OpenAI and Broadcom announced Jalapeño, the first private artificial intelligence processor developed as a result of their joint work. The new accelerator, which the company describes as “Intelligence Processor”, is designed specifically to run the inference processes of large language models (LLM). OpenAI stated that this processor was developed taking into account the needs of future artificial intelligence models, and announced that the first distribution is planned to start in data centers in late 2026. Thus, the company took one of the first concrete steps towards reducing dependence on external hardware that it has been using for a long time.
OpenAI and Broadcom announced that they would cooperate to develop a private artificial intelligence accelerator in October 2025. Jalapeño, which emerged after the work carried out in the intervening period, is the first processor developed by the two companies together. According to OpenAI, this chip was designed considering the needs of larger and more complex artificial intelligence systems that will be developed in the future, not just for current models. This approach is aimed to provide long-term advantages in terms of both performance and operational efficiency.
Jalapeño comes with claims of high energy efficiency
According to the information shared by OpenAI, Jalapeño aims to significantly exceed the current state-of-the-art solutions in terms of performance per watt. However, the company also emphasizes that the final performance tests for the processor have not been completed yet. Therefore, the information announced today is based on preliminary evaluations. More detailed technical specifications and independent performance results are expected to be shared in a comprehensive technical report to be published in the coming months.
The development process of the processor also contains remarkable details. OpenAI states that the time from Jalapeño’s initial design stage to production readiness was only nine months. The company attributes this short development time to the combination of OpenAI engineers’ work on the software side and Broadcom’s experience in semiconductor design and manufacturing. In addition, it is stated that OpenAI models are also used in certain parts of the design and optimization processes. Thus, it seems that artificial intelligence technology plays an active role in the development of its own hardware.
For OpenAI, Jalapeño’s announcement doesn’t just mean the introduction of a new processor. Until now, the company has been running ChatGPT and other generative AI services largely on NVIDIA’s data center GPUs. Developing its own custom accelerator may enable it to gain more control at the hardware layer. In addition, targets such as reducing extraction costs, reducing energy consumption and using capacity more efficiently are among the main reasons for this approach.
On the other hand, Jalapeño is positioned as the first product of the collaboration between OpenAI and Broadcom. The two companies state that the partnership will not be limited to a single-generation project, but that they aim to create an artificial intelligence computing platform that will be developed over different generations. In this direction, it is planned to develop more powerful processors and increase the scalability of the artificial intelligence infrastructure in the future.
The first Jalapeño processors are aimed to be used in data centers towards the end of 2026. However, the real-world performance, energy efficiency and cost advantage of the processor will only become clear after large-scale deployments. In addition, OpenAI’s development of its own artificial intelligence hardware is among the developments that show that competition in the industry is accelerating not only in software but also in special processor design. Especially today, when inference costs have become a determining factor in artificial intelligence services, such private accelerators are expected to play a more widespread role in the coming years.