Matt McIlwain of Madrona argues Meta's earnings drop was overblown and all four hyperscalers benefit from a rising tide of AI investment. He calls their AI services "reasoning machines" that combine models, context, and user data, and examines their custom chips: Google's TPUs, Amazon's Trainium, and Microsoft's lagging hardware. Nvidia hit $5 trillion market cap but faces AMD competition.
Chapters
For more on the big takeaways from mega-cap tech earnings, let's bring in Madrona managing director Matt McIlwain. Matt, good to see you. We got a— we got an overall look that we can show you of this tech earnings and the AI story, looking at the CapEx that's out there and the degree to which they're going to be able to prove that it's worth it, at least what they've shown so far. So we'll put that up in a moment. There it is. It's, it's up on the wall. What was your impression of Meta earnings in particular and the degree to which the negative reaction was justified? It seems like a perceived leader in this group and then a perceived laggard in Amazon sort of switched positions, at least for a while this week.
John, I completely agree with your comments just a few minutes ago. I think it's an overreaction on Meta. And the macro point is that there's a rising tide that's lifting all hyperscaler boats right now. All 4 of these companies, I think, are must-own companies. And now you get Meta at a little bit of a better rate here. You know, notably, as you also point out, until this earnings today or yesterday, rather, Amazon was actually down for the quarter and was really a laggard amongst this group of hyperscalers. You know, over the past year, that's clearly turned around today in the market and deservedly so. You know, all of these companies have the capital and the wherewithal to build out data centers and to enable models to be trained and run on those data centers. And that sets you up for what they're really doing, which is building what we think of as reasoning machines. You know, all those applications Meta has that they're monetizing today are effectively reasoning machines that combine models and context and data and help solve real-world problems or help solve real-world interests of people being connected and communicating together. Amazon's doing that as well. Google's doing that as well. Of course, Microsoft to.
It strikes me that we as investors, when I say we still don't know who's building an iPhone in their data center strategy and who's building a Wintel PC. And what I mean by that is the degree to which these efforts to engineer chips specifically for AI that are more performant than competitors is going to be successful for one or more of these hyperscalers and give them an operational advantage over the competition? Right now, it seems like we're paying attention to top line growth numbers to share, but not yet thinking about who's going to be able to operate this stuff effectively and at lower cost.
I think it's an— all 4 of them are saying it's an all of the above strategy for me. And I think we all have skepticism on the degree to which they can do different things. Google more optimistic on TPUs, their ability to build chips. Microsoft further behind in that area. Amazon, I think with the announcement today, gets more— yesterday gets more credibility around their Trainium chips and what they're doing with that and in particular how they're partnering with folks like Anthropic. When you go up the stack, then you say models. Again, Google, a year ago there was a lot of skepticism around Gemini. They've had a very strong year with their models. Microsoft, you know, a big week for Microsoft, a little bit overshadowed by the earnings. Is the partnership realignment with OpenAI. I think that sets Microsoft up well for the next 5+ years. Amazon probably underappreciated going into this week on their relationship with Anthropic and their Bedrock offering, which allows what Amazon always does, which is said price, convenience, and selection. We're going to give you the selection so you can build these reasoning machines on top of these models.
The name behind all these names, NVIDIA, which topped $5 trillion in market cap for the first time this week, might have ended up a little bit below there to end the week, but still an important milestone. To what degree is its position protected given the strength we've seen in AMD shares? This talk about custom chips, but then this overwhelming overall demand for AI and data center?
I think the overwhelming demand is what's the winner here. I mean, to think that the 5 companies that announced earnings this week are going to spend over half a trillion dollars on CapEx next year. A fair bit of that comes Nvidia's way, even if they have their own chips that they're building. And recall that Nvidia is not just about amazing GPU and GPU systems, but they have CUDA and they have capabilities up the stack, and they're the ones that are playing the most offense in investing into private companies and even partnering in new ways. Take the Nvidia and Intel relationship. And how that's come together in the last couple of months to, you know, kind of facilitate growth in partnership for both. You need a lot of CPUs along with those GPUs to build these reasoning machines that I'm talking about. And those reasoning machines are going to run on all of these hyperscalers data centers.
