AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

Hot Chips: OpenAI’s Jalapeño, Cerebras CS-5, Groq 3 LPX, Apple M6

· Latent Space

OpenAI announced its custom inference chip, Jalapeño, at the 37th Hot Chips conference, claiming materially better efficiency and latency than NVIDIA GB200/GB300 systems. Jalapeño reportedly delivered 1.5–1.9× more work per watt and 1.7–3.6× lower end-to-end latency, with 2.1–4.1× higher performance for interactive workloads. The chip is rated at 700W but stayed at or below 550W during tested runs. OpenAI stated that GPT-Astra + Codex helped write and optimize low-level kernels for Jalapeño, bringing three previously unplanned open-weight models to high performance in about two months. Deployment into OpenAI's infrastructure begins by year-end, with Gen 2 and Gen 3 already in development. This development suggests a shift in inference economics for frontier labs, potentially reducing their reliance on NVIDIA.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.