A must read
Two weeks ago I sat in the @RAEngNews watching a quiet Oxford physicist make the most aggressive claim I have heard in British deep-tech this year.View quoted post on X
Ninety-five per cent less energy than an @nvidia GB200 by 2029. Ninety per cent lower cost. A British company doing AI inference with light instead of electrons.
I followed up. The result is this week's Sunday Signal.
The company is @Lumai. The CEO is Dr Xianxin Guo. The Iris Nova server, launched last month, is currently running @AIatMeta's Llama 8B and 70B on optical compute. The only system in the world doing it.
Three lines from the interview that I cannot stop thinking about.
"Cost per token. That is the only metric that matters."
"Jevons' paradox is real, and we will see more AI used, not less."
"Where the company is built is the question that matters, and the answer is the UK."
His ask of the @GOVUK state was not subsidy. It was procurement. A committed government anchor purchase of Lumai hardware into the AI Research Resource, this financial year, or watch the next generation of AI compute be created outside of the UK again.
Britain has watched @GoogleDeepMind, Magic Pony, @vocaliq, Latent Logic and @OxfordIonics walk out of the door.
Light is the next one. Buy the thing, or read the press release.
๐ก The full issue: open.substack.com/pub/davidjrichโฆ
๐ก The full interview with Dr Xianxin Guo: open.substack.com/pub/davidjrichโฆ