Analysis
I don't think the "Wall Street vs. Silicon Valley" framing on AI spending is really a disagreement about whether AI is valuable -- it's a disagreement about time horizon, and this week made that split impossible to paper over. The Information reported Wednesday that stock-market angst has already delayed several mid-sized IPOs and is now cooling sentiment around Anthropic's own giant IPO, which keeps slipping later into the year than investors expected.
Silicon Valley's model is: spend now on compute, capture the market, worry about unit economics once you have pricing power. That's the model behind Anthropic's roughly $965 billion post-money valuation on its last round, OpenAI reportedly in talks at a $1.4 trillion mark, and every hyperscaler capex number getting bigger every quarter. Wall Street's model runs on a different clock: public-market investors need to underwrite a return within a normal fund or pension timeline, and they're the ones who have to explain a multi-year gap between spending and revenue to their own limited partners and retirees every single quarter, not once a decade.
“Silicon Valley's model is: spend now on compute, capture the market, worry about unit economics once you have pricing power.”
That mismatch is exactly why Oura pulled its IPO, why multiple mid-sized listings have slipped, and why Bain is now putting a number on the gap: as Pulse covered today, the firm estimates the AI industry needs $6 trillion in annual revenue by 2031 against maybe $1.8 trillion current products can plausibly deliver. Private markets can underwrite that gap on faith and momentum. Public markets, by construction, can't -- they have to price it today, every day the stock trades.
What I'd actually watch is Anthropic's S-1, once it's public: not the valuation number, but the delta between disclosed infrastructure commitments and disclosed revenue run-rate. That single number is the one both Wall Street and Silicon Valley are implicitly arguing about right now without either side actually putting it on the table.
Room for disagreement: it's entirely possible Wall Street is simply wrong here, the same way public-market skeptics were wrong about Amazon's spending in 1999 and wrong about cloud infrastructure capex for most of the 2010s. Compute spend that looks irrational against today's product revenue can look cheap in retrospect if the new product categories Bain says are missing actually show up on schedule. The delayed IPOs prove investors are nervous right now -- they don't prove the underlying spending thesis is wrong.