OpenAI and Broadcom jointly unveil Jalapeño, a custom chip designed for LLM inference workloads

Decision brief

What changed

OpenAI and Broadcom jointly unveiled Jalapeño, a custom chip optimized for LLM inference workloads, marking OpenAI's first entry into chip design; the chip was developed over nine months and is expected to begin deployment in late 2026, with broader scaling in 2027-2028.

Why it matters

This signals OpenAI's strategic push to control more of its infrastructure stack and reduce reliance on Nvidia GPUs, which could materially affect its long-term compute costs and supply chain risk. For competitors and partners, it raises questions about future inference pricing, capacity allocation, and the competitive dynamics of the AI chip market. The multi-year deployment timeline means near-term operational impact is limited, but it reshapes expectations about vertical integration in AI infrastructure.

Affected roles

CEO CFO COO CTO

Evidence

All three sources (OpenAI's own blog, TLDR, and TLDR Dev) consistently report the joint development, the nine-month design timeline, and the late 2026 deployment target; specific performance benchmarks were not disclosed by OpenAI itself, though TLDR Dev cites early testing claims of better performance-per-watt.

What remains uncertain

No independent, third-party verification of performance claims exists yet since benchmark numbers were not officially disclosed; it's unclear how much this will actually reduce Nvidia dependence in practice, what the cost structure will look like, or whether Broadcom will supply similar chips to OpenAI's competitors.

Monitor next

Watch for disclosed benchmark data or independent performance validation of Jalapeño as deployment approaches in late 2026.

Analytical support, not advice — assumptions and open questions stated above.

Source coverage