Microsoft has been reportedly developing its own chips specifically for artificial intelligence (AI) programs that can be used to train large language models (LLMs) such as ChatGPT and GPT-4 and process huge amounts of data to generate human-like content.
According to a report from The Information, who cited two people with direct knowledge of the project, said that the Redmond giant has been secretly working on the chip internally codenamed “Athena” – a first-party chip built to power the supercomputers that train AI – since as early as 2019.
This secret project has 300 employees dedicated to it. Currently, the chip is being tested by a small group of Microsoft and OpenAI employees that can be used to power AI projects for OpenAI and its own internal teams, the report added. The plan is to launch Athena as early as 2024 following the success of the AI-powered chatbot, ChatGPT.
At present, Microsoft is using chips made by NVIDIA for its AI services. If Microsoft is able to successfully create its own in-house processor that can train and power AI, it will not only reduce its dependence on NVIDIA for its AI chip supply but also could result in significant savings for the company that can be used to improve the AI its services rely on.
Of late, many well-known tech behemoths, including Amazon, Alphabet (Google), and Meta (Facebook), have developed their own in-house chips for their AI projects.
The Information report notes that Microsoft reportedly doesn’t see its own in-house chips, as a “broad replacement for NVIDIA’s products”.
Instead, they are meant to enhance Microsoft’s existing infrastructure. It remains to be seen how Microsoft plans to use NVIDIA GPUs and its own chips for its AI-based services.