OpenAI and Broadcom Unveil ‘Jalapeño’ Chip for LLM Inference

The Update

OpenAI and Broadcom have introduced Jalapeño, a custom AI accelerator architected specifically for large language model inference. The chip was designed from the ground up to handle current and future LLM workloads, including OpenAI’s GPT‑5.3‑Codex‑Spark model. Engineering samples are already running production target frequencies and power levels in the lab. The companies plan to deploy the first generation at gigawatt scale with data center partners by the end of this year.

Why It Matters

This partnership marks a significant step in OpenAI’s strategy to build a full-stack platform behind its models and products. By designing the chip internally, OpenAI aims to reduce dependence on external suppliers and improve performance per watt. The move comes as the industry faces a global compute crunch, making efficiency and custom silicon critical for scaling AI systems.

What to Watch

OpenAI has not yet released detailed performance benchmarks. The companies will present a technical report in the coming months. It remains unclear if the chip will be available to third-party developers or remain exclusive to OpenAI’s infrastructure.

Sources

• OpenAI News — Official announcement and technical details about Jalapeño
• Ars Technica — Context on the broader AI silicon race and industry implications