Why AI is not a bubble* (www.derekthompson.org)

Despite a chorus of investors and CEOs calling AI “a bubble,” the reality looks more complicated: capital and corporate behavior suggest people are still racing to build and deploy AI at scale. High-profile warnings (Jeff Bezos, OpenAI’s Bret Taylor, JPMorgan and Goldman executives) contrast with massive commitments — Amazon’s reported $100B AI infrastructure plan, OpenAI-triggered cloud deals and AI companies driving a large share of 2025 US stock gains — which imply both enormous demand and speculative fever. The piece lays out five concrete reasons for skepticism and also why this might not be a classic bubble: runaway valuations for preproduct startups (e.g., a $2B seed round), unprecedented private-sector capex (estimates of $400B in AI data-center spend vs. ~$60B AI revenue in 2025, a 6–7x gap), risky accounting and depreciation practices (chips replaced every 2–3 years but amortized over five), off‑balance‑sheet vehicles, and deep corporate entanglements (Amazon–Anthropic, Microsoft–OpenAI, Nvidia equity deals). Technically and economically, that means huge infrastructure load (examples cite multi‑GW power needs), lumpy capital risk, and opaque revenue recognition — so the AI/ML community should expect both real, sustained platform buildout and acute financial fragility unless transparency, realistic accounting and durable product-market fit catch up with the hype.