OpenAI has unveiled Jalapeño, its first custom AI chip developed in partnership with Broadcom [1].
The move signals a strategic shift for the AI company as it seeks to reduce reliance on third-party hardware providers to meet escalating compute demands. By designing its own silicon, OpenAI can optimize hardware specifically for its large-scale models, potentially lowering costs and increasing speed.
Greg Brockman, president of OpenAI, discussed the new hardware during a broadcast of CNBC’s ‘Squawk on the Street’ on June 24, 2026 [1]. He said the Jalapeño chip provides a real performance improvement for the company's inference systems [1].
Brockman said the hardware specifically targets the efficiency of how models generate responses. "The Jalapeño chip gives us a tangible boost in inference throughput and latency," Brockman said [2].
The unveiling comes eight months after the company first announced a deal to develop custom chips [1]. During the interview, Brockman said broader industry questions regarding whether the current AI market is a bubble and how the company plans to manage its massive compute needs [1].
Custom silicon allows companies to tailor the architecture to the specific mathematical operations required by their software. This partnership with Broadcom enables OpenAI to move from general-purpose GPUs to a more specialized environment, a trend currently being mirrored by other major tech firms seeking hardware independence [2].
“"This is a real performance improvement."”
The development of the Jalapeño chip indicates that OpenAI is moving toward vertical integration. By controlling the silicon layer, OpenAI can optimize the relationship between its software and hardware, which is critical for reducing the latency of AI responses and managing the immense energy and financial costs of running global inference systems.
