OpenAI and Broadcom Unveil Jalapeño, First Custom AI Chip for LLMs
What happened: OpenAI announced Jalapeño, its first custom inference chip for large language models, co-designed with Broadcom.
What happened: OpenAI announced Jalapeño, its first custom inference chip for large language models, co-designed with Broadcom. The chip was designed and taped out in nine months—a record pace for high-performance ASICs—and is already in lab testing with GPT-5.3-Codex-Spark. OpenAI claims Jalapeño will cut inference costs by up to 50% and deliver superior performance-per-watt, though no third-party benchmarks have been released. Broadcom shares rose 3% on the news.
Why it matters: Jalapeño marks OpenAI’s entry into the custom hardware race, challenging Nvidia’s dominance and aiming for "Apple-like" vertical integration. While the chip’s real-world performance remains unverified, the move signals a strategic shift toward controlling the full AI stack. If successful, Jalapeño could lower costs and democratize access to advanced AI, but skepticism remains until independent data emerges.
Source: Decrypt