The United States is currently leading to AI hardware and software, but Deepseek and Huawei of China continue to push advanced fleas, India has announced an indigenous GPU program targeting production by 2029 and political changes to Washington reshuffle the rules of the game. In the second quarter, the decline in export restrictions allowed American companies AMD to conclude several billion dollars agreements in Saudi Arabia.
JPR categorizes suppliers in five segments: IoT (ultra-available inference in microcontrollers or small SOCs); Edge (survival or almost peripheral inference in the range of 1 to 100W, used outside the data centers); Automobile (sufficiently distinct to burst from the edge); data center training; and the inference of the data center. There is a certain overlap between the segments because many suppliers play in several segments.
Of the five categories, inference has the most startups with 90. Peddie says that the list of inference applications is “gigantic”, with everything, from portable health monitors to intelligent vehicle sensor networks, personal items at home, and each machine imaginable in all imaginable manufacturing and production lines, as well as robotic boxes and surgeons.
Inference also offers the most versatility. In the past, “smart devices”, such as washing machines or coffee makers, could do one thing and could not adapt to any change. “Inference -based systems will be able to hide and forge, adapt in real time and find alternative solutions quickly,” said Peddie.
Peddie said that despite his apparent cynicism, this is an exciting period. “There are really new ideas that are tried like processors of analog neurons and processors in memory,” he said.
