Microsoft looks to have entered the race to introduce custom chips designed to power computationally intensive artificial intelligence workloads to its public cloud, based on job postings reported this week.
Business news network CNBC spotted three listings in March seeking chip designers in Microsoft's Azure Cloud division, and another in April for a software/hardware engineer to work on design and optimisation for AI acceleration.
AI developers increasingly rely on graphics processors for computationally intensive tasks like machine learning, and the three largest cloud providers: Amazon Web Services, Microsoft and Google, offer instances powered by NVIDIA GPUs. Now it appears all three are also betting next-generation AI workloads will demand even-more-specialised processors.
Google kicked off the custom-chip trend in 2016, when it introduced the tensor processing unit (TPU) at its Google I/O event. That chip is now in its third iteration of development, and available to provision on Google Cloud Platform to accelerate software built with the TensorFlow framework.
Amazon reportedly is also working on custom chips to allow its Echo home assistants to do more processing on the device, and reduce requests to the cloud.
Apple relies on custom-built chips from the likes of ARM for some of its smart devices, but has also started developing chips in-house for the iPhone.
More-traditional hardware developers are also working on next-generation chips and servers with features geared to enable AI.
IBM in December released a new server line, AC922, that speeds connectivity between CPUs and GPUs for computationally intensive tasks like training machine learning models and advanced analytics. Big Blue concluded that faster internal bus between embedded devices was key to accelerating AI processes.
Intel, particularly, has a lot riding on the market trend toward processors tuned for AI computation.
Last year, the chip giant introduced its Artificial Intelligence Products Group to build out a portfolio that aligns with trends for AI adoption. Intel also has created an applied AI research lab to explore architectural and algorithmic approaches for future generations of artificial intelligence.