News
AI Chips · Broadcom · Custom Silicon · OpenAI
OpenAI and Broadcom officially launched their first custom AI chip, the Jalapeño Intelligence Processor (JIP), designed specifically for AI inference and agentic products like ChatGPT and Codex, marking a strategic move to control the full AI stack and reduce dependency on Nvidia.
OpenAI President Greg Brockman stated ChatGPT acted as a virtual chip architect, accelerating the design and delivery to just nine months. The JIP is inference-native, featuring ASIC silicon surrounded by eight HBM stacks to minimize latency.
OpenAI, in partnership with Celestica and Broadcom, plans to build the complete rack systems and network infrastructure. This "full stack" approach, mirroring strategies by Apple, AWS, and Google, aims for lower prices, enhanced feature differentiation, and faster market entry.
Broadcom CEO Hock Tan confirmed a small prototype development by late 2026, with global rollout ramping up in 2027 and reaching full deployment in the first half of 2028. The collaboration, first rumored in July 2024 and officially announced at GITEX in October 2025, positions Broadcom as a key provider for custom AI accelerators, attracting potential customers like Google, Meta, and ByteDance.
This initiative directly challenges Nvidia's market dominance, with OpenAI explicitly stating the platform is "not a general-purpose accelerator adapted from earlier AI workloads." The Jalapeño is the first in a multi-generation compute platform, with future iterations expected.
