The consensus that artificial intelligence is disinflationary rests on a category error. In the long run, cheaper cognition should expand supply. In the transition — the phase markets and central banks are actually living through — AI is a capital-hungry, debt-financed, input-price-raising investment cycle that pushes against disinflation and against the natural rate of interest. The question for policy is not whether AI eventually lowers costs. It is whether the Federal Reserve and its peers can hold nominal rates high enough, long enough, to accommodate a neutral rate that may already be drifting upward.
The scale of the capital commitment is the first fact to internalise. Google, OpenAI, Anthropic, Meta, Amazon and Oracle committed roughly 300 billion dollars to capital investment across semiconductor supply chains, power grids and specialised labour in 2025, according to Liberty Street Economics. That spending has now outrun operating cash flow: the same firms raised over 100 billion dollars of new debt in late 2025 once capex began exceeding what the businesses generate internally. This is no longer an equity-funded moonshot. It is a leveraged build-out competing with the real economy for scarce inputs — power, chips, engineers, transformers — and bidding their prices up in the process.
This is what Liberty Street Economics calls the productivity J-curve: diffusion temporarily raises production costs during adoption even as the technological frontier expands. The disinflationary payoff is real but back-loaded. The inflationary impulse — construction bottlenecks, power-price pressure, reorganisation spending, wage competition for a narrow talent pool — is present tense. Whether AI raises productivity faster than adoption costs, not productivity gains alone, determines its net effect on inflation. Framed that way, the near-term signal for the price level is ambiguous at best and pro-inflationary at worst.
The neutral rate is moving
The second-order consequence sits in r-star (the natural real interest rate — the level at which policy neither stimulates nor restrains). Lukasz Rachel argues AI could lift r-star by around one percentage point, because faster expected growth and higher future incomes raise the demand for capital today. The Economist puts the same logic more starkly: even under benign superintelligence, rates must rise steeply to coax people into saving rather than spending against expected future income. Simone Lenzu draws the useful distinction — a one-time level shift in productive capacity raises r-star temporarily; a sustained acceleration raises it permanently. The dossier is unusually one-sided on direction: every named forecaster expects r-star to move up, and AI capex to keep rising through 2026 and into 2027. The disagreement is about duration and magnitude, not sign.
the bigger the hype, the more rates may need to rise to prevent overheating
Warsh's view is the coherent dovish counter, and it will find sympathisers on the FOMC. But it requires believing the supply response arrives before the demand and financing pressures work through the price level — a sequencing bet the current capex-versus-cash-flow gap makes hard to defend. The operationalisable claims to attach to this cluster are narrow but sharp. Rachel's call — r-star up by at least one percentage point — is the highest-conviction forecast in the dossier at caliber 70. Liberty Street and Lenzu both expect AI capex to keep climbing through 2026 and into 2027, funded increasingly by credit markets. If both are right, the financial-stability tail — leveraged concentration in a handful of hyperscalers, exposed to a growth path no one can yet calibrate — becomes the story of 2026, not the productivity miracle.
Briefings are synthesised by the Ledger Desk from multiple sources cited in the sidebar. They are distinct from Articles, which are written by named contributors and carry a tracked Calibration Index. The Desk does not currently carry a Brier score; this is a deliberate choice for the v0.1 editorial layer and will be revisited.