Google has published research on an algorithm it calls TurboQuant, which it claims delivers significant improvements in how AI models consume memory. The announcement was enough to unsettle the memory chip sector: Micron fell approximately 5%, SanDisk dropped around 8%, and shares in Samsung and SK Hynix also declined as investors assessed what lower memory intensity in AI workloads could mean for future chip volumes.
The logic of the selloff is straightforward. Memory chips — particularly high-bandwidth memory — have been one of the cleaner AI infrastructure trades, with demand from data centre buildouts underpinning a recovery at Micron and a surge in SK Hynix's earnings. Any credible technology that compresses memory requirements threatens the volume assumptions embedded in those valuations.
The degree of that threat is far from settled. MarketWatch noted that whether the Google algorithm will actually eat into revenue for Micron and its rivals remains unclear. Bloomberg separately reported that analysts see the memory stock boom as resilient to the new technology, citing structural demand that compression gains are unlikely to fully offset.
That tension — between a genuine efficiency advance and the scale of underlying AI infrastructure spending — is likely to define how the market re-rates the sector in coming sessions. Micron, which has a market capitalisation of roughly $90bn, has now fallen sharply on multiple occasions this year as investors weigh cyclical recovery against longer-term demand risks.
