Broadcom's CEO has officially laid out the production roadmap for OpenAI's custom AI accelerator, 'Jalapeño,' signaling a major shift in the AI hardware landscape.
The announcement confirms that large-scale deployment of Jalapeño will begin in 2027, with the goal of reaching high-volume production in 2028. This is significant because OpenAI has already started testing the chip for inference tasks and plans to use it for customer queries later this year. The conversation in the industry has now shifted from if OpenAI's custom chip will arrive to how fast it can be deployed at scale. This reframes 2028 as the year of full production, not just the first prototype.
So, why is this timeline considered credible? There are three main reasons. First, the supply chain is preparing for it. The bottlenecks in advanced packaging (like TSMC's CoWoS) and high-performance memory (HBM) that plagued the industry in 2026 are expected to ease by 2027. With more suppliers for key components like HBM4 being certified, the risk of production delays is reduced.
Second, the financing is already in place. Broadcom recently partnered with investment giants Apollo and Blackstone to create the AI XPV financing platform. This platform is designed to fund over 20 GW of AI deployments through 2028, providing the necessary capital to move from small-scale testing to massive production. This financial backing removes a major hurdle for such an ambitious project.
Finally, this initiative is a key part of a broader strategy called 'vertical integration.' By designing its own chips, OpenAI aims to gain more control over its technology stack, from software models down to the hardware they run on. This allows for better optimization, potentially leading to faster and more cost-effective AI services in the long run, reducing reliance on third-party chipmakers like NVIDIA. The pieces of this plan have been visible for some time, with initial reports of the partnership emerging in late 2025, and Broadcom's own financial forecasts for 2027 already account for a massive ramp-up in AI chip revenue. In essence, this announcement connects all the dots, providing a clear and well-supported timeline for a powerful new force in the AI hardware market.
- Inference: The process of using a trained AI model to make predictions or generate outputs based on new data. It's what happens when you ask an AI a question.
- HBM (High Bandwidth Memory): A type of high-performance memory crucial for AI accelerators, allowing for faster data access than traditional memory.
- Vertical Integration: A strategy where a company controls multiple stages of its supply chain. Here, it refers to OpenAI designing the hardware its software runs on.