AI Was Supposed to Lower Prices

Goldman Sachs just put a number on the bill Wall Street spent three years promising would never arrive: AI is already adding roughly 20 basis points to U.S. core PCE inflation — and may hit 50 basis points by December, before any productivity payoff shows up in your paycheck.

On July 11, Goldman economist Megan Peters published fresh estimates quantifying what executives euphemistically call "infrastructure buildout costs" and what consumers experience as higher laptop prices, pricier software subscriptions, and creeping power bills.

The number: AI-driven price pressure is lifting the Fed's preferred inflation gauge by about 20 basis points annually right now. Goldman projects that punch rises to 50 basis points by year-end — more than double today's drag.

Canada, Australia, Europe, the U.K., and Japan will see roughly 10 basis points on average.

"While not completely negligible, these effects are far below the 50bp peak we estimate for US PCE, suggesting that for the most part AI-driven inflation is a US story," Peters wrote, per Business Insider.

The irony: In May, the same Goldman economics team said the long-run story was still disinflationary — but conceded the near term looked different.

"We expect artificial intelligence to deliver large productivity gains over the next several years... So far, however, AI is boosting US inflation," they wrote in an earlier note.

That sequencing matters for the Fed. Officials who priced in AI-driven disinflation to justify rate cuts now face a rival scenario: the buildout tax arrives first.

Wave 1 — Memory: Data-center demand has crushed the consumer supply chain. The average price of an 8 GB DDR5 module climbed to roughly $148 in the last week of reporting — more than triple the $35 average in the same week a year earlier, per Pangoly data cited by Business Insider.

Goldman expects U.S. software-and-accessories inflation to peak near 30% year-over-year in November 2026, contributing an estimated 36 basis points to core PCE on its own.

Wave 2 — Software: Vendors are not hiding the markup. Microsoft lifted flagship 365 pricing after bundling Copilot. Goldman flagged Adobe, Duolingo, Atlassian, Intuit, and Apple for similar AI-feature upcharges in its May analysis.

Software and accessories account for roughly 1% of U.S. PCE inflation — less than half a percent in other developed markets Goldman measured. The AI tax is structurally American.

Wave 3 — Electricity: The average U.S. city price for one kilowatt-hour hit $0.19 in May 2026, up roughly 27% since May 2022, according to Bureau of Labor Statistics data cited by Goldman.

Data centers are on track to consume 11% of total U.S. power demand by decade's end, up from 6% today, Goldman estimated.

The receipt, same day: On July 11, Meta announced a C$13 billion ($9.17 billion) AI data center in Sturgeon County, Alberta — its first in Canada and 33rd globally. The one-gigawatt facility is expected to consume electricity equal to 800,000 homes, per Reuters reporting via the Economic Times.

Meta is not waiting on the public grid. Pembina Pipeline and partners reached a final investment decision on the $4.6-billion Greenlight Electricity Centre — a 932-megawatt gas-fired plant built to serve the data center, scheduled for the second half of 2030.

Separately, Capital Power signed a 250-megawatt supply deal with Meta for the back half of 2028 — bridging the gap before Greenlight comes online.

Alberta technology minister Nate Glubish told reporters: "This is the first of its kind, the first of its size, the first of its scale, but it won't be the last."

Meta VP Gary Demasi said the company will offset consumption with clean-energy investments and use closed-loop liquid cooling designed to use less water than a typical golf course.

Capital Power CEO Avik Dey said: "AI infrastructure will be built where power is available, reliable and scalable — and, with the support of Capital Power's fleet, Alberta meets the mark."

Why investors should care: Goldman's 50-basis-point estimate does not fully account for spillover effects — the bank acknowledges the true hit could be larger. If core PCE carries an extra half-point of AI-driven pressure into year-end, the Fed's rate-cut window narrows precisely when IPO-week labs need cheap capital to fund the buildout causing the inflation.

What we cannot verify: Goldman has not publicly released Peters's full research note. Business Insider is our primary source for the July 11 figures and quotes. Secondary outlets echo the 50-basis-point projection but we have not independently confirmed the component-level 36-basis-point memory contribution.

Convina's view: The AI trade sold disinflation on a productivity timeline nobody can invoice yet — and Goldman just priced the interim bill at half a point of core PCE. Memory scarcity, Copilot upcharges, and dedicated gas plants are not side effects of progress; they are the mechanism. Washington can debate export controls and IPO gates all it wants. The inflation Fed officials need to explain is already in your laptop cart and your power bill — and Meta just proved hyperscalers would rather contract private gas plants than wait for the grid to catch up.