Reports say OpenAI is working with TSMC and Broadcom to make its own AI chip
It is also starting to train its AI with AMD and Nvidia chips.
According to Reuters, OpenAI has given up on its plans to build a network of chip plants, at least for now. The company will instead focus on making chips in-house. Reuters claims that OpenAI and Broadcom have worked together for months to make an AI chip for running models. This chip could be ready as early as 2026.
On the other hand, OpenAI wants to train its models on AMD chips through Microsoft’s Azure cloud platform. Reuters says that OpenAI is looking into other options because of chip shortages and delays, as well as the high cost of training. Previously, the company relied almost entirely on Nvidia GPUs for training.