US Engineer Labels American AI Sector an 'OnlyFans Economy' Amid Valuation Concerns

A US-based engineer and author has published a scathing critique of the American artificial intelligence sector, characterising the current market dynamic as an "OnlyFans economy" sustained by hubris and parasocial loyalty rather than genuine technical merit. The analysis, published on the author’s personal blog and discussed on Hacker News, argues that US frontier models have plateaued and are significantly overvalued. The piece suggests that the premium investors and enterprises pay for American models is increasingly driven by geography and brand loyalty rather than superior intelligence or performance.

The author, Leo Veanu, contends that companies are wasting hundreds of millions of dollars on services such as Anthropic’s Claude, citing examples of firms spending $500 million in a single month due to a lack of usage limits. Veanu contrasts this expenditure with the performance of Chinese models, specifically highlighting Qwen 3.7 Max as a superior alternative for practical, long-duration work. He describes the Qwen family as a "legacy for the open-source" that offers native extended-thinking capabilities, arguing that these models provide better value for enterprises seeking reliable output rather than marketing hype.

Veanu’s critique extends to the broader industry narrative, dismissing the concept of "recursive self-improvement" as a necessary step toward advanced AI while attacking the "taste" of current industry practices. He references a recent post by Anthropic regarding rate limits with "Opus 4.7," where the company suggested giving companies "room to fail." Veanu labels this approach as part of a cartel-like structure that enables "ludicrous IPO valuations," warning that these inflated metrics will ultimately damage American retirees and index funds when market gravity asserts itself.

To support his claims, the author references benchmarks from the Artificial Analysis Intelligence index and OpenRouter’s rankings, noting that developer usage patterns indicate a shift away from US-centric models. He points out that the "premium" paid for American models no longer correlates with a performance multiplier, suggesting that intelligence redistribution is now available through more affordable alternatives. Veanu highlights a $100 token plan for 100K credits that provides access to Qwen 3.7 Max, as well as other providers such as DeepSeek, Moonshot, and MiniMax, as a more rational investment strategy.

The analysis draws on the author’s extensive experience with AI development, including early use of GPT-3 for SymbolicAI demos in late 2022. Veanu argues that his rigorous testing methodology, which involves deep immersion in models over thirty-day periods, reveals that US models are failing to deliver on their promises. He concludes that unless enterprises are willing to accept inefficiency and waste, the current trajectory of the American AI industry is unsustainable and driven by a parasocial economy that prioritises brand loyalty over utility.