They want to build the systems AI runs on
Inference, GPU efficiency, training platforms and serving. Agent work rises a little (23% vs 18%). Infrastructure is where the jump is.
“Deep technical work on GPU infrastructure, distributed training systems, or ML serving at scale.”