OpenAI teams up with Broadcom to produce its own AI chips

Following the announcement of multi-billion dollar contracts with NVIDIA and AMD, OpenAI, a leading AI developer, confirmed a strategic partnership with Broadcom to develop and manufacture its own AI chips.

What is known

The produced accelerators will be used to scale computing clusters needed for training and launching next-generation AI models.

OpenAI aims to create a more efficient and scalable AI infrastructure, and its own chips will allow for deeper integration of model architecture into hardware, improving energy efficiency and speeding up training.

Key points of collaboration:

• OpenAI is developing its own AI chips to optimize performance and reduce dependency on third-party suppliers;
• Broadcom will become a key partner providing infrastructure, including networking solutions and optical technologies;
• The first deliveries of accelerators are expected in the second half of 2026, with full deployment to be completed by the end of 2029.

The project will cover OpenAI data centers and partner sites worldwide.

The partnership with Broadcom will allow OpenAI to deploy additional infrastructure with 10 GW capacity — this enormous figure is comparable to the power generation of ten nuclear reactors.

Source: OpenAI