A I Accelerator - A New Approach to Artificial Intelligence

Training a large a i agarithm can cost tens of millions of dollars. With general purpose chips like c pus or older ai chips, this training would take much longer and cost orders of magnitude more. The opportunity is so large that there are hundreds of a i accelerator start ups designing their own chips. They use intersecting coherent light beams rather than switching transissors to perform matrix multiplies. Computation happens in pico seconds and requires only power for the lazer. Though off chip digital transition still limit power savings.