Semiconductors

Inference Optimisation

The set of techniques — quantisation, batching, caching, distillation and specialised silicon — that reduce the cost and latency of running a trained model.

Liquid Cooling

Moving heat away from servers using liquid — either cold plates attached to chips or full immersion in a dielectric fluid — instead of moving air.

Cirrus Inference

Inference serving stack aimed at cost per token rather than peak throughput.

Austin, United States · Founded 2022

Helio Silicon

Inference accelerators designed around memory bandwidth instead of raw FLOPS.

Hsinchu, Taiwan · Founded 2021

Inference, not training, is the cost line

As AI products reach volume, attention is shifting from training runs to the recurring cost of serving. Routing, compression and purpose-built silicon are where margin is won.