OpenAI develops proprietary Jalapeño chip for LLM inference

💻 OpenAI develops proprietary Jalapeño chip for LLM inference

OpenAI has introduced Jalapeño — a specialized processor created in collaboration with Broadcom and Celestica. The chip went from design to tape-out in just 9 months, which is a record cycle for high-performance ASICs. The architecture is optimized for model workload patterns (kernels, memory movement, networking), and engineering samples are already being tested on the GPT-5.3-Codex-Spark model.

🌍 The shift by major AI labs toward owning the full stack (from chip to model) accelerates the innovation cycle and reduces dependence on general-purpose accelerators. These record development timelines are setting a new industry standard.

👤 This means that future models (such as GPT-5) will run faster and cheaper thanks to hardware that "understands" their architecture at a physical level.

Source 1: https://the-decoder.com/openai-and-broadcom-unveil-jalapeno-a-custom-chip-built-for-llm-inference/