The company has been secretly developing chips since 2019. According to the source, some Microsoft and OpenAI staff already have access to these to test how well they work with the latest major language models like GPT-4. The technology will be used to train future versions of language models and neural networks. Generative AI requires a lot of resources. processing large amounts of data, pattern recognition and text rendering to simulate human speech.

Goodbye to NVIDIA

Currently, NVIDIA is a major provider of computing accelerators for AI. OpenAI is estimated to need more than 30,000 NVIDIA A100 accelerators to commercialize ChatGPT. The latest NVIDIA H100 GPUs sell for more than $40,000 on eBay, demonstrating the demand for high-performance chips that can help deploy AI software.

Microsoft, its the chip will outperform solutions from other manufacturers, saving time and money in expensive artificial intelligence systems. Athena’s cost must be at least 30% lower. The first version of the chip will likely use TSMC’s 5 nanometer process. Other big tech companies, including Amazon and Google, are also making their own neural network chips.

It is unclear whether Microsoft will offer these to Azure cloud customers. The company reportedly plans to expand the use of AI chips within Microsoft and OpenAI as early as next year. There is also information about what Microsoft has a big plan for the future that spans several generations.

Athena is being developed by multiple people. 300 employeesand the creation of the chip has accelerated significantly against the background of the popularity of chatbots with artificial intelligence. According to The Information, Athena processors may become available in early 2019. in 2024.