Microsoft wants to spend less money on more powerful chips for AI, so Redmond is developing them itself. The accelerators, codenamed Athena, are already being tested and should be available from next year.

AI is an expensive business because both training and inference require proprietary hardware. After all, CPUs work hand-in-hand with accelerators like Nvidia’s H100 GPUs. Training models like GPT-4 requires tens of thousands of accelerators. Subsequent inference, effectively using a trained model, also requires a lot of large-scale hardware. In this respect, Microsoft wants to take its destiny more into its own hands.

Athena

The manufacturer has therefore been working on its own chip since 2019: Athena. The information knows that. Athena needs to be more powerful than third party chips. In addition, Microsoft does not have to buy the microchip externally, which has a positive effect on the final bill.

After all, Microsoft has big AI plans. Almost the entire range of the company, especially Office, will be integrated with AI functionality in the near future. The search engine Bing is also getting an AI flavor. Every time someone uses such an AI function, the workload has to be handled by suitable hardware.

Complementary to Nvidia

Microsoft doesn’t plan to become the new Nvidia. The chips of this AI specialist remain relevant. Athena is a necessary extra that will support future workloads. The first Athena chips are currently being tested by Microsoft itself together with OpenAI. The goal is to launch the first generation of Athena in 2024. In principle, new generations would then follow.

It is unclear whether Microsoft only wants to use Athena to satisfy its internal hunger for AI or also to serve customers. Finally, cloud competitors AWS and Google offer their users access to AI-specific instances using their own chips. In this respect, it seems more logical that Microsoft will also offer such instances based on Athena in Azure at some point.