On October 27th, it is reported that OpenAI is planning to purchase artificial intelligence chips independently developed by Google in order to expand its computing power and diversify its chip supply sources. This potential transaction involves Google's Tensor Processing Unit (TPU), but the final terms of the agreement have not yet been determined by both parties.

With the continuous growth in the training and inference demands for the AI model, OpenAI is seeking sources of computing power other than NVIDIA GPU. At the same time, it is collaborating with Broadcom to develop a self-developed AI chip codenamed “Jalapeno” to further enhance its computing power supply system.

I. OpenAI plans to introduce Google TPU to expand the training and inference computing power of AI.

According to people familiar with the matter, OpenAI is considering purchasing Google's TPU chips to integrate them into its existing artificial intelligence computing infrastructure. The specific scale of the potential transaction, the purchase price, and the final terms have not yet been determined.

TPU is a processor specifically developed by Google for machine learning workloads. Similar to NVIDIA GPU, it can be used for the training and inference of AI models. As the computational demands for large AI models continue to increase, TPU is gradually becoming an important alternative to NVIDIA GPU.

For OpenAI, introducing Google's TPU helps to increase the sources of available computing power, reduce dependence on a single chip supplier, and provide more hardware options for different AI workloads. However, whether the expected performance and cost advantages can be achieved still depends on chip deployment, software adaptation, and actual operational efficiency.

II. Collaborated with Broadcom to develop a self-developed chip, codenamed Jalapeno.

In addition to planning to purchase Google's TPU, OpenAI is also collaborating with Broadcom (Broadcom) to develop its own AI chip codenamed Jalapeno. The goal is to start providing computing support for its AI models later in 2026.

Unlike purchasing off-the-shelf GPU or TPU, custom-designed chips can be tailored to meet the computational requirements of the OpenAI model, thereby optimizing performance, power consumption, and operating costs under specific workloads. However, custom-designed chips still require development, manufacturing, and deployment processes, making it difficult for them to completely replace existing computing infrastructure in the short term.

OpenAI simultaneously promotes external procurement and independent chip research and development, reflecting that large AI enterprises are expanding the supply of computing power through various technical routes to meet the continuously growing demands for model training and inference.

III. Google Expands TPU External Sales, AI Chip Market Competition Intensifies Further

This potential collaboration also reflects a strategic shift in Google's chip business. In the past, Google's chips mainly served its own products and the Google Cloud platform, but now the company is pushing these chips into a broader external market.

If OpenAI ultimately reaches a procurement agreement with Google, it will provide Google TPU with new major customers, which also means that NVIDIA faces further competition from Google's self-developed chips in the AI computing market.

However, OpenAI expanding the sources of chip procurement does not mean that the use of NVIDIA GPU will be stopped. As the demand for computing power from AI continues to grow, products from different suppliers may play a role together in model training, inference, and other computational tasks.

The core of this message is that OpenAI is simultaneously advancing three technical routes: NVIDIA GPU, Google's TPU, and a self-developed chip in collaboration with Broadcom. Meanwhile, Google is also pushing TPU from its own cloud computing ecosystem to external customer markets.