Most technological revolutions arrive dressed in vulgar costumes. A robot dog does a backflip, a chatbot writes a sonnet, some founder in a tight T-shirt announces that history has been abolished. Then, every so often, the really consequential thing turns out to be a piece of engineering so dry that only infrastructure obsessives initially notice. Google’s new TurboQuant looks rather like that sort of development. It is not a glamorous new model. It is not a humanoid servant or a silicon Shakespeare. It is, instead, a method for squeezing more usefulness out of the memory that AI systems already consume in yuge quantities. Markets paid attention.
