On October 13, 2025, OpenAI announced a strategic partnership with Broadcom to jointly develop and deploy up to 10 gigawatts of AI accelerators and network infrastructure. According to the official announcement, OpenAI will be responsible for the accelerator's architecture and system design, while Broadcom will provide hardware manufacturing and underlying communication solutions such as Ethernet, PCIe, and optical interconnects. The project is expected to begin deployment in the second half of 2026, with full implementation planned for 2029.
OpenAI stated that this collaboration aims to achieve a deep integration of model development and hardware, improving the energy efficiency of large-scale training and inference through its own accelerators. Broadcom emphasized that this collaboration will promote the convergence of standard Ethernet and customized AI hardware, providing a reference for future data center construction. Currently, the accelerator's specific technical specifications, energy consumption parameters, or commercialization plans have not been officially announced. Industry analysts believe that this move may mark a new stage in the integrated collaboration of "model-chip-network" in AI infrastructure.
Frequently Asked Questions
Q: Why did OpenAI partner with Broadcom?
A: OpenAI hopes to improve the performance and energy efficiency of its own accelerators and reduce its dependence on general-purpose GPUs by jointly designing hardware with Broadcom.
Q: What does the so-called “10 GW scale” mean?
A: This scale refers to the overall computing power deployment target, involving data centers in multiple locations and tens of thousands of acceleration chips. It is a large-scale AI infrastructure plan on a global scale.
Q: When is the project expected to be completed?
A: According to the official announcement, the deployment will start in the second half of 2026, with the goal of being fully completed by 2029.
Q: What is Broadcom’s main responsibility?
A: Broadcom is responsible for accelerator manufacturing and network interconnection layer construction, including Ethernet, PCIe and optical communication modules.
Q: What impact will this collaboration have on the AI industry?
A: If the plan is successfully implemented, OpenAI may gain advantages in hardware autonomy and energy efficiency optimization, affecting the AI computing power ecosystem.
