Tomohiro Ohsumi/Getty Images
OpenAI is considering building its own AI chips to power ChatGPT, in the latest sign that the company is diverging from its partner Microsoft, according to a report from Reuters.
OpenAI is exploring plans to build its own AI chips as it faces a shortage of the previous technology behind the AI revolution. The company, which has not yet committed to the plan, has even discussed acquiring a chipmaker to help with what would be an enormously expensive and time-consuming process, Reuters said.
AI chips such as Nvidia's H100 are the most valuable resource in tech right now, as they are crucial for training the large language models that power the likes of ChatGPT.
The mad dash to acquire as many as possible has led to a global shortage, with many tech giants like Meta and Microsoft now seeking to develop their own chips as a result.
If OpenAI was to join them, it could help bring down the eye-watering cost of running ChatGPT, which one analyst estimated at up to $700,000 a day earlier this year.
The reported move is also further evidence of divergence between OpenAI and Microsoft, with the Information reporting that Microsoft is trying to reduce its reliance on OpenAI by developing its own in-house large language models.
Microsoft invested $10 billion in OpenAI at the start of 2023. The company has provided OpenAI with the enormous computing power needed to train its AI models. In 2020, Microsoft built the startup a huge supercomputer that uses 10,000 advanced Nvidia GPUs. In return, it has been able to integrate OpenAI's models into its own products, such as Bing.
However, unnamed sources told the Information that Microsoft has instructed researchers to build smaller and cheaper alternatives to OpenAI's GPT models after growing concerned over the spiralling cost of more expensive models like ChatGPT.
OpenAI did not immediately respond to a request for comment from Insider, made outside normal US working hours.