TPU

What people use it for

Performs low-precision dense linear algebra for machine-learning inference with lower energy use and latency than the CPUs and GPUs described in the transcript.

Y Combinator