OpenAI unveils Jalapeño, its first custom inference chip, co-developed with Broadcom

OpenAI unveiled Jalapeño, a custom inference processor co-developed with Broadcom for real-time coding workloads, with early results showing better performance-per-watt than current alternatives — though pre-training workloads will still run on Nvidia. With this move, OpenAI joins Google (TPU), Amazon (Trainium/Inferentia), and Microsoft (Maia) as a frontier-lab operator with proprietary inference silicon, directly pressuring Nvidia's pricing power in the AI accelerator market.

Source: techcrunch.com ↗

We have a deep understanding of the workload. We've really been looking for specific workloads that are underserved, [and asking] how can we build something that will be able to accelerate what's possible?

Greg Brockman, OpenAI president

Why this matters

→ Reduces inference costs, improving AI service profitability at scale
→ Breaks Nvidia's stranglehold on custom silicon for frontier labs
→ Enables full-stack optimization across models, hardware, and infrastructure

Breaking Nvidia's grip