Intel CEO Pat Gelsinger announced the Gaudi 3 accelerator on Tuesday, alongside a new processor optimized for retrieval-augmented generation, a process for making artificial intelligence systems more accurate. (The Logic)

Talking point: Modern machine-learning models—particularly the largest multipurpose ones powering generative tools—are mostly trained and run on specially designed chips. Nvidia’s H100 has been the AI boom’s hottest commodity, but other semiconductor firms and tech giants have been working on their own alternatives. Intel says Gaudi 3 trains open-source models significantly faster and delivers their output more effectively than the H100. And while Nvidia keeps developers on its chips with its accompanying software, Intel is touting an “open ecosystem” of device and application makers.