The investor literature presents a genuinely uncomfortable confrontation between two internally coherent but mutually incompatible narratives. The bubble thesis is stark: $85 billion in cumulative AI infrastructure investment at Google has generated $21 billion in annualized revenue at thin margins, while capex now exceeds net profit at a company historically celebrated for capital-light economics. The implied demand required to justify industry-wide infrastructure spending runs to something like 333 million paying users — roughly the entire US population, against 20 million paying users as of late 2025. The telecoms parallel is explicit and historically specific.
The counter-narrative is that the bubble framing is structurally blind to demand-side dynamics. DeepSeek V4-Flash pricing at $0.28 per million output tokens — two orders of magnitude below frontier pricing — confirms that inference costs are collapsing even as model capability advances. The Jevons effect is real: cheaper intelligence expands the frontier of tasks worth attempting, meaning usage growth may substantially exceed what any static projection of current demand would suggest. The compute efficiency trajectory is the bull case: a model achieving 80.9% image recognition accuracy in 2021 required 16,500 times less compute than one achieving the same result in 2012. The cost curve is bending the right way.
What the investor articles do not resolve — and what remains the most consequential open question for anyone allocating capital — is the scaling law uncertainty. The relationship between training compute and model performance is highly uncertain and rapidly changing. If scaling continues, the infrastructure investment is defensible on a long-horizon GPT basis. If it does not, the capex cycle accelerates toward commodity economics before the revenue base materializes.
Europe’s structural position makes this worse: European AI investment records are being set while more than half the capital funding Europe’s best companies originates abroad, meaning the returns from European innovation are increasingly flowing to American limited partners. We know where the value is being created. The question of who captures it is, for European investors, still very much open.