The chip promises Nvidia Blackwell-level performance at half the cost, but investors aren't exactly throwing a party yet.
Broadcom and OpenAI just pulled the wrapping off Jalapeño, a custom AI inference chip designed to power the kind of large language model queries that keep ChatGPT humming along. The chip emerged from a partnership that kicked off in October 2025, and early samples are already being tested.
Broadcom’s stock is sitting roughly 24% below its early 2026 highs, a gap that opened after disappointing AI revenue guidance on June 4. A shiny new chip announcement hasn’t been enough to close it.
What Jalapeño actually is
Jalapeño is a custom ASIC, or application-specific integrated circuit, built specifically for AI inference workloads. In English: it’s a chip designed to do one thing really well, which is running the computations that happen every time someone asks ChatGPT a question.
OpenAI’s engineers led the design process while Broadcom handled the production side. The whole design-to-tape-out cycle took nine months, which is genuinely fast for custom silicon.
The performance claim is the headline grabber. Jalapeño reportedly delivers performance comparable to Nvidia’s Blackwell architecture at roughly half the cost.
Initial deployment is planned for late 2026, with production ramping through 2029. The broader collaboration targets approximately 10 gigawatts of OpenAI-designed AI accelerators.
Key figures involved include OpenAI CEO Sam Altman, President Greg Brockman, and hardware lead Richard Ho on the OpenAI side, with Broadcom CEO Hock Tan and President Charlie Kawwas representing Broadcom.
Why Broadcom’s stock isn’t celebrating
Broadcom shares dropped roughly 12-14% on June 4 after the company’s AI revenue guidance came in below what the market had priced in. The stock has been trading 24-26% below its early 2026 peak since then, and the Jalapeño reveal hasn’t sparked a meaningful recovery.
The Jalapeño partnership is a long-term play that won’t meaningfully impact revenue until late 2026 at the earliest. Broadcom remains one of the dominant players in custom AI accelerators, with clients including Google.
The bigger picture for AI hardware
OpenAI is following the playbook that Google, Amazon, and other hyperscalers have been running for years: building proprietary silicon to reduce dependence on Nvidia’s GPUs.
OpenAI’s move represents the construction of a fully integrated ecosystem, from the models themselves down to the silicon running them. The company even used its own AI models to help optimize the chip design during development.
