OpenAI expands AI Infrastructure with Major Broadcom Chip Deal
OpenAI has announced a significant partnership with Broadcom, securing a multi-million chip supply to bolster its generative AI capabilities.This deal marks the third major chip agreement for the ChatGPT creator in recent days, following similar arrangements with Nvidia and AMD.
Unlike standard purchases, the chips provided by Broadcom will be custom-designed specifically for OpenAI’s needs, a collaboration that has been underway for the past 18 months, according to OpenAI CEO Sam Altman. These advanced processors are engineered to handle the parallel processing demands of AI, a departure from the architecture of more traditional chips.
Broadcom is slated to deliver processors with a combined power draw of 10 gigawatts (GW) to OpenAI. While the exact number of individual processors this represents is difficult to quantify,it equates to several million units. Deployment of servers utilizing these chips is expected to begin in late 2026.
Altman clarified that the Broadcom chips will primarily be used for inference - generating responses to user prompts – rather than the more computationally intensive tasks of building or training AI models.
“On several levels, we can say that the construction of AI infrastructures constitutes the largest industrial project in the history of humanity,” Altman stated.
The scale of OpenAI’s chip orders has raised concerns among some analysts, given the company’s current financial situation – burning billions annually and not projected to achieve profitability until 2029. Estimates suggest potential chip orders could reach hundreds of billions of dollars over several years. OpenAI is relying on continued rapid revenue growth and its ability to secure further funding to support these investments.
