Many economists now expect the AI buildout to add roughly half a percentage point to core consumer prices by the end of 2026, according to an Associated Press analysis published July 13 [1]. Core inflation, which strips out food and energy, ran 4% in May and may ease only slightly by year-end, staying well above the Federal Reserve's 2% target. The AP notes that such pressure could push the central bank to raise its key interest rate later this year to cool spending.
The mechanism is physical, not abstract. Data-center investment is likely to top $700 billion this year, and that spending has bid up memory chips, processors, and electricity. JPMorgan Chase estimates the cost of some computer memory chips will rise as much as 400% between 2024 and the end of 2026. Apple, citing "an extraordinary surge in demand for memory and storage," raised laptop and iPad prices roughly 15% to 25% last month. Shoppers already pay more for laptops, smartphones, game consoles, and PCs.
This tracks the paper's July 10 report, which separated today's measured AI costs from tomorrow's modeled productivity gains. AP's number sharpens that distinction: the half-point is a near-term estimate of demand pressure now, not a verdict on whether AI eventually lowers prices through efficiency.
The framing gap is real. Booster commentary on X leaps to a deflationary future in which AI productivity drags prices down; alarm-side commentary treats the shock as permanent. AP does neither. Analysts at Evercore ISI told the wire that the "wave of AI-related cost pressures spilling over into consumer prices is still in the early stages of building" — early stages, not endpoint. The half-point is bounded: smaller than the 2021-2023 spike, when inflation peaked at 9.1%, and attached to a specific year and a specific set of goods.
The reader loses if either frame wins. Fed officials will watch the June inflation report, due Tuesday, for further signs of AI's imprint on prices, and last month's reading likely cooled as gasoline fell after the Hormuz ceasefire. Whether the half-point holds depends on that data, not on the confidence of the loudest post [1].
-- THEO KAPLAN, San Francisco