Share

    • OpenAI and Broadcom Unveil Custom AI Chip Named 'Jalapeño'

    By Any June 24, 2026
    OpenAI announced it has begun testing its first in-house AI processor, named 'Jalapeño,' developed in collaboration with Broadcom. The custom chip is optimized for large language model inference and is expected to be deployed for commercial use by the end of 2026.
    OpenAI and Broadcom Unveil Custom AI Chip Named 'Jalapeño'

    OpenAI and Broadcom Unveil Custom AI Chip Named ‘Jalapeño’ OpenAI’s move into custom silicon is accelerating, as the company and Broadcom push a new in‑house AI chip from lab prototype toward commercial deployment, reshaping its relationship with traditional GPU suppliers.

    Early design and unveiling

    On June 22, 2026, OpenAI formally introduced “Jalapeño,” describing it as its first “Intelligence Processor” and an accelerator “architected around OpenAI’s vision for the future of LLM inference.” The company framed the chip as the first in a multi‑generation compute platform, built with Broadcom and Celestica to make advanced AI “faster, more reliable, and more accessible to more people.”

    OpenAI said the chip was designed “from scratch” around its understanding of large language models and future product needs, with Broadcom providing silicon implementation and high‑performance networking to scale Jalapeño into production systems.

    Lab testing and technical ambitions

    Engineering samples are already running machine‑learning workloads, including models such as GPT‑5.3‑Codex‑Spark, at target production frequencies and power levels, with early results indicating “performance per watt substantially better than current state‑of‑the‑art,” according to OpenAI. A detailed technical performance report is promised in the coming months.

    From lab to customers

    By June 24, OpenAI confirmed it had begun testing Jalapeño as the first in a family of homegrown chips, with plans to start handling customer queries later in 2026. Broadcom expects initial commercial use at Microsoft and other partners by year‑end, with “real volume” coming in 2027.

    The chips are aimed at inference rather than training and are “specifically designed for handling current and future models,” promising better performance per watt than off‑the‑shelf options.

    Strategic stakes and industry context

    Broadcom CEO Hock Tan argued that leading AI firms need their own silicon, saying companies “cannot, should not rely on some other third‑party GPU” for such a critical capability. Richard Ho, who leads OpenAI’s chip efforts, said the project gives OpenAI “full stack control” to deliver compute more efficiently and cost‑effectively in pursuit of AGI that benefits all of humanity.

    Even as it develops Jalapeño and targets 10 gigawatts of custom‑chip compute by 2029, OpenAI stresses that Nvidia remains a key partner, especially for training new models.

    Continue reading https://foxvector.com/stories/019efaaf-2c8d-00ed-7077-00f832f5fb02

    Write a comment