Writing about AI, fintech, risk and agent commerce. Thinking about what trust looks like when buyers are agents and not people. Advising a small group of teams.
Through the end of 2027 the median European industrial deeptech company raising €15m to €20m still assembles it from three or more instruments. If one European instrument starts writing that size pre-revenue at volume, I am wrong.
Europe has 45% of the world's deeptech startups and 17% of the funding. That usually gets read as European investors being timid, except they are putting a record 32% of the entire European venture pool into deeptech. 🧵
The call: watch order books, not spreads. The next real test is the Apollo paper written against Anthropic's TPUs, roughly $15bn of which becomes tradable early next year. That is the first honest secondary mark anyone gets on AI infrastructure credit.
Last Thursday the market erased $797 billion from the Magnificent Seven after Alphabet and Tesla both guided AI spending higher and both posted negative free cash flow. I think it repriced the wrong number. 🧵
[7] My bet, and the timing could be off: by Q1 2027 at least one of these company-building companies quietly reprices off revenue share onto a flat subscription, once someone runs the math on inference cost per dead business.
[1] A dead Shopify store is free. A dead AI-run company bills you every month. That is the part everyone calling Polsia the next Shopify keeps getting wrong. 🧵