Chips and compute
Chips, data centres and capacity, where they change what models cost or who can train them.
3 stories, newest first
- OpenAI says its models improved AMD GPU design by up to 14%19 November 2026. The PPA gain took under five months, and OpenAI says AMD has put some of the changes into a design sent to manufacturing.
- Nvidia shows GLM-5.3 at 607.8 tokens per second on two DGX Stations15 November 2026. Nvidia splits the 464 GB checkpoint across two GB300 GPUs; throughput falls to 385.0 tokens per second at concurrency 32.
- Anthropic says Mythos ported Claude to AMD GPUs in three days12 November 2026. The claim comes from an investor presentation and is unverified. Anthropic says its model wrote or ported nearly all the kernels itself.