OpenAI reveals its first AI processor: Jalapeño

OpenAI has disclosed its first custom AI processor, an inference chip called Jalapeño developed with Broadcom, as the company moves to reduce its dependence on Nvidia GPUs [1]. Jalapeño is an application-specific integrated circuit, or ASIC, built to handle AI inference — the process by which models such as ChatGPT generate responses to user queries [1]. The disclosure arrives nine months after OpenAI first confirmed it would collaborate with Broadcom on chip design [1]. Broadcom, a Palo Alto-based semiconductor and infrastructure software supplier that surpassed a $2 trillion market capitalization in April 2026, has become a central partner for several large technology firms seeking alternatives to Nvidia hardware [5]. Broadcom CEO Hock Tan told Reuters that Jalapeño matches the performance of Nvidia’s Blackwell chips and Google’s Tensor processing units [1]. OpenAI, headquartered in San Francisco, has built its business on large language models including the GPT series, and its November 2022 release of ChatGPT helped accelerate a global surge in generative AI investment [2]. The organization was founded as a nonprofit in 2015, later adding a for-profit subsidiary, and in 2025 restructured into a public benefit corporation partially controlled by a nonprofit foundation [2]. Microsoft has invested more than $13 billion in OpenAI and supplies Azure cloud computing resources [2]. With Jalapeño, OpenAI joins Microsoft, Meta, and Amazon, each of which has recently introduced custom AI chips for training or inference while still trailing Nvidia on overall performance [1]. OpenAI described the processor as the “first step in a multi-generation compute platform” that it expects to deploy by the end of 2026 [1]. The company said early testing indicates the chip “will deliver performance per watt substantially better than current state-of-the-art,” though final performance measurements are still under way [1]. Broadcom’s semiconductor portfolio serves data center, networking, broadband, wireless, and storage markets, and roughly 58 percent of its revenue came from semiconductor products as of 2025 [5]. The company completed its $69 billion acquisition of VMware in November 2023, further expanding its infrastructure software footprint [5]. OpenAI’s move to custom silicon reflects a broader industry shift as model developers seek to manage the cost and supply constraints of leading-edge GPUs [1].