Something strange is happening at the intersection of finance and prophecy. Ramp just raised $750M at a $44B valuation, nearly tripling in a year, on the strength of an AI story layered over a fintech foundation. Meanwhile, Anthropic, OpenAI, and their cohort barrel toward Wall Street debuts at valuations that would make even the dot-com faithful blush. The market is casting spells, not spreadsheets.
Risk Pricing in the AI Investment Frenzy
The timing is uncomfortable. A 2026 Delphi study in arXiv by Saeri et al., surveying 272 international AI experts, found that consensus among researchers places near-term economic disruption and misuse risks far above the existential scenarios that dominate headlines. In other words, the risks most likely to actually bite investors are the ones least baked into valuations. Ramp's pitch is efficiency and intelligence. The study's finding is that efficiency tools deployed without governance frameworks are a category of risk unto themselves. The gap between those two sentences is where valuations live. TurboFund's breakdown of investor research mistakes flags exactly this pattern: founders who lead with AI narrative without addressing governance get funded fast and scrutinized slow.
When the IPO Window Opens on an Unaudited Industry
The Dow's rotation out of chipmakers after Broadcom's underwhelming outlook is a preview of what happens when the AI story hits a moment of friction. TSMC's own admission that it can only support so much demand introduces supply-side reality into a market that has been pricing for infinite scale. The Delphi paper's most pointed finding is that disagreement among experts, not consensus, is where the real signal lives. A companion arXiv paper by Wawer and Chudziak argues that reasoning-trace disagreement in multi-agent systems is itself a knowledge-representation signal. Apply that to VC: the investors who are quietly hedging, not the ones beating the drum, are probably seeing something the market isn't pricing. The spell only works until it doesn't.