OpenAI, Broadcom Unveil Jalapeño AI Chip
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI and Broadcom unveiled Jalapeño, an LLM-optimized inference chip developed from design to manufacturing tape-out in nine months, aided by OpenAI's own models.
- OpenAI says Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products, with the chip designed from the ground up rather than adapted from existing silicon.
- OpenAI's early testing claims the first-generation accelerator will deliver performance per watt substantially better than current state-of-the-art inference chips.
- CNBC frames the chip as part of OpenAI's effort to "build the full stack," while Yahoo Finance calls it "a strike at Nvidia," making Nvidia the implicit target of OpenAI's vertical integration.
- Analyst Ben Bajarin publicly pushed back on the inference-only framing, writing that the chip "looks more like a training chip" despite OpenAI's positioning.
- Professor Thibault Schrepel noted the move contradicts prior antitrust assessments that concluded the chip market "could not possibly be competitive" — a direct rebuke to those papers.
- OpenAI confirmed Broadcom as its manufacturing partner, with the chip developed from scratch around OpenAI's understanding of LLM fundamentals and its roadmap of models, kernels, and serving systems.
Why it matters: OpenAI designed its first in-house inference accelerator with Broadcom in just nine months, and outlet framing — CNBC's "build the full stack" and Yahoo Finance's "strike at Nvidia" — makes clear the chip is aimed at loosening Nvidia's grip on AI compute. By designing silicon around its own model and serving roadmap rather than buying off-the-shelf GPUs, OpenAI gains control over the cost and performance curve beneath ChatGPT and its agentic products.
