Meta’s decision to more than double the planned capacity of its Louisiana data centre to 5 gigawatts, at a cost exceeding $50bn, delivered an immediate split-screen on Wall Street. Meta shares jumped 8.8% on the news, but semiconductor stocks tumbled—Micron fell 10.6%, AMD nearly 7%—as investors began to price in the risk that the artificial-intelligence buildout may be running ahead of demand. The move came just twelve days after reports that Meta was developing plans to sell “excess” AI computing capacity, a contradiction that crystallised a question the entire sector has been dodging: how much of the compute already bought is actually being used.

The scale of the spending is now feeding through to consumer prices. JPMorgan Chase economists estimate that the cost of some memory chips will have risen as much as 400% between 2024 and the end of this year, driven by the torrent of data-centre investment—likely to surpass $700bn globally in 2026 from just four hyperscalers. Apple has raised laptop and iPad prices by 15–25%, Microsoft added $100 to its Xbox, and Sony and Dell have followed suit. Electricity costs are also climbing as data centres absorb a growing share of new grid capacity. The US Federal Reserve is watching the June inflation report closely for signs that AI-related price pressures are offsetting disinflation elsewhere, with some economists forecasting a half-point boost to core inflation by year-end.

In Louisiana’s Richland Parish, the local impact is starkly different. Meta says the project has already delivered $1.6bn in contracts to local businesses and year-end bonuses of up to $50,000 for teachers—a 400% increase on the previous year. Yet the company’s AI push is generating friction elsewhere. Advertisers report that Meta’s automated creative tools are producing garbled images and unapproved product alterations, forcing brands to double-check every campaign. Meta’s response—that its terms place responsibility on advertisers to review AI outputs—has done little to quell frustration among agencies managing hundreds of millions in annual spend. Privacy tensions are also rising: Meta is adding tamper-detection to its Ray-Ban smart glasses, but it is simultaneously testing a “super-sensing” prototype that would capture photos every few seconds and record audio continuously, prompting alarm from privacy advocates.

The next milestone is whether the hyperscalers can demonstrate utilisation rates that justify the capital outlay. Meta’s plan to resell spare capacity may be a prudent cloud-style move, but without a disclosed utilisation figure, it reads to some as a hedge against overbuilding. The Fed’s inflation reading, due Tuesday, will offer the first hard data on how AI spending is rippling through the real economy. For now, the market is beginning to grade not just the size of the buildout, but what it actually returns.