Microsoft Is Late to the AI Party. But At Least It's Bringing Its Own Chips.

Microsoft Is Late to the AI Party. But At Least It’s Bringing Its Own Chips.

Microsoft introduces its new "Athena" chip to power its machine learning workloads. This will make the already-crowded AI party even more interesting.

April 18, 2023

There was an excellent report from The Information today about how Microsoft is finally developing its own “Athena” custom AI chip. Athena will power the machine learning workloads for Azure’s cloud computing, while also being used internally for Microsoft’s own AI needs.

An AI-generated image using DALL-E. “An artificial intelligence party with robots dancing in low light and with potato chips in the foreground.”

I say finally because Microsoft has been riding the slowest horse in this AI hardware race. Just about every other hyperscaler has already designed their own AI chips and are commercially manufacturing them. Here’s a look at who’s ahead of them on the track:

Alphabet (NASDAQ: GOOGL): TPU chips for Google Cloud & TensorFlow AI operations. Co-designed by Broadcom and manufactured by Taiwan Semiconductor (NYSE: TSM).
Amazon (NASDAQ: AMZN): Inferentia chips for AWS, specifically for EC2 instances. This is the hidden force behind Alexa. Manufactured by TSMC.
Apple (NASDAQ: AAPL): Neural Engine for ML functions on Smart Devices. The hidden force behind FaceID and Siri. In-house design that expanded upon its A1 chips. Manufactured by TSMC, though continually being wooed by Intel (NASDAQ: INTC).
Facebook/Meta (NASDAQ:META) : Custom ML chips for recommendations and video transcoding on the social network. In-house design in partnership with Qualcomm. Metaverse-specific workloads still primarily being handled by NVIDIA.
Tesla (NASDAQ: TSLA): Dojo ML chips for vehicle FSD computer vision and data center neural networks. Designed in-house and manufactured by TSMC (who won the contract from Samsung in 2022).

While Microsoft may be late, it’s still making quite an entrance. Its investment in OpenAI and its introduction of large language models into its enterprise-embedded applications will serve it well for decades to come.

Now that it has the applications figured out, it can design and create its own in-house chips to optimize its performance. Or potentially save costs, displacing NVIDIA’s (NASDAQ: NVDA) expensive A100 GPUs that it’s currently buying and using to power much of Azure’s computing.

Things are about to get interesting. Cue up the music from the android DJ. It’s time to get this AI party started.

Already a 7investing member? Log in here.