On Tuesday morning, Google published a research paper called TurboQuant. By Thursday, SK Hynix had fallen 6% in Seoul, Samsung had fallen nearly 5%, Kioxia had dropped 6%, and Micron was lower in US premarket trade. The Cloudflare CEO called it Google's DeepSeek moment. Financial media ran headlines about the death of the memory boom.
Almost all of it missed the point.
TurboQuant compresses the KV cache — the working memory AI models use during inference — by 6x with zero accuracy loss and no retraining required. The stock market read this as: AI needs less memory. Chip stocks fell.
The stock market was wrong. Or at least, premature. Morgan Stanley clarified it within 24 hours. What TurboQuant actually does is allow systems to handle 4 to 8x longer context windows on the same hardware. It does not reduce the ceiling of what you need. It raises the floor of what you can do.
This week we also cover Arm ending 35 years of neutrality, a bottleneck nobody is talking about, AI stealing CPUs from your laptop, and two quieter signals that will matter more than most people realise.
[Lots of exciting stuff. Continue reading the Chip Brief - CLICK HERE]