Gary Marcus用FT数据揭了AI公司的底：五年资本支出翻四倍，营收大头被托管服务吃掉，商业模式比想象中更危险，做投资的别错过。
Gary Marcus引用FT报道指出，OpenAI和Anthropic等前沿AI公司资本支出相对收入在五年内翻了两番。Exponential View报告显示，过去一年生成式AI行业营收1100亿美元，但基础模型仅占11%，82%流向托管服务。Marcus认为，尽管AI技术获得赞誉，但可持续商业模式仍未明确，与90年代互联网泡沫相似但风险更大。
when capex quadruples relative to revenue in five years it just can’t be good.
when capex quadruples relative to revenue in five years it just can’t be good. Trevor Noren @trevornoren FT: "It is not yet clear that either [OpenAI or Anthropic] has a sustainable business model. For sure, both have built astonishing AI models that have won ecstatic user acclaim and generated rocketing revenues. But the maths is daunting. The costs of remaining at the frontier of AI are punishing. The penalties for falling behind may be even worse...A new report from Exponential View calculated that the tech sector had generated actual revenues of $110bn from generative AI over the past year, a far faster rate of growth than in any previous tech cycle. But it estimated that foundation models only accounted for about 11 per cent of this revenue in the first quarter of 2026, with 82 per cent going to hosting services." So often in my conversations with clients and prospects I'm asked the dotcom versus AI bubble question and the business model question is core to why I warn it's a poor reference point. Yes, I do believe that excessive enthusiasm has driven a disconnect between priced-in ROI expectations and the likely timeline to actually realizing that ROI. That's a clear similarity. I dissected exactly how I see that equation with AI in my December report on "GenAI & Productivity" ( sageroadresearch.com/collections/re… ). However, Pets.com didn't fail because the founders didn't know their business model and the path to profitability. Amazon, Walmart, and Chewy have reaped huge rewards with Pets.com 's business model. Pets.com was just too early. It remains unclear the hyperscalers can achieve profitability on any conceivable business model, whether it's ad sales, consumer subscriptions, enterprise contracts, or government contracts. Obviously, the hope is that persistent CAPEX will keep market participants from confronting the business model issue as all that spending circulates between the biggest AI stakeholders. That'll give hyperscalers the time needed to define and prove out a viable business model before a "trough of disillusionment". I remain skeptical markets will have that patience and believe all investors should be focused on hedging that risk. Far more thoughts on hedging opportunities in my reports! Learn about Sage Road Research here: sageroadresearch.com . Interested in subscribing? Message me. FT link: ft.com/content/7bff5a… 🔗 View Quoted Tweet 💬 1 🔄 3 ❤️ 18 👀 2435 📊 3 ⚡