In the rapid development of the artificial intelligence field, OpenAI has reached an important cooperation agreement with Broadcom to jointly develop an AI accelerator with a 10-kilowatt output capacity. This collaboration demonstrates OpenAI's further layout in hardware technology, aiming to enhance its computing power to meet the growing market demand.
According to the agreement, OpenAI will be responsible for the chip design of the new accelerator, while Broadcom will provide key network components, including Ethernet and PCIe technologies. This division of labor not only improves development efficiency but also ensures the performance and compatibility of the new device.
The new AI accelerator is scheduled to start shipping in mid-2026, with installation expected to be completed by the end of 2029 in OpenAI's and its partners' data centers. The total value of this project is approximately $10 billion, indicating OpenAI's ambition in pursuing technological advancement and market competitiveness.
Notably, this collaboration is not OpenAI's first attempt in the hardware field. Previously, OpenAI had also reached an agreement with AMD to provide 6-kilowatt GPU capacity. These series of collaborations aim to reduce OpenAI's reliance on Nvidia, which currently leads the generative AI market and has become an industry benchmark.
Through collaborations with Broadcom and AMD, OpenAI is building a more diversified hardware ecosystem, which will provide more flexibility and choices for it in the fierce market competition. Such a strategic layout not only accelerates technological upgrades but also opens up broader possibilities for future AI application scenarios.
Key points:
🔧 OpenAI and Broadcom are collaborating to develop a 10-kilowatt AI accelerator, expected to ship in 2026.
🔌 OpenAI is responsible for chip design, while Broadcom provides key network components, with installation expected to be completed by the end of 2029.
💵 The cooperation agreement is worth approximately $10 billion, aiming to reduce dependence on Nvidia.