Microsoft has developed a new supercomputer for artificial intelligence for hundreds of millions of dollars

According to Bloomberg, Microsoft has developed a unique supercomputer by combining tens of thousands of Nvidia graphics processing units (GPUs) on its Azure cloud computing platform. The supercomputer is expected to improve the performance of OpenAI’s ChatGPT artificial intelligence.

Microsoft’s Executive Vice President of Artificial Intelligence and Cloud Technologies Scott Guthrie did not name the specific cost of the project, but said that «the price of the supercomputer exceeds several hundred million dollars.» From now on, OpenAI specialists will be able to train more powerful models, and new ChatGPT capabilities will be «unlocked.»

«We have created a system that can work and be reliable on a very large scale. There will be many, many more developments in the future,» emphasized Microsoft’s general manager of Azure AI infrastructure Nidhi Chappell.

Microsoft is already working to make Azure’s capabilities even more powerful with the launch of its new virtual machines that use Nvidia H100 and A100 Tensor Core GPUs. According to Microsoft, this should allow OpenAI and other companies that rely on Azure to train larger and more complex AI models.

The update of the most famous artificial intelligence GPT-4 will be released on March 16. Earlier, it was announced by CTO of Microsoft Germany Andreas Braun. «GPT-4 will include multimodal models that will offer users completely different capabilities, such as video ,» Brown said. He called the updated language model «a game changer». Brown emphasized that the technology has advanced so far that it «works in almost all languages.» For example, a user can ask a chatbot a question in German and get an answer in Italian.