Broadcom and OpenAI unveil Jalapeño, a custom chip aimed at speeding large language model inference

Broadcom (AVGO) and OpenAI have introduced Jalapeño, OpenAI’s first “intelligence processor,” a custom chip intended to speed up inference for large language models (LLMs). Inference is the stage where an AI model generates responses, as opposed to training the model, and it is where compute demand can be relentless as services scale.

According to reporting that cited the companies’ announcements, Jalapeño is designed specifically for OpenAI’s large-scale inference needs. The launch is being framed as a step forward in a broader, long-horizon semiconductor effort between the two companies. The same reporting characterized the OpenAI-Broadcom initiative as a major growth lever for Broadcom’s custom chip business over the coming years, while noting the timeline between chip design milestones and meaningful revenue can stretch out.

The wider deal context is that Broadcom and OpenAI have been working toward large-volume deployment, sometimes described in terms of an electricity-consuming data center scale. In a related OpenAI product announcement, OpenAI and Broadcom said Jalapeño is built to improve performance and efficiency for inference and to support the scale required for modern AI workloads.

Market-facing analysis around the initiative suggests Broadcom does not expect substantial revenue from this specific OpenAI chip effort immediately. One reason is that custom ASIC-style development, testing, and deployment take time before production shipments flow. The reporting also highlighted that investors may be recalibrating expectations as the companies move from prototypes and design iterations toward customer deployment.

For Broadcom, Jalapeño reinforces its strategy of taking a systems role in AI infrastructure, not just selling standard components. Broadcom’s AI semiconductor bet is closely tied to customers’ willingness to adopt custom chips that can offer efficiency advantages compared with more general-purpose hardware. If OpenAI’s internal deployments expand, Broadcom’s custom design work could translate into large, repeatable revenue streams tied to production runs.

Even so, important details remain opaque outside of what the companies choose to disclose. The publicly available product and media coverage in this instance does not lay out, in the reporting consulted, the exact unit volumes, pricing, or near-term financial guidance specifically attributable to Jalapeño. That means investors are still likely to look to later disclosures, including Broadcom’s segment commentary and any operational updates from OpenAI, to assess how quickly the chip moves from “unveiled” to “shipped at scale.”

The next watch items are straightforward: whether Broadcom and OpenAI provide further information about deployment timelines, performance metrics in production use, and how the chip’s rollout maps onto Broadcom’s forecast for AI-related revenue. Analysts will also likely track whether other major AI labs follow similar pathways toward custom inference accelerators, because once those designs are productionized, the economics of AI compute can become more favorable for the chipmakers positioned for scale.