Reisner: Generative AI Is a Trillion-Dollar Engineering Disaster

There's a running assumption in the AI buildout that scale itself is progress, that if you keep pouring capital and silicon into larger models the industry will eventually earn its trillion-dollar valuation. Alex Reisner's new piece in The Atlantic argues something less flattering: the work of building efficient, scalable AI systems has not actually been done, and the resulting inefficiency is starting to cost people outside the industry.

The specific detail worth sitting with is the memory number. Reisner writes that large language models such as ChatGPT and Claude are so resource-hungry that tech companies may be purchasing 70 percent of the world's supply of high-end computer memory, causing a shortage. That claim tracks with TechRadar's read on 2026 memory forecasts, which describes data centers set to grab 70 percent of all high-end memory chips in 2026 as the AI boom leaves consumers in the cold. If both are pointing at the same underlying figure, the binding constraint on the buildout right now is memory rather than pure compute, and everyone who buys hardware for a non-AI reason is quietly paying the surcharge.

Reisner's framing of a trillion-dollar project as an engineering disaster is the sharper piece though. The implicit charge is that the industry has substituted capital for engineering discipline, buying its way past hard problems that a more mature field would have had to solve. If that read holds, the payoff curve for the current wave of hyperscaler capex is not the curve investors have been pricing.

The honest caveat is that this is one journalist's argument and the article itself sits behind the Atlantic's paywall, so most of the specifics available secondhand come from the outlet's own social summaries and one corroborating memory report. What the coverage does not give you is a named list of which companies drive the 70 percent, or which specific engineering shortcuts Reisner thinks the industry took. The forward-looking read, if you take his thesis seriously, is that the winners over the next couple of years may be the labs and vendors selling efficiency rather than more scale.