Everything About AI
NextFin News - Artificial intelligence is doing two opposite things to inflation at the same time, and central banks are being forced to admit it. The Federal Reserve cut its policy rate by three-quarters of a percentage point across its final three meetings of 2025, then held at 3.50%–3.75% as policymakers weighed a disinflationary productivity boost against a fresh wave of AI-driven price pressure. The latest data show why the pause has teeth: core PCE inflation ran at 3.0% year over year in August 2026, with a 0.2% monthly increase, while Fed officials in September lifted their median core-inflation forecast for the year to 3.4% and their median preferred policy rate for year-end to 4.1% — above the current target range. The age of easy disinflation is giving way to something harder to tame.
The question now facing the Fed, the European Central Bank, and their peers is whether AI is a structural cure for the low-growth, low-productivity world that followed the global financial crisis, or a cyclical inflation shock dressed in technological clothing. The answer determines whether the rate cuts of 2025 were the start of a new easing cycle or a policy error that will have to be reversed.
The Two-Faced Shock: Productivity Gains Meet a Capex Boom
On the disinflationary side, the data are real and they are recent. Total factor productivity in the US private nonfarm business sector rose 0.8% in 2025, with output up 2.6% while combined inputs grew only 1.7%. In the second quarter of 2026, nonfarm labor productivity increased 1.4% at a seasonally adjusted annual rate and was up 2.2% from a year earlier. That keeps productivity growth running at a 2.1% annualized pace through mid-2026 — above the 1.5% rate of the previous business cycle from late 2007 through 2019 and matching the long-term average of 2.1%.
The Federal Reserve Bank of Kansas City has traced the pickup toward the industries most exposed to generative AI, noting that labor productivity has remained above its pre-pandemic trend since late 2022 — the same window in which generative AI tools moved from research labs into commercial use. The timing is suggestive rather than definitive; the Fed's own researchers report a correlation with an R-squared of 0.19, and causality can run both ways. But the direction is hard to dispute.
Against that stands the inflationary leg. Electricity prices rose 6.9% year over year through December 2025, more than double the 2.9% headline PCE reading at the time. Goldman Sachs Research expects consumer electricity inflation to hold near 6% through 2026–2027 before decelerating to roughly 3.5% in 2028, and warns that if non-AI customers bear half of the incremental cost of data-center capital spending — rather than the one-third assumed in its baseline — electricity prices would rise about 8% on average in 2026–2027. The transmission is already visible in wholesale power markets: PJM's capacity clearing price for the 2026–2027 delivery year reached $329.17 per megawatt, more than ten times the $28.92 level of 2024–2025, with rapid data-center growth identified as a major contributing factor.
Inside the Fed, the split is no longer theoretical. The minutes of the September 15–16, 2026 meeting recorded that "surging AI-related investments were contributing to inflation pressures," that the effects of the AI buildout on core goods prices "appeared to increase," and that "certain items in the PCE price index—software and portfolio management fees in particular—had made relatively large contributions to recent PCE inflation readings." Several participants raised their estimate of the neutral federal funds rate, several described the current policy rate as "not restrictive or only mildly restrictive," and a few argued that a higher policy rate would help prevent AI-driven demand from "broadening out and generating more persistent inflation dynamics."
AI is lifting the economy's speed limit at the same time it is pressing on the accelerator.
Why the Disinflationary Payoff Arrives Late — and the Inflationary Hit Arrives Early
The asymmetry of timing is the crux of the central-bank problem. The costs of AI arrive upfront and in hard currency: chip orders, data-center construction, power purchase agreements, grid upgrades. The benefits arrive later, diffusely, and only if the technology actually changes how work gets done rather than merely shifting tasks around.
Isabel Schnabel of the ECB put the point precisely in a March 2026 speech at the US Monetary Policy Forum: "Price and wage rigidities imply that even if cost savings from new technologies were to arise immediately, the associated disinflationary impulses would come with a measurable lag." She also drew a distinction that most market commentary blurs. If AI is labor-augmenting, it raises the productivity of remaining workers and lifts potential output. If it is primarily labor-substituting, "it will automate tasks previously performed by workers without enhancing the productivity of remaining jobs" — a path that cuts labor income without expanding supply, which is inflationary in composition if not in level.
There is a second channel working in the same cautious direction. Schnabel noted that if AI raises labor insecurity, workers may increase precautionary savings, which would dampen the natural rate of interest; similarly, if AI widens income and wealth inequality, upward pressure on the natural rate would be mitigated. In other words, even the direction of AI's effect on the neutral rate — the single most important unknown in monetary policy — depends on a social outcome that no central bank can forecast.
The lag structure matters because monetary policy works with lags of its own. A rate cut decided today affects spending and prices over the next year or more. If the Fed cuts on the strength of today's disinflation while the AI capex cycle is still feeding through electricity bills, business software fees, and commercial rents for data-center space, it risks discovering the mistake only when core services inflation has already re-accelerated. That is the scenario officials were pricing in September, when the median "appropriate" funds rate for the end of 2026 moved to 4.1%, above the 3.50%–3.75% range then in force.
The Neutral-Rate Problem: AI May Be Raising r-Star While Lowering Prices
Here is the uncomfortable implication that most of the AI-inflation debate misses. Central banks do not set policy against inflation alone; they set it against the gap between the actual rate and the neutral rate — the level that neither stimulates nor restrains the economy. AI plausibly pushes that neutral rate in two directions at once.
On one side, a genuine productivity boom raises the return on capital, draws in investment, and lifts the natural rate. On the other, labor substitution and precautionary saving pull it down. The Fed's own officials are already moving: a couple of participants in September said they had raised their neutral-rate estimates, and the median "appropriate" funds rate for the end of 2026 now sits above the current target range. Markets agree on the direction, if not the urgency: a survey of 102 economists conducted in June found 72 expecting the federal funds rate to hold unchanged through the end of 2026, with major banks including Goldman Sachs, JPMorgan, Barclays, and Morgan Stanley all forecasting no rate change through year-end.
That gap between policymakers and markets is the real story. If the neutral rate has risen because of AI-driven investment demand, then a policy rate that looks appropriate against today's inflation print may already be too loose — and the inflation consequence will show up only after a lag. Chair Jerome Powell has been candid about the uncertainty. At his January 2026 press conference he said that "ultimately technology increases productivity, which is the basis for rising wages," but added: "we always ask, well this is going to be different, is it going to be different? And we don't know." A month earlier, in December 2025, he had been more effusive, saying he "never thought I'd see this many years of really high productivity" and that AI spending would continue to support growth.
"We always ask, well this is going to be different, is it going to be different? And we don't know." — Jerome Powell, Federal Reserve Chair, January 2026
The ECB has moved more cautiously. Christine Lagarde, at the December 2025 press conference, tied AI to a broad investment upturn focused on computing capacity, telecommunications networks, and intangible assets rather than traditional physical capital, but warned against drawing premature conclusions about its impact on the neutral rate. The euro area projections released alongside that decision showed growth of 1.4% in 2025, slowing to 1.2% in 2026, with headline inflation averaging 2.1% in 2025 before dipping below target and returning to 2.0% in 2028 — a path that prices in a productivity tailwind but refuses to bet the policy rate on it.
The Strongest Case Against Worrying — and Why It Is Only Half Right
The dovish counter-thesis is straightforward and it has serious backing. History's general-purpose technologies — electricity, the internal combustion engine, the computer — took decades to show up in the productivity statistics, and their biggest price effects were downward. The post-2022 productivity pickup, on this view, is partly a statistical artifact of post-pandemic reallocation and a government shutdown that distorted the 2025 data. The Kansas City Fed's own caution applies: the AI-productivity correlation is moderate, and causality can run both ways. If the productivity numbers revert toward the 1.5% pace of the 2007–2019 cycle, the disinflationary leg disappears and the whole debate dissolves into a normal, manageable inflation cycle.
There is also scale. Even the most aggressive estimates put data centers at 1%–2% of global energy demand today, with the International Energy Agency projecting a doubling by 2026 and some scenarios reaching as high as 21% by 2030. That is a large number in growth terms but still a slice of the economy. Electricity carries a small weight in the consumer price index, and a 6% rise in one component is not the same as economy-wide inflation. If natural gas prices fall as expected, the Goldman Sachs baseline has electricity inflation decelerating to 3.5% by 2028 and to about 2% for non-AI businesses.
But this counter-thesis underestimates two things. First, the inflationary channel is not just electricity; it is the whole capital-intensity shift — software fees, data-center construction, grid congestion, and the pass-through of power costs into every digital service. The September Fed minutes explicitly named software and portfolio management fees as upward contributors. Second, the disinflationary channel is not just a promise; it is already in the data, at 2.1% annualized productivity growth, and it is already changing policy deliberations. The dovish view requires believing that the productivity numbers will revert while the capex numbers will not — a combination that has happened before, but not often.
The strongest version of the counter-argument does not claim AI is disinflationary. It claims that central banks cannot tell the difference in real time, and that over-reacting to a supply-side boom risks choking off the very productivity gains that would solve the inflation problem without a recession. That is a real risk. It is also the risk that a central bank with a 2% mandate is paid to take.
What to Watch: The Signals That Will Settle the Debate
Three data streams will determine whether AI is a central banker's friend or foe. First, nonfarm business output per hour: if it holds above 2% year over year through 2027, the structural productivity story survives; if it stalls below 1%, the disinflationary leg was cyclical after all. Second, electricity and software prices: consumer electricity inflation staying above 6% through 2027, combined with re-accelerating core services, would confirm that the capex channel is broadening into persistent inflation. Third, the neutral-rate estimates themselves: if more Fed officials raise their longer-run rate projections in the December 2026 Summary of Economic Projections, the policy path will follow, regardless of what futures markets currently price.
The base case is that AI is structurally disinflationary over a multi-year horizon but cyclically inflationary over the next 12 to 18 months. That combination keeps central banks cautious rather than dovish: rate cuts remain possible if the labor market weakens, but the burden of proof has shifted to anyone arguing for aggressive easing. The upside case is a 1990s-style productivity surge in which output growth absorbs the capex boom and inflation falls faster than forecast. The downside case is a 1970s-style supply shock in which energy and capacity constraints dominate, productivity disappoints, and the neutral rate rises faster than policy can follow.
The verdict: central bankers should be worried, but not about AI making inflation permanently higher. They should be worried about the timing mismatch — about cutting rates for a disinflation that is real but slow, while an inflation that is cyclical but fast is still working its way through the grid, the data center, and the software bill. The policy error to fear is not staying too tight for too long; it is mistaking a lag for a trend.
Explore more exclusive insights at nextfin.ai.
