An AI-chip startup founded by former Alphabet chip engineers. MatX builds on the systolic-array concept used in Google's tensor processing units (TPUs), in which data flow rhythmically through a grid of processing elements—each calculation passing its result directly to the next unit, bypassing the need to store intermediate results in memory. MatX proposes a "splittable" systolic array that divides the processor into several smaller grids, allocating computing resources differently depending on whether the chip is handling prefill (processing a user's prompt) or decode (generating the response token by token). Traditional systolic arrays are fixed in size, leading to inefficiency; the splittable design aims to overcome this. As of early 2026 the chip remains in development.
I went to a Grateful Dead Concert and they played for SEVEN hours. Great song.