The Six Five team discusses Microsoft & BlackRock Sitting in a Tree
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.
Transcript:
Daniel Newman: Microsoft-BlackRock partnership, $30 billion rising to $100 billion. The goal is to invest in infrastructure for powering these data centers. Pat, rather than talking too much about this fund, I think it’s more of a theme. Yesterday, I shared a job description in a Data Center Dynamics article that talked about Amazon hiring a principal nuclear engineer for AWS. We heard about the reopening of Three Mile Island. We are hearing Larry Ellison on stage talking about nukes and building specialized nuclear power facilities to power these data centers as he talks about thousand percent increase in data centers over the next handful of years. We’ve got Nvidia hogging more power than any other company on the planet right now. Of course, we’ve got multi hundred billions of CapEx rising at fast rates for every cloud company in the world. Problem is we are running out of power. So, now we’re starting to see a new cycle, a new news cycle. The new cycle is holy crap, wind, solar is not going to get this done. We are going to need something more robust. We’re not going to do more coal and fire. We’re going to have to find a new way. The cleanest available energy, despite its historic bad press, is nuclear. Problem is in parts of the world like Germany, they’ve shut it all down. In the US here, we’ve basically not invested in a long time in any meaningful capacity.
So, we’ve got this interesting inflection now where we want AI, we’re thirsty for it, not thirsty enough to buy an iPhone 16, but we’re thirsty for more GPUs. Then Pat, of course, this also accelerates the accelerator conversation. We are accelerating our conversation about accelerators for AI because as we know, as these workloads and needs become more well understood, we can build specialty chips that are going to be more efficient for doing those specific things that we’ll use less power, but nonetheless a lot of power is needed. So, Pat, to me, this is not so much a big moment with just the BlackRock-Microsoft thing, but this is going to be like we heard about all these LLM partnerships, Anthropic and Amazon and Microsoft. We’re going to hear the next wave is going to be companies tying up major investment funds to figure out how to get more power because the next oil is not GPUs. It’s going to be enough power to power those GPUs. What do you think?
Patrick Moorhead: There’s so much here. You know what? I’m going to be late to my workout.
Daniel Newman: Yeah, so go quick so you can get to your workout.
Patrick Moorhead: Oh, I don’t want to go quick.
Daniel Newman: Go slow and tell your trainer to… Anyways, he’s not-
Patrick Moorhead: Well, I built some time in knowing that you never show up on time and we get out late. I’m just kidding.
Daniel Newman: I was right on time today.
Patrick Moorhead: I know. Please don’t show up on that. Okay, so a couple gap fillers here. China competitiveness, I think, it’s just… Is that a helicopter? Oh yeah. China competitiveness is kicking in here. China has been putting up new nukes and new coal fire like absolute crazy here. Here we are and let me give you a good example. East Grid in Virginia where AWS has its largest data centers has 3% power remaining to support more people, EVs, and data centers. There will not be new data centers being installed in a place where there’s 3% energy. Our regulatory environment is so strict that there likely won’t be any new energy put in there for five years. I very can safely say even if you put up a coal fire, you can’t drive enough power with solar that’s consistent. The challenge with solar is that it’s consistent only with the sun being out. Nothing against solar. I’m just spitting out the facts here. With what Greenpeace did and other activists did on nukes, it became a bad thing to install any new nukes out there, even though it’s literally the cleanest consistent energy that you can do. It’s amazing. Dan, were you even born when Three Mile Island almost had a meltdown? How about Chernobyl? Probably. Okay. Lawn people are probably there. Okay, I get it.
Daniel Newman: I don’t want to distract you. I know how that noise can be…
Patrick Moorhead: Yeah, I’m sorry. But yeah, I’m just struck at how little foresight that we have as a country to predict things and we just go along for the ride as corporations. Just another example where we can find ourselves just on the complete ass end of a trend. I am very inspired though by some of these companies that used to be very activist. They’ve got their ESG reports that are now fricking going all in on nukes and absolutely love that. Dan, you brought up essentially the ASIC versus GPU. Jensen Wong was quoted in this and Jensen does need a heck of a lot of power to see his growth, but I really hope that it drives the conversation on power efficiency. 100%, I’ll take this to the bank and I’ll die on this vine that ASICs are more efficient than GPUs, up to 10X efficient. GPUs to me, you can bank on for two or three generations of it. At least as it relates to TPU, even ASICs are coming out new ones every year from these folks. So, I really hope this drives the conversation about, “Why aren’t we seeing more Gaudi, Intel Gaudi? Why aren’t we seeing more Groq out there? Why aren’t we seeing more Qualcomm A-100?” I know those are primarily inference, but when it comes to Maya and TPU and Trainium, those do LLM training and LLM inference. So, I hope it drives this conversation. It’s a good conversation to have.
Daniel Newman: Yeah, I mean there have been some wins. I was cited in a press note about Groq and Aramco digital partnering 19,000 LPUs going into the Middle East to run the inference of Aramco Digital. So, we are seeing it, Pat, but our forecast has about a 20% delta in terms of how fast ASICs are going to grow versus GPUs. I actually think it’s going to get bigger than that. When you’re building bottoms up, it tends to come from data from people with less foresight. So, you got to always take a little bit of a risk top down. My assessment is the GPU CapEx investment won’t stop growing, but it might normalize more. You’ve got Maya coming online, you’ve got AWS doing their thing. You saw ByteDance, Alibaba, Meta. You just go down the list. They’re all building their own. You know that ARM’s got something up its sleeve. I don’t know. I’m suspicious that it does. What I mean by this is these companies aren’t building these to not use them. Of course, Google TPU hogs close to $7, $8 billion a year of spend on TPUs for their own use cases. Pat, this is going to happen. This is going to happen and it has to happen. By the way, it doesn’t mean there’s no cycle for continued buying of GPUs. It just means it’s going to change. We all knew the training will continue and 10 to 20 companies will train mega models. Everyone else is going to use more efficient use case specific chips that can train for a need or more importantly deliver inference for a specific workload. That’s going to happen.
