OpenAI Rents Google AI Chips to Power ChatGPT Products

Post by : Bianca Suleiman

Photo: Reuters

OpenAI, the company behind ChatGPT, has started renting artificial intelligence chips from Google to run its products. A source shared this news with Reuters on Friday.

Usually, OpenAI buys Nvidia’s special graphics chips. These chips help train AI models and also help the models think and give answers when people ask questions. This process is called inference.

Earlier this month, Reuters reported that OpenAI planned to use Google Cloud services because it needs more computer power. This is surprising because OpenAI and Google are strong competitors in the AI world.

For Google, this deal is important. Google is now letting other companies use its special AI chips called tensor processing units (TPUs). Before, only Google used them. Because of this, Google has new customers like Apple and AI startups such as Anthropic and Safe Superintelligence. These two startups were started by former leaders of OpenAI and also compete with ChatGPT.

OpenAI also plans to finish designing its own special AI chip this year.

Using Google’s TPUs is a big change for OpenAI. It is the first time it is using AI chips other than Nvidia’s in a big way. It also shows that OpenAI is moving away from only using Microsoft’s data centers. Some reports say Google’s TPUs could be cheaper than Nvidia’s chips.

OpenAI hopes that using TPUs will reduce the cost of running its AI models. However, Google is not giving OpenAI its most powerful TPUs, according to The Information. This was shared by a Google Cloud employee.

When Reuters asked, Google chose not to comment. OpenAI also did not reply immediately.

By adding OpenAI as a customer, Google is using its own AI hardware and software to grow its cloud business faster.