Hourly · 2026-06-25 12:00
OpenAI Just Built Its Own Chip. NVIDIA Should Be Worried.
OpenAI unveils Jalapeño, its first custom AI chip built with Broadcom — an inference processor that cuts costs 50% and signals the beginning of the end for GPU monoculture.
OpenAI took the wraps off its first custom silicon on Tuesday — an inference processor codenamed Jalapeño, co-designed and manufactured with Broadcom. The chip is built for one thing: running already-trained models as cheaply as possible. Early testing shows it delivers 50% lower cost versus typical GPUs.
OpenAI president Greg Brockman: "We have really been looking for specific workloads that are underserved, and asking how can we build something that will accelerate what is possible." The chip is the first output of a 10-gigawatt Broadcom partnership, with rack deployments beginning H2 2026.
Broadcom CEO Hock Tan did not mince words: "Every frontier lab will have a custom chip." Google has TPUs. Amazon has Trainium. Now OpenAI has Jalapeño. Real-time coding models are expected to be the first major workload.
Nine months from design to tape-out — believed the fastest ASIC cycle ever for high-performance semiconductors. Engineering samples are already running GPT-5.3-Codex-Spark at production frequency and power.
