Meet Jalapeno, OpenAI's custom AI chip designed with Broadcom
What's the story
OpenAI has unveiled its first custom artificial intelligence (AI) chip, named Jalapeno. The groundbreaking processor was developed in collaboration with Broadcom and is designed to enhance the tech giant's infrastructure. The move comes as part of a broader effort by AI labs like OpenAI and Anthropic, who are struggling with computing power shortages for their advanced chatbots and coding apps.
Chip development
Jalapeno designed for specific AI task called inference
The Jalapeno chip has been specifically designed for a particular AI task known as inference, where data is processed to respond to user queries on chatbots like ChatGPT. OpenAI's engineers collaborated with Broadcom in creating this innovative processor. Broadcom CEO Hock Tan said that the performance of the new chip matches NVIDIA's Blackwell chips and Alphabet's Google-designed Tensor processing units.
Performance assurance
Chip optimized for speed and efficiency with LLMs
Richard Ho, OpenAI's hardware chief, said that the Jalapeno processor is optimized for speed and efficiency with large language models (LLMs) that power many AI apps. He added that it will perform well on all future iterations of these models. OpenAI plans to deploy the new chip by year-end as part of a multi-generation chip development strategy.
Manufacturing
Celestica to build server systems for Jalapeno
Canadian electronics manufacturer Celestica will build the server systems for the Jalapeno chip. These systems will be exclusively used by OpenAI, just like the chips themselves. The company has already tested samples of the new processor in its labs with its GPT-5.3-Codex-Spark AI model, achieving target power and performance levels during these tests.
Production timeline
Design of the chip took around 9 months to complete
The design of the Jalapeno chip took OpenAI's engineers around nine months to complete. They then sent it for manufacturing to Taiwan's TSMC, partly by using AI to speed up certain parts of the process. This is a major step in the tech giant's journey toward developing its own in-house chips, a trend also seen at Meta Platforms, Amazon, and Google.