Cost of Intelligence
Cost of Intelligence is the blog where I write about things I find interesting at the intersection of economics and AI. Some posts are short, some go deeper, all of them start with something that made me curious enough to dig in.
Subsidising the Build, Losing on the Bill: Why Europe's AI Catch-Up Targets the Wrong Cost Line
Europe's response to the AI gap is a capex instrument. The race is being lost on opex, where a one-time grant cannot reach a power-cost disadvantage that compounds for a decade.
The Adversarial Stack: When Cross-Model Routing Becomes a Spec, Not a Hedge
Six papers across 2024-2026 converge on the same finding: same-family verification breaks down at scale. The reframe is technical. What it implies for foundation-model margins is not.
The Hyperscaler Capex Stress Test
Part 4 of a four-part series on AI infrastructure financing. Big-4 hyperscaler capex guidance for 2026 has moved past free cash flow. The bond market is filling the gap.
The $250B Footnote: Off-Balance-Sheet Compute Commitments
Part 3 of a four-part series on AI infrastructure financing. A quarter-trillion in compute commitments lives in 10-K footnotes — not on the balance sheet, not in the headline capex number.
Vendor Financing Loops: What 1999 Telecom Tells Us About 2026 AI
Part 2 of a four-part series on AI infrastructure financing. The vendor-financing loops between NVIDIA and its largest customers rhyme uncomfortably with Lucent and the CLECs in 1999.
Stranded GPUs and the SPV Problem
Part 1 of a four-part series on AI infrastructure financing. The SPV structures funding GPU clusters look healthy on paper. Underneath, the residual-value assumptions are not.