TurboQuant promises to be a breakthrough in memory compression algorithms for LLM tasks related to generative AI models. Google claims to have developed a method to maintain the highest accuracy of models while reducing RAM usage by a factor of six.
The market is already reacting to the new product. In the last week alone, shares of Micron, a memory manufacturer, have fallen by more than 15%. SK Hynix has lost about 13% on the stock exchange in recent days.
The situation gives hope to the entire hardware market that the monstrous demand for RAM from AI data centers will decrease. This would improve the situation with the insane prices for RAM and NAND Flash.