A call to halt Anthropic's IPO: ~$895 billion in commitments and a debate over transparency

By Tech Policy Press · November 2, 2026.

Matt Scherer, a researcher at the Open Markets Institute and co-author of the report “Brace for Impact,” argues in an op-ed that the SEC (Securities and Exchange Commission, the U.S. securities regulator) should block Anthropic's stock market listing. It is the author's position, not an official decision or investigation.

According to Scherer, the IPO (initial public offering) filing reveals in its notes some $895 billion in spending commitments, most of them stretching into the 2030s. Of that, at least $582 billion would be non-cancelable commitments with five suppliers: Amazon, AMD, Broadcom, Google and Microsoft. Against that, Anthropic has $96 billion in cash, according to the same piece.

The author adds two criticisms of the financial disclosures. First, the 2026 results are unaudited, and by his reading of the third-quarter figures, revenue growth slowed over the summer. Second, those commitments do not appear on the balance sheet, although he considers them debt in practice. He calls on the SEC to require two consecutive quarters of audited accounts and full disclosure before allowing the debut. These are the author's claims; the article does not include a response from Anthropic.

His economic argument rests on four pressures on margins: uncertain demand (he cites a study according to which AI use at work among Americans fell from 46% to 33% between the summers of 2025 and 2026), commoditization (models of similar quality and a price war), high costs (he estimates $31 billion per gigawatt of data center capacity; Anthropic has reportedly committed to 15 GW and estimates it needs 20-25) and external constraints such as local opposition to data centers and the supply chain.

Our take. The piece mixes data verifiable in a prospectus with projections and adjectives, and the two should be separated. That there are compute purchase commitments worth hundreds of billions is a matter of record that an investor must be able to assess; there, the call for transparency is reasonable. That the business model is “deadly” or that the IPO will lead to a collapse is, by contrast, a prediction, and the description of Fable 5's commercial reception as “lukewarm” is the author's assessment. More broadly, AI becoming cheap and commoditized is bad news for a lab's margins, but good news for society.

That is the interesting point for our thesis. If models get cheaper due to competition, Chinese open-weight models and local AI, abundance arrives sooner, though not necessarily at the expense of whoever financed the infrastructure. In the short term, the risk is real and resembles other infrastructure bubbles: capacity built with debt or fixed contracts that depends on demand that has yet to be proven. Who ends up bearing that risk, retail shareholders included, is a legitimate public policy question.

Our view: neither a block nor a blank check. The sensible course is to demand audits and clarity on off-balance-sheet commitments, and let the market set the price with that information. The industry's compute ambition is what could bring us closer to curing diseases and to long-term abundance; financing it transparently is the best way to ensure that any adjustment, if it comes, is a correction and not a crisis.