Microsoft’s Custom Silicon at Microsoft Ignite 2023

By Patrick Moorhead - November 20, 2023

The Six Five team discusses Microsoft’s Custom Silicon at Microsoft Ignite 2023.

If you are interested in watching the full episode you can check it out here.

Disclaimer: The Six Five Webcast is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we ask that you do not treat us as such.


Patrick Moorhead: Yeah, it’s a big week there, and probably four or five years, Dan, I was always questioning how could Microsoft keep moving from a performance standpoint, less of a performance standpoint and more of a cost standpoint, and compete with AWS who has been doing first-party silicon for the data center for close to a decade. And we did get our answer there. And Net-net, there’s an accelerator for AI inference and training called Maya, and then there is a CPU, an ARM-based CPU, called Cobalt. You can go and read all the details. I’ve had over 100,000 people read my stuff on X and LinkedIn. I’ll eventually get it up on my own website, maybe Forbes, I don’t know, but it’s been a very provocative type of thing.

I’m going to try to keep this strategic versus geek. Now there is some information that we don’t know yet, for instance, how does this compete performance-wise versus AMD and Intel on the merchant side, but I think, more importantly, versus AWS and Google on the custom side. Microsoft, to me, I did see, under NDA, some performance figures and I did think they were compelling. Those aren’t public yet, but I’m looking forward to that. From a rollout standpoint, Microsoft is focusing Maya on Copilot. I consider that a SaaS product, and then Azure OpenAI Service. Interesting they said Microsoft Copilot or Azure OpenAI service, which I’ll admit, I don’t completely understand the or versus the and there.

Daniel Newman: And?

Patrick Moorhead: Yeah. And one thing that Sam Altman said that leads me to believe that current and maybe some future trillion, 10-trillion-parameter models might be done for OpenAI, as Sam Altman said, that it was a co-collaboration to produce more capable and cheaper models. I talked about performance, and I talked about cost. The CPU used on Maya’s a fourth gen Intel CPU, go Intel. Note that they did not use their own custom ARM or AMD there. Some details on Cobalt, it’s based on ARM Neoverse N2, and N2 is the higher efficiency, not the max performance that we see from other makers. And they said up to 40% improvement performance per core versus the previous ARM server that they have in there right now.

Couple big things that were interesting is I always thought that if Microsoft were going to do that, they were going to part with a custom provider for maybe AMD or Marvell or Samsung Semi. They did not. They told me they went all the way from core IP design to SOC design to tape out and validation on their own all the way to TSMC. How they stealth that, I have no idea, and I don’t consider one information article as a giant leak. I’d said it, what I want in the future is performance data and pricing versus the competitors. I will also want to know what the dates for full IAS support as well as SaaS support for Microsoft 365 and Dynamic 365 and official GA dates. Congratulations, Microsoft. You gave me more than I had expected. You did more than I expected as well as opposed to partnering, and it is game on at this point.

Daniel Newman: Yeah, absolutely. That’s some good insight, Pat. Look, I’ve had a lot of media reach out to me. This was a big moment. They’re asking questions. Is this the take on moment? Is Microsoft going after Merchant silicon? I do think there’s very much an AWS playbook here, and like I said, each company is their own and Microsoft doing some really innovative things with OpenAI. It’s not a one good, one bad or one copying, but I’m saying there’s some very valuable lessons that could be learned from the past near decade that AWS has spent building its own silicon in the hyperscale cloud and offering that to their customers. This is something that’s been a long time coming. People that are surprised, I am asking them why? Literally there’s been a thousand stories about this particular thing happening and now we know what’s happening and at what pace it is happening.

This is an opportune moment. I mean, let’s be very clear about what’s going on here. Vertical integration is the winning formula, especially for Wall Street. Apple made this model very evidently successful by in the Tim Cook era, verticalizing everything in its supply chain. Now again, we’ve seen there’s been vulnerabilities there if you can’t handle best in breed, but in things like an AI ASIC, this is very achievable. It’s been very achievable, there’s many building it and Microsoft is more than capable to do this. You look at a core compute like what they’re doing with Cobalt, very achievable. The refactoring of workloads for ARM, the optimizing workloads for ARM has been going on for some time now. There was a period of time where this was a much heavier lift. The last several years, AWS has really paved the way to make these workloads run more and more efficiently.

And let’s be very candid about what this is. This is economics. I basically chalk this up to three things. I said it’s performance and optimizing performance of workloads on Azure. Two, it’s power and efficiency. It’s a sustainability story and an efficiency story. Every company’s looking to get more efficient. And three, it’s economics. Those two things drive economics. It’s spend less from a standpoint of Microsoft and potentially offer customers to spend less, and that’s how they spend less net. The idea is they keep raising up their bill, but more and more of that bill is going a hundred percent to Microsoft.

Does this mean they’re going to run away from Merchant silicon? Absolutely not. You saw actually at the event they made announcements of Merchant silicon, AMD and NVIDIA were on display. They will continue to be a general store for consumption. So if you want Intel, you’ll get Intel. You want NVIDIA, you get NVIDIA. You want AMD, you’ll get AMD. This is Microsoft basically saying we have an answer. Choose your own adventure silicon wise, but if you want highest performance optimized for Azure, optimized for the workloads that we’re building around open AI and you want to do it potentially at a lower price while being more cost-efficient, they’re putting their hand up and saying, we’re going to play the game. It’s going to be a lot of fun, Pat to watch where this goes next.

Patrick Moorhead
+ posts

Patrick founded the firm based on his real-world world technology experiences with the understanding of what he wasn’t getting from analysts and consultants. Ten years later, Patrick is ranked #1 among technology industry analysts in terms of “power” (ARInsights)  in “press citations” (Apollo Research). Moorhead is a contributor at Forbes and frequently appears on CNBC. He is a broad-based analyst covering a wide variety of topics including the cloud, enterprise SaaS, collaboration, client computing, and semiconductors. He has 30 years of experience including 15 years of executive experience at high tech companies (NCR, AT&T, Compaq, now HP, and AMD) leading strategy, product management, product marketing, and corporate marketing, including three industry board appointments.