Bloomberg published a major analysis on March 18, 2026 asking whether the AI bubble is set to burst, citing stretched infrastructure valuations, slowing venture deal velocity in pure-play AI, and increasing investor pressure for demonstrable ROI.

The valuation bubble in AI infrastructure — propped up by compute bets and model arms races — may well correct. That pattern is consistent with every prior technology cycle. But the productivity gains already flowing through teams that actually deployed AI tools are not going back. The split that matters is not between AI believers and sceptics. It is between two types of organisations: those that bought subscriptions and called it a strategy, and those that redesigned actual workflows, trained their people, and are now compounding efficiency quarter by quarter. When markets demand ROI evidence — and they will — the first group will struggle to justify spend. The second will have the receipts.