Team tapes out physical-compute chip in 5 months, generates images for under 900 nJ each
NaveenGRao · x · 2026-09-17
A founder shared chip progress: from no team in January to tape-out in 5 months, with hardware now back. The chip is billed as the first large-scale demonstration of a causal physical dynamical system doing real compute, running the Un-0 model released in June—the images in the post come from the actual hardware.
Key number: generating an image costs <900 nJ, orders of magnitude less energy than conventional machines. The author credits AI for dramatically tightening research loops and enabling the speed.
More from Infra
- Ciena sets 3-year targets: ~30% annual revenue growth, ~50% gross margin as quiet AI winner — kevinsxu · 2026-09-17
- If compute demand outpaces supply, distributed general-purpose computing could win — gajesh · 2026-09-17
- Framework Desktop with 128GB unified memory: Qwen 27B-35B beats bigger local models — lebbi · 2026-09-17
- Alchemy ships official coding-agent prompt for its TypeScript IaC tool — samgoodwin89 · 2026-09-17
- Close-up of Ayar Labs optical interconnect solution on GPU tray — BenBajarin · 2026-09-17
- Apple reportedly building enterprise AI server with its own M8 Ultra chips, launching no earlier than 2029 — The Decoder · 2026-09-17