AI inference hardware

Etched, a startup building AI inference chips, is reportedly fielding funding offers at a valuation exceeding $40 billion, according to sources cited in press reports. The figure would place the young semiconductor company among the most valuable private AI hardware firms, reflecting intense investor demand for alternatives to Nvidia's dominant chips. Details on the size and terms of the round have not been confirmed.

Cedana has published a tool and guide that translates large language model token volumes — millions, billions, or trillions — into the number of GPUs needed to serve them. The resource is drawing attention on Hacker News, where commenters are discussing how engineers can estimate infrastructure requirements for AI workloads and what token throughput actually means in terms of hardware cost and capacity.