The birth of ChatGPT was not free. Aside from the multi-billion dollar investment in OpenAI, Microsoft has already spent hundreds of millions on the hardware needed to train the neural network.

In January, Microsoft announced it would invest around $10 billion in OpenAI. However, the cooperation between the two companies has lasted much longer. In 2019 Microsoft invested the first billion in OpenAI. Then the infrastructure needed to support OpenAI research also began. In the years leading up to the launch of ChatGPT, Microsoft is said to have spent hundreds of millions of dollars on the hardware needed to develop ChatGPT.

That money went toward building an AI supercomputer within the Azure cloud infrastructure, according to Bloomberg. This supercomputer is made up of tens of thousands of Nvidia GPUs linked together with fast connections. When it was built, Microsoft had no idea what such a system would look like. The virtual supercomputer was a pilot project that required Microsoft to figure out how an HPC installation would work with so many GPUs in the cloud.

HPC architecture for everyone

The result of this is not just ChatGPT. Microsoft eventually developed a model for a cloud supercomputer. Building the virtual HPC system started as a custom project for OpenAI, but had to lead to a more widely applicable product from the start. Today Microsoft announced instances within Azure built around Nvidia H100. These can also be linked together, which basically gives customers the opportunity to build such an Azure AI supercomputer themselves.

OpenAI was thus the first practical user of a cloud supercomputer. With this, Microsoft proves that the architecture works and has a practical use in the context of AI development.