Qualcomm’s AI 100 is a 7nm AI inference acceleration ASIC for the edge. They have been drumming up interest for this product for a while now, but recently at The Linley Conference, they divulged further details. Their main two focuses were performance per watt and latency. These are two pieces in the AI chain where Qualcomm believes they can beat Nvidia and CPU based solutions by a large margin. Furthermore, they seperate themselves from the AI ASIC pack by providing actually good software support. According to SemiAnalysis, this combination of hardware and software has landed Qualcomm a Super 7 Hyperscaler win in addition to a myriad of wins in the edge market.
How does this compare with Brainchip's Akida?
I did some work on that for a client, so can't discuss much publically. Sorry.
12 trillion parameters! That's bonkers