Could this be the Frontier Labs’ WeWork moment?

Gary Marcus published a piece asking whether frontier AI labs are approaching a WeWork-style reckoning, arguing their sought-after valuations look harder to justify. He points to several pressures: token spending that peaked around May, Chinese open-weight models driving price wars, questions around OpenAI revenue projections, and a new U.S. appeals court copyright decision he says undermined generative AI companies. He also notes that OpenAI has pushed its IPO back to 2027, while Anthropic is said to be planning an IPO next month but has not publicly released its S-1.
Marcus frames the moment through the lens of potential liability. As poorly controlled agents cause errant, and in his words possibly criminal, incidents, liability becomes a major issue for labs. He quips that LLMs might someday be renamed Large Liability Modules. He also references the so-called rogue AI fiasco, which he characterizes as really just bad software, and notes it had not yet made the news at the time of earlier valuation talk.
His central claim is that against these headwinds it is increasingly difficult to see how the companies merit the enormous valuations they seek. He recalls that in May, at the peak of tokenmaxxing, people discussed OpenAI IPOing at a trillion dollars and Anthropic at two trillion dollars. He does not rule out that happening, but argues it probably should not.
Marcus says he was reading two additional items that morning, jet lagged in New Delhi, and found them even more ominous, without detailing them in the excerpt. The post is partly a reminder about paid subscriptions to his newsletter.
Why it matters: For developers and product teams, the practical signal is that AI platform economics and legal exposure are becoming central planning assumptions, not background noise. If liability and copyright rulings shift costs, teams building on proprietary frontier models may face higher prices or more restrictive terms, while open-weight alternatives could gain appeal — though that remains inference from Marcus's argument, not a reported outcome.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Some ominous signs