It was inevitable and expected, but we now have a new power player in the ever-expanding semiconductor landscape: Microsoft. It announced its Cobalt 100 CPU and Maia AI accelerator at its Ignite conference this week. In simpler times, we had Intel and AMD playing leapfrog with their competing x86 CPUs. It wasn’t a very exciting time. All that changed — almost stealthily, until the silence grew into a cacophony of chipmakers, architecture variants, and geopolitical tensions over the past few years. It’s boring no more.
Recent developments saw other tech giants deciding that they can design their own chips. The tectonic event was the 2020 introduction of Apple’s M1 processor. Apple proved that it could replace Intel CPUs with its own, a move that demonstrated superior performance per watt. The floodgates were now open.
Cloud giants such as Google and AWS followed, designing their own processors. Meanwhile, NVIDIA was taking the world by storm with expanded demand for its GPUs to perform machine learning and other AI tasks. A little-known UK firm called Arm emerged as a powerhouse because its ARM architecture lent itself well to the new generation of CPUs, including Apple’s M-series. Previously the choice for smartphones and other low-power devices, ARM (the architecture from “Arm,” the company) became a viable contender for high-end CPUs in data centers and PCs. It was only a matter of time before Microsoft jumped in … and now it has.
Microsoft’s move erodes lucrative business for Intel, AMD, and NVIDIA, all of whom are key suppliers to Azure. All three will remain part of Azure’s menu of customer options, but the main growth for Azure will come on its own chips. This is especially impactful for NVIDIA, since Microsoft is focusing so much of its Azure services on generative AI, which currently depends on huge quantities of NVIDIA GPUs.
Democratized Processor Design Presents Broad Choice
The new Cobalt 100 instances will offer superior price performance for some workloads. Maia will make Microsoft’s Copilot services more powerful (and maybe cheaper, but that’s TBD). We advise enterprise application developers and cloud decision-makers to evaluate these new offerings for new applications or for migrating existing applications. After years of force-fitting workloads into one or two instance types, cloud providers can now tailor instances to workload type and/or cost efficiency when that’s a higher priority than sheer power.
Custom processors are inevitable. As great as Intel’s and AMD’s x86 processors continue to be, they are general-purpose devices — a jack of all trades and master of a few. Although superior for AI needs, even GPUs like NVIDIA’s are also general-purpose chips. Don’t even think of kissing x86 goodbye. It is here to stay for a long time, and it keeps getting better, as AMD and Intel are adding new capabilities and significantly improving power efficiency. We expect ARM to gain even more momentum, though. Although ARM is also fundamentally general-purpose, the building blocks that Arm provides allow its partners great flexibility in customizing their designs.
Arm is making it easier to build chips on the ARM architecture. Qualcomm, NVIDIA, Samsung, and plenty of other chipmakers already provide ARM-based processors, and they’re starting to work their way into hardware for corporate data centers, end users, and consumers. NVIDIA’s much-hyped Grace Hopper processor teams an ARM-based CPU with its GPU into a formidable choice for AI applications. Dell now offers a Windows PC based on Qualcomm’s Snapdragon CPU. We expect it to be popular, as its power consumption (and resulting heat dissipation) and pricing make it an attractive alternative to Chromebooks.
Democratized Silicon Brings Power To The People
All of this means that tech buyers have more choice even beyond cloud. In fact, today’s options come from the tech giants, but the trend will extend to enterprises that don’t currently identify as tech companies. Chip design will get easier as chipmakers and software companies evolve design automation in a manner similar to low-code application development. Low-code empowers a far richer community for creating digital business value. Our friend Diego Lo Giudice introduced TuringBots, which now extend this concept whereby generative AI can assist in code development. The ideas and much of the same technology will allow you to software-define custom processors even if you don’t have a Ph.D. in electrical engineering.
“Power to the people” is taking on a profound new meaning in tech. John Lennon would be proud. Microsoft’s silicon announcements represent one more step forward in that democratization pursuit. Embrace the power and enjoy the ride!
Caution: Weigh The Advantages And Risks
Today’s hyperscalers are no longer just cloud providers — they are tech juggernauts. You see clear signs of vertical integration from chips to software to artificial intelligence. While it makes technology choices easier for many, it rings alarm bells for many others, whether that is lock-in, being too big to fail, or concentration risk. Weigh your risks and advantages before you enthusiastically embrace one hyperscaler and put all your eggs in one basket.