In the era of rapid development of artificial intelligence, AI chips, as the core driving force, have attracted much attention for their technological innovation and industrial dynamics. Recently, according to industry sources, OpenAI and Broadcom are working together to complete the design of the first custom AI processor in the next few months, and mass production is expected to be achieved in 2026.
Background of cooperation
Currently, NVIDIA dominates the AI chip market, and its GPU products are widely used in AI model training and inference. However, as AI technology continues to evolve, so do enterprise requirements for chip performance, cost, and autonomy. As a leader in the field of artificial intelligence, OpenAI decided to devote itself to the development of customized AI processors in order to enhance its bargaining power in the supply chain and reduce its dependence on a single supplier. The partnership with Broadcom is a key step for OpenAI to achieve this goal. By integrating the technical advantages and resources of both parties, it is expected to create an AI processor that is more in line with OpenAI's own needs.
Technical features: pulsating array and advanced process
This custom AI processor features a pulsating array design that offers significant advantages when performing matrix or vector computations. Pulsating arrays are able to efficiently process large amounts of data, enabling fast and accurate calculations through the flow of data through the array and the interworking of computing units. In terms of memory, it will use HBM memory, although it has not yet been determined whether it will be HBM3E or HBM4, but the high bandwidth characteristics of HBM memory can meet the strict requirements of AI computing for data transfer speed. In terms of process technology, the processor aims at TSMC's mature N3 series (3nm-level) manufacturing process, and the advanced process technology will bring higher integration and lower power consumption, further improving the performance of the processor.
Development team: professional talent gathering
OpenAI has assembled a professional development team, led by Richard Ho, who has worked on Google's TPU project. The team has recently doubled in size to 40 people, and while still small compared to Amazon Web Services or Google's teams, the team members have a wealth of industry experience and expertise. In the partnership with Broadcom, OpenAI is responsible for developing key differentiated intellectual property, while Broadcom is responsible for adding the rest of the components such as general-purpose CPU cores, memory, I/O controllers, and physical interfaces to the final design, drawing on its deep experience in chip design. This mode of division of labor and cooperation gives full play to the advantages of both parties and provides a guarantee for the smooth progress of the project.
Figure: OpenAI and Broadcom work together to customize AI chips (Source: OpenAI)
Industry impact: Break the pattern and drive innovation
The cooperation between OpenAI and Broadcom will have a profound impact on the AI chip industry. On the one hand, it is expected to break NVIDIA's monopoly in the AI chip market and bring more competition and choices to the market. Other companies may be inspired to accelerate the pace of developing their own custom AI processors, thereby driving technological innovation and development across the industry. On the other hand, the emergence of customized processors will make AI applications more suitable for the personalized needs of enterprises, improve computing efficiency, and reduce costs. This is of great significance for promoting the wide application of artificial intelligence in various fields.
Future Trend: Continuous Optimization and Widespread Application
From the perspective of future trends, OpenAI is expected to gradually optimize and upgrade this custom AI processor after its successful launch. With the continuous advancement of technology and the change of application requirements, the performance of the processor will continue to improve, and the functions will be more perfect. In terms of applications, the first processor is mainly used for inference workloads of AI models, and is expected to expand into the field of training in the future. With the in-depth application of AI technology in healthcare, finance, transportation, and other fields, customized AI processors will play an important role in these fields to help various industries achieve intelligent transformation. At the same time, as more companies join the development of custom AI processors, the market competition will become more intense, which will promote continuous innovation in chip technology and push the entire AI industry to new heights.
OpenAI's cooperation with Broadcom to customize AI processors is an important attempt in the field of AI chips. It not only reflects the active exploration of enterprises in technological innovation and supply chain management, but also brings new opportunities and challenges to the development of the AI chip industry. We look forward to the smooth launch of this custom AI processor and play an important role in future AI applications, injecting new impetus into the development of artificial intelligence."
