label to DoorDash, likewise comparing it to WeWork, and the delivery company today trades at a market capitalization of $83 billion, well above the $60 billion it closed its first session at.

The report's underlying argument is not only about valuation: it is about the business model. Growing competition from open-source models, New Constructs maintains, leaves closed AI labs — Anthropic included — without "a viable business". Anthropic argues the opposite with its own data: at the end of July it said its annualized revenue had multiplied sevenfold in a year, to $65 billion, and the New York Times reported in September that the company was on track to close 2026 at $100 billion annualized.

It is worth separating two questions the report blends together. One is whether Anthropic deserves a $2 trillion valuation; the other is whether the technology it sells has a future. They are different things, and at Zendoric we have spent months pointing out that the open frontier — China's Qwen, GLM, DeepSeek and Kimi, but also Western open models — is advancing faster than the dominant narrative admits, which puts pressure on the margins of anyone charging for access to a closed model. On that, New Constructs points to a real problem: if a cheap or free model covers most use cases with a fraction of Claude's capabilities, the question of what margin Anthropic can sustain over the long term is legitimate, and it is not settled by showing a chart of growing revenue.

But Trainer's record is also a forced lesson in humility: he was right about WeWork because the deterioration in corporate governance and the accounts were verifiable before the collapse; he was wrong about DoorDash because he mistook operating losses typical of a growth phase for a business with no future. That Anthropic is multiplying its revenue sevenfold while losing $42 billion proves, by itself, neither fraud nor solidity: it is the phase of massive spending on compute that every frontier lab is going through, from OpenAI to Google. The risk for anyone buying shares is not that Anthropic lacks revenue — it has revenue, and it is growing fast — but that the public market has yet to learn how to tell "growing fast" from "worth what it is asking".

That is, in fact, the real test coming with this offering, beyond Anthropic itself: if Wall Street pays $2 trillion for a company losing $42 billion a year, the market will be certifying that spending in the AI era has no rational valuation ceiling, with the risk of an abrupt correction that implies if the narrative cracks. Our underlying thesis — that generative AI can, over the long term, bring us closer to a real abundance of resources and capacity — does not depend on this particular IPO succeeding, nor on Trainer being right again. But it does depend on the capital financing that race resting on foundations capable of withstanding a down cycle, and that, whether or not the analyst is right, is precisely what this report puts to the test.