A new FT piece argues there is no AI capex bubble. It is poorly argued, tendentious, reminiscent of pre-housing collapse chatter, and assumes its conclusions. But it is also usefully wrongheaded.
Let's dive in.
What It Argues
Here are the main arguments the piece makes:
- Capacity is Tight
Current AI “bubble” chatters misreads the supply-demand situation: capacity is tight, not overbuilt, so conditions for a classic bursting aren’t present. - Deals Aren't Real Commitments
Many megadeals are mostly options rather than firm capex, so the headline $1.4tn “overbuild” fear is premature; only ~10% or so is actually committed. - Technology, Not Finance, is the Risk
The real risk is in the technology, like hallucinations, inference costs, and uncertain monetisation timelines. It is not asset-stranding or imminent capital flight. - Big Players Benefit
Any correction would be uneven: weakly capitalised players (OpenAI, neoclouds) get hit, while balance-sheet giants (Microsoft, Google) gain share. - The Peak is Far Away
Valuations may fall hard in some names, but the overall trajectory isn’t dot-com-style collapse; chip demand remains strong and the cycle peak is not yet visible.
Why It's Appealing
The draw is its contrarian stance: a calm, clean dismissal of bubble talk at a moment when many are newly forecasting collapse. It highlights genuine optionality in the OpenAI/Nvidia megadeals and makes the overbuild narrative look jumpy rather than grounded. It also leans on the near-term scarcity story — tight capacity, rising chip demand — which collectively gives the argument surface plausibility.