OpenAI Developed Custom Jalapeño AI Chip
The firm partnered with Broadcom to build its own hardware for running advanced AI inference models.
Updated on Sept. 28, 2026 in Semiconductors
Live Poll
Do you believe proprietary chips developed by AI companies will lead to better consumer technology?
In August 2026, OpenAI revealed it developed a custom AI inference ASIC named Jalapeño to run internal model workloads. The company designed the chip to improve system efficiency and secure its proprietary research from external silicon vendors.
Why it matters
By moving to custom silicon, OpenAI gains control over its infrastructure and reduces reliance on commercial hardware providers. This strategy aims to optimize performance for large-scale models while keeping proprietary research IP internal.
Engineers deployed models including GPT-OSS, DeepSeek R1, and Kimi K2.5 onto A0 sample chips within two months. The hardware was validated using the InferenceX benchmark platform with 8,000 input and 1,000 output tokens.
The players
OpenAI
An artificial intelligence research organization that develops large language models and associated infrastructure.
Broadcom
A global technology company that designs, develops, and supplies a broad range of semiconductor and infrastructure software solutions.
Nvidia
A multinational technology corporation known for designing graphics processing units and data center accelerators for AI.
The details
OpenAI leveraged codesign of its hardware architecture alongside internal model research to ensure high system efficiency. The company intends to keep its hardware resources fully occupied for internal compute demands for the foreseeable future.
Timeline
OpenAI and Broadcom revealed their hardware partnership in 2025.
OpenAI presented Jalapeño benchmarks at Hot Chips 2026 in August 2026.
An interview detailing the hardware was published on September 28, 2026.
The Tech Race
The development of the Jalapeño ASIC signals a transition toward vertically integrated AI infrastructure among leading model labs. This effort places OpenAI in direct competition with Nvidia Vera Rubin platforms as it seeks to replace third-party chips with custom hardware.
This hardware development will likely accelerate model response times for users by optimizing inference efficiency. Consumers will not see direct changes to their service pricing, but the increased compute throughput will facilitate the deployment of more complex, real-time AI tools.
The takeaway
OpenAI's pivot to custom silicon highlights the immense compute pressures facing leading AI research labs. Other companies may soon follow suit as high-performance AI inference becomes the primary constraint on growth.
Further reading
For more on the industry's shift toward custom hardware, visit the Semiconductors section.
Live Poll
Do you believe proprietary chips developed by AI companies will lead to better consumer technology?
