Server hardware and software co-design for a secure, efficient cloud.
Adding big blocks of SRAM to collections of AI tensor engines, or better still, a waferscale collection of such engines, ...