Skip to content

Broadcom News: Latest Releases & Events

Broadcom partners with OpenAI on Jalapeño, an inference chip OpenAI designed using its own LLMs, IEEE Spectrum reported in September. In August, SemiAnalysis found it beat every Nvidia, AMD and Google chip tested on throughput per MW without speculative decoding, after OpenAI's June announcement. In April, Anthropic committed to about 3.5GW of Google TPU capacity through Broadcom from 2027.

All 4 Broadcom entries of the AI News timeline, newest first.

2026

September 2026

2026-09-14

  • OpenAI's LLMs Designed Jalapeño Chip in Record Time
    IEEE Spectrum detailed how OpenAI used its own LLMs, from o3 up to internal precursors of GPT-6 Astra, to design its debut AI accelerator chip Jalapeño, compressing the timeline from first concept to first silicon to under 20 months, with only nine months from first RTL to tape-out, achieved by a team of fewer than 100 people partnering with Broadcom. OpenAI claims Jalapeño cuts end-to-end inference latency up to 3.6x versus Nvidia's GB300 while consuming less power, and that AI-driven software optimization lifted a DeepSeek attention kernel benchmark from 0.31% to 88.94% of its theoretical ceiling in roughly 40 hours. (spectrum.ieee.org)
    OpenAI Nvidia

August 2026

2026-08-25

  • OpenAI's Jalapeño inference chip beats Nvidia Blackwell
    SemiAnalysis published detailed benchmarks of OpenAI's first custom inference ASIC, Jalapeño, built with Broadcom and taped out in roughly 16 months, finding it beats every Nvidia, AMD, and Google chip tested on InferenceX throughput per MW without speculative decoding, and surpasses Vera Rubin's published perf/MW figures while matching its TCO. The results, achieved via extreme hardware-software co-design and Codex-driven kernel development, suggest a credible threat to Nvidia's CUDA moat, with production ramping through 2027. (newsletter.semianalysis.com)
    OpenAI Nvidia AMD

June 2026

2026-06-26

  • OpenAI announces Jalapeño custom AI inference chip
    OpenAI revealed plans for Jalapeño, a custom inference chip built in partnership with Broadcom, joining Google, Apple, and SpaceX in a broader industry trend of reducing single-supplier dependence on Nvidia. The move aims to give OpenAI more control over hardware tuned to its specific needs and unlock performance gains similar to what Apple achieved when it transitioned away from Intel. (techcrunch.com)
    OpenAI

April 2026

2026-04-07

  • Anthropic hits $30B run rate, commits to 3.5GW Google TPUs
    Anthropic revealed its run-rate revenue has surpassed $30 billion—up from approximately $9 billion at the end of 2025—and committed to accessing approximately 3.5 gigawatts of next-generation Google TPU-based AI compute capacity through Broadcom starting in 2027. Broadcom's regulatory filing noted that Anthropic's ability to consume this capacity is dependent on its continued commercial success, flagging it as a risk factor. (go.theregister.com)
    Anthropic Google