AI infrastructure may not add up without trillion-dollar revenue

AI infrastructure may not add up.

The conversation is increasingly shifting from growth to payback.

I was struck by David Cahn’s Sequoia estimate: by 2026, AI infrastructure spending could reach $1.5 trillion, while the industry may need about $3 trillion in revenue to justify that investment.

At the same time, several risks are building:
— companies are opting for cheaper open-weight models;
— token prices are falling;
— new models are more efficient, but that does not necessarily mean faster revenue growth for infrastructure players.

Why it matters: if hyperscalers fail to deliver the expected cash flow, the pressure could extend far beyond the AI market.

How do you see it: a temporary imbalance, or a sign of overheating?