OpenAI and Broadcom Debut Jalapeño Chip: 9-Month Design, Gigawatt-Scale Deployment by 2026

OpenAI and Broadcom have introduced Jalapeño, which's the first custom-designed Intelligence Processor made by the people who created ChatGPT. This chip is a part of OpenAIs plan to control everything behind its models and products from the software and models to the actual silicon chips.

Broadcom CEO Hock Tan said, "This is the start of a long-term plan that will have many stages." He also said, "By working with OpenAI to make our industry-leading silicon we can set up huge data centers with Microsoft and other partners starting in 2026."

Jalapeño is made specifically for Large Language Model Inference, not for use. It is not like AI accelerators that were adapted from older technology. Instead Jalapeño was designed from scratch to work well with large language models. The design is based on the systems that OpenAI uses every day for ChatGPT, Codex and other products.

Tests have shown that the chip will work better than current state-of-the-art technology according to the company. The architecture is designed to reduce the amount of data that needs to be moved and to balance the use of computing, memory and networking resources. This helps the chip work closer to its potential.

The design and manufacturing of Jalapeño took nine months, which is a record for high-performance semiconductors. This was possible because of the collaboration between the software and hardware teams and the use of OpenAIs own models to speed up parts of the design process. OpenAI President Greg Brockman said, "We were surprised by how our models were able to accelerate the process."

Samples of the chip are already being tested in the lab. Are running machine learning workloads at the target frequency and power. This includes GPT-5.3-Codex-Spark.

Jalapeño will help OpenAI make progress by providing infrastructure, which will make computing more efficient. This will allow for training and serving of models which will lead to more capable models, better products and more revenue to invest in the next generation of infrastructure.

The chip is designed to be flexible and work with all Large Language Models. It combines the power and speed of AI accelerators with the low latency of specialized inference systems. The companies are calling it an Intelligence Processor and the first AI accelerator in a multi-generation compute platform.

Broadcom CEO Hock Tan said that after some small prototype development in 2026 the chip will start being produced in larger quantities in 2027 and will be widely available in the first half of 2028. OpenAI is also working with companies, such as AWS, AMD and Cerebras to develop silicon chips.

One report noted, "When a company controls the application layer, the middle model layer and the bottom silicon architecture that's what we call a full-stack advantage." Jalapeño is OpenAIs step towards achieving this full-stack vision. OpenAI is making Jalapeño to have control, over its products and models. OpenAI wants to use Jalapeño to make its models and products. Jalapeño is a part of OpenAIs plan to make the full stack.