Overview
Broadcom shares rose 3.4% Wednesday morning following OpenAI's announcement of a custom AI accelerator named "Jalapeño." Market participants reacted quickly to the revelation that the chip, described by OpenAI as its first "Intelligence Processor," was created specifically for large language model (LLM) inference and transitioned from design to production in nine months.
Design and production
OpenAI said Jalapeño was developed from the ground up to match its particular model requirements. Broadcom and Celestica are handling implementation, rack integration, and networking tasks for the platform. According to the announcement, engineering samples are already running active machine learning workloads in laboratory environments, including OpenAI's GPT-5.3-Codex-Spark, operating at production-target frequencies and power levels.
Performance and architecture
Early lab testing, as reported in the announcement, indicates the architecture delivers a substantial performance-per-watt advantage compared with current market alternatives. That benefit is attributed to minimizing data movement and optimizing the relationship between compute, memory, and networking within the chip and system design.
Deployment plans and partnerships
OpenAI plans to deploy the Jalapeño processor at gigawatt scale across data center partners beginning in 2026, naming Microsoft among those partners. For its large-scale deployment capabilities, the platform will make use of Broadcom's Tomahawk networking silicon.
Company commentary
Hock Tan, President and CEO of Broadcom, framed the collaboration as a long-term effort to scale the physical infrastructure needed for AI, saying the project represents a commitment to that scaling and that Jalapeño is the start of a multi-generation roadmap.
Market reaction
The immediate market response included the roughly 3.4% uptick in Broadcom shares following the OpenAI reveal. The announcement ties semiconductor design, systems integration, and data center networking together as OpenAI moves toward deploying its own custom hardware at scale.
Note: The article reports only the details provided in the announcement, including lab testing results and deployment intentions. It does not assert outcomes beyond the information presented.