OpenAI and Broadcom have concluded a large-scale strategic partnership to develop and implement specialized AI chips and computing systems with a power of 10 gigawatts - enough to run the largest data centers and support the most advanced neural networks such as "ChatGPT" and "Sora".
The essence of the partnership: OpenAI develops the architecture of the chips with integrated experience from leading AI models directly into the hardware, and Broadcom takes over the production and implementation - using its Ethernet solutions for optimization and scalability.
The system will be fully scalable and designed for OpenAI's own and partner centers - the start is in 2026, and by the end of the decade it will cover the infrastructure of global AI. The power of 10 GW is comparable to the consumption of a large city.
OpenAI aims to reduce its dependence on NVIDIA and AMD in order to obtain an independent and budget-friendly technology stream and integrate advanced developments directly into the new chips.
Broadcom shares jumped 9% after the announcement, and analysts estimate the deal at tens of billions of dollars - the technological rebound is already being felt in the markets.
OpenAI CEO Sam Altman commented: "Creating our own accelerators is the decisive step towards the infrastructure needed for the potential of AI and the real benefits for business and people." Broadcom defines the partnership as a turning point for the industry - the new accelerators and network solutions will become the basis for the development of super-intelligent models.
OpenAI and Broadcom are positioning themselves in the global technological landscape - the new chips will reduce computing costs and open up new horizons for scalable AI with a real impact on the economy and society.