OpenAI and Anthropic will soon have to explain to investors why they are each supposedly worth $2tn–$3tn, likely drawing on the template Elon Musk used for the SpaceX IPO in June, where a modestly profitable satellite-launch and telecoms business attached to a tremendously unprofitable AI startup achieved a valuation far above what the company’s financials justify. With private markets showing signs of exhaustion, their expected IPOs promise access to a much larger pool of capital, opening equity markets and the deeper investment-grade debt market, and bankers are already pressing rating agencies to grant them investment-grade marks after their listings. Their pitch must therefore give shareholders a market vast enough to justify today’s valuation while convincing creditors that today’s cash burn can be contained. The frontier labs are relying on two narratives: that AI will capture a meaningful share of corporate wage bills by replacing human workers, recasting software as labour and competing for the wage bill, with “agents” carrying out multi-step tasks under limited human supervision; and that AI poses an “existential risk” to humanity, a claim advanced by Alan Turing in 1951 and more recently by Stuart Russell, Yoshua Bengio and others.