According to rumours, Microsoft is developing its own AI chips so that it can train large language models without having to rely too heavily on Nvidia. The Information claims that the company began secretly developing the chips in 2019, and that some Microsoft and OpenAI employees have already had access to them to test how well they work with the most recent large language models, such as GPT-4.
Currently, Nvidia is the leading provider of AI server chips. Businesses are vying for these chips, and predictions indicate that OpenAI will require more than 30,000 of the company’s A100 GPUs for the launch of ChatGPT. The price of Nvidia’s most recent H100 GPUs on eBay exceeds $40,000, demonstrating the high demand for expensive chips that can support the use of AI software.
Microsoft is reportedly looking internally in an effort to cut costs on its AI push, while Nvidia rushes to produce as many as possible to meet demand.
According to reports, Microsoft has sped up work on Athena, a project to create its own AI chips. The Windows maker is reportedly planning to make its AI chips more widely available inside Microsoft and OpenAI as early as next year, although it’s unclear if the company will ever make these chips available to Azure cloud customers. According to reports, the software giant has a roadmap for the chips that includes several additional generations.
Microsoft’s in-house efforts could significantly reduce costs as it pushes to roll out AI-powered features in Bing, Office apps, GitHub, and other places. The company’s own AI chips aren’t expected to be a direct replacement for Nvidia’s, however.
For many years, the redmond based tech giant has also been developing its own ARM-based chips. Bloomberg claimed that Microsoft was considering creating its own ARM-based processors for servers and perhaps even a future Surface device in late 2020. However, Microsoft has collaborated with AMD and Qualcomm to develop custom chips for its Surface Laptop and Surface Pro X devices.
Also Read:
