The AI Compute Shortage Could Last Until 2028

What's up guys? You're listening to Milk Road and I want to tell you that today's video has actually aired before on our channel, but we thought that this conversation was so good that we should air it twice. So, this is a little excerpt from a past conversation. And just a reminder that if you want to see what our analysts are calling next across AI stocks and crypto, all of that is in Milk Road Pro at the link below. Enjoy the video. >> The moment everybody on finance finance Twitter was talking about is Jensen named Chet Nevius. He said on stage that startups like companies like Nevius are actually doing incredibly well and Nevius is right in the middle of AI factory layer on Nvidia's Europe slide as well. And who was also at the event? You guessed it, Arkady. Arkady, the CEO of Nevius, was also also there and what he said is honestly one of the clearest statements about the AI compute shortage I've heard all year. He said, "Yes, the prices are still there. Demand is there is still there. Actually, the it's the main thing which is happening on the market. Demand is out there." And then he said the problems are how to finance, how to build, and where to find power. And then someone asked, "How many times and Nevius raised GPU prices in the last few months?" And he said, "Nobody counts them anymore. They now reprice their spot actions every 12 minutes." And he said the price mostly just goes up. Every 12 minutes guys, think about how different that is from the you know, story the market was saying like just couple months ago. The market was saying GPUs are going to become a commodity, prices are going to collapse, the neo clouds are going to go bust. Oh, no. And the one of the CEO of one of the biggest neo clouds is telling you the prices goes up every few minutes. I've seen projections in the last like two one day, two days alone, the prices have went up like 30, 40%. If you look at Nevius has its own website where you can track all these prices, which is pretty cool. Nebius also said it could sell out its entire 2027 capacity today, but it's holding some of it back because why? They want to sign these short-term deals like Elon signing like for 50 MW 50 million per MW deals. So they're holding back as well. And they're targeting about 8 billion in annual recurring revenue by the end of end of the year as well. And uh >> [snorts] >> Nebius also announced like as I said like market-based pricing and it added like agentic workflows to its physical AI like workbench. All this boring stuff. All you got to know is like they're just like executing at a very fast very very highly high level and nobody is building and building a team just like Nebius. So like this is why I keep on saying like market has been selling off companies like Nebius and Corvy. Corvy I understand the debt concerns. I understand their yields going up. Nebius is an incredible company guys. I think they closed at around maybe 170 172 today, but you should be buying Nebius at these levels. to say that to all our members as well. This is easily when the market macro conditions get better this is easily a 250 to 300 dollar stock. Like that's a no-brainer for me. So I continue to make the case for Nebius here. So what else? And also like the the biggest takeaway for me in Berlin is this. Every country wants its own compute. The United States, the Middle East, Asia, now Europe as we seen that's more demand on top of hyperscalers, on top of the neo clouds, on top of SpaceX, and the supply chain, the the chips from TSMC, the HBM from SK Hynix, Micron. You name it. The power of the transformers. None of that can scale overnight. That's that's the core issue. Now before we get to Meta, I want to take a few minutes on the question I get all the time. If everybody needs compute and everybody has the money, why can't they just build more of it? The honest answer is that compute is not one thing, it's a stack. Like every layer has its own bottleneck, and the whole system can only move as fast as its slowest layer. This is one of the reasons why we look back like 2 years or a year ago how brilliant Jensen has been. Jensen has been going to these companies begging them to increase supply chain, investing in these companies. Yet, you know, a lot of them are still not doing that because they they're not as like they they don't have the they're not seeing the big picture like Jensen's seeing. And we're starting to see that it, you know, come out through these bottlenecks that I'm going to mention. Like, so let's start at the bottom, right? You have power. To turn on every gigawatt of data center, you need a gigawatt of power. You need it every single day, every day. In the in the US, the grid was not built for this. Utilities are telling data center developers to developers to wait for a connection. The transformers, the big pieces of equipment that like step up and down have lead times of more than 2 years in some cases, and capacity prices in PJM, which is like the biggest like uh grid operator in the country, have basically exploded. And that's why you keep seeing like Nvidia put nuclear on their keynote slide. Um that's why everyone is talking about gas turbines and small modular nuclear reactors. And power is the first wall. If you don't have power, you can't turn anything out. Then you have buildings, right? You need land, permits, cooling, and a lot of people who knows how to build these things. And it's just getting harder because just this week like activists said that they have a blueprint for campaigns against hyperscalers across country disrupting like Oracle data center in New Mexico. So, even when you have the money and the chips, local fights can like really slow you down. And then you have the chips themselves, right? Like the like almost every AI chip, whether it's Nvidia, AMD, custom chips to Google, Amazon, goes through TSMC. There are some other foundry businesses as well. You have Intel and Samsung, which are like much smaller. And the advanced packaging that glues all the GPU and the memory together has its own bottleneck as well. And underneath TSMC, you have ASML, the only company on Earth that makes the EUV machines you need for for the most advanced chips. ASML said the AI chip shortage last until 2028 in the recent earnings as well. Then you have memory, which is in my opinion the tightest of all right now. HBM uses three times the DRAM. I'm not going to go into this. You know how all this all this works. Basically, what I'm trying to say is there's a there is immense amount of bottlenecks. So, when you hear somebody AI capex is going to fall off a cliff next year, ask them one question. Which layer of the stack is going to suddenly have too much supply? Because right now, I can't I can't find a single one. >> Guys, I'm just going to pause here to tell you about one of my biggest problems in the market, which is seeing what some of the big names in the world actually hold in their portfolios. I'm talking about like Gavin Baker, how much SpaceX does he hold, how much Nvidia does he hold, Cathie Wood, Peter Thiel, Ken Griffin. We always hear about these big names, but nobody actually shows you individually what stuff they hold, which is why we went and built a website that does exactly that. So, we have relaunched milkroad.com to basically aggregate all of these insider portfolios, we're calling them. Basically, puts it all into one spot and lets you set alerts to tell you when they make moves. I'm talking about Warren Buffett, freaking Donald Trump is on there, Ray Dalio, all these big names you hear about. You can see their full holdings updated in real time on milkroad.com, so make sure you check that out and sign up today. It's 9:00, the kids are finally asleep, the dishes are done. It's time to chill or maybe start another project or think about buying a house or listen to an awesome podcast. Either way, you've got a few minutes to yourself and researching something like life insurance is It's the last thing you ever want to spend those minutes on. So, it ends up on that list for another few months and you put it off again. So, Ryan's, a partner of today's show, is actually built to fix that. At surance.com, you can request a free, no obligation quote for term life insurance, and the whole process is online. You can compare options from established carriers and talk to a licensed advocate if you like a person to walk you through it, and some policies don't even need a medical exam or blood test depending on eligibility. You've already done the hard part, which is building a life worth protecting. This is the step that makes sure it's covered. Go to surance.com for your free, no obligation quote. >> And that brings us to Meta tonight. So, Meta reported Q3 just just like an hour ago, and this is a perfect example of everything I I've been talking about. Let me start with the numbers, um because the revenue was 64.6 billion, up 26% year-over-year. Wall Street was expecting about 63.3 billion, so a clear beat. Advertising revenue was 63.1 billion, and also up 26%. Ad impressions were up 10%, average price per ad was up 14% as well, and daily active users across the family of apps hit 3.64 billion, and the Q4 guide was 73 to 76 billion against a consensus of consensus of 73.6 billion. So, the midpoint is above the street as well, and um and the stock is up roughly 4 to 6% um it's like consolidating at the moment. Um because because they heard the street really like what they heard from the from Meta from this earnings. Now, here's the real story in Meta, um CapEx. We all know Meta had huge CapEx. Um in Q3 alone, Meta spent roughly 34.9 billion on CapEx. Operating cash flow was 32.08 billion, and so because of that, for the first time, free cash flow was negative in a really long time. Um so, let me put that in context. Meta, one of the best generating businesses in the history of business, in this quarter they spend more on capex than they generated in operating cash flow and if you look at this chart going back to 2015 this is the first negative cash flow quarter on it and for the full year they narrowed the 2020 capex to 135 to 145 billion the range was 130 to 145 billion so they raised capex again so why did they raise capex why are they spending this so much money it's because of something called Muse Mark said in just about 6 weeks more than 10 million people in the United States and Canada have set up Muse personal agents it's been number one like on the on the app store almost every day and about 25% of people who set up Muse they use it on any given day he said he also said that earliest adopters they use it longer they use much the longer they use the better the Muse gets so 10 billion in just six two six six weeks in just two countries that's crazy but here's the important thing to understand about Muse it like it's extremely computer intensive it's running models it's taking actions it's an agent running on your behalf it's connecting your app and every single user costs a lot more compute than a normal like average Instagram user that they use these data centers for and and this is the key line in this episode right like the reason Muse is not launched global globally yet is because of compute meta does not have enough compute to give Muse to everybody in the world right now a company with 3.64 billion daily users spending up to 145 billion in capex this year is telling you that it cannot launch its successful product globally because it does not have enough compute guys that you guys is the compute shortage in one sentence so what are they doing about it? Two things. One, they're working on cost efficiency. Meta said it's roughly targeting 10x improvement in the cost efficiency of serving Muse. That's huge, and it's the same idea as, you know, Nvidia power efficiency that that I mentioned earlier. And when you get 10x more efficient, it doesn't mean you buy less compute. It just means you can serve way more people with the compute you have. And two, they're building capacity for more than 1 billion users. Meta is planning on using um its infrastructure around the world um where Muse has a billion people plus using it. And if 10 million users already needs as much, imagine a billion people on Muse, guys. And they said they also plan to invest aggressively in 2027. So, if you were hoping that capex would like slow down next year, it's not going to happen at at Meta. Or it's not going to happen at SpaceX with Grok bot. Or it's not going to happen at OpenAI when they raised their in private rounding of $30 billion. Or it's not going to happen at Anthropic, who is going to IPO potentially next month, who's roughly looking to raise probably 80 to 100 billion dollars. Where is all that money is going to go to? All that money is going to just go into more compute and more compute, guys. That's That's That's where That's the world we're in right now. We're so compute constrained. And all that money will go through and falls through um the supply chain as well. So, a lot of Pixel shovels will obviously benefit from that as well. And Muse is also like or Meta is also building the enterprise platform. Um Muse for small businesses launched in late September. It basically plugs into Instagram analytics, Facebook pages, Meta ad accounts, plus a dozens of like third-party like tools like Shopify, QuickBooks, Canva, you name it. It can analyze your app and content performance, draft campaigns, customer responses, and flag issues. And Susan, um who was talking in their in the earnings call, pointed out in their call that meta has touch points with about 500 million self-identified small businesses and if one small slice of those AI agent every day it's a lot more compute and hopefully a lot more revenue for them as well cuz they will start to charge subscriptions for this as well. So and they're planning on adding Muse to meta glasses next year that's going to be more compute they're they're going to hopefully going to have to find some more so and they also said on the on the earnings call about reality Labs they said that reality Labs could be lost could be around 15 billion next year and that's on top of the AI spending so meta is essentially funding glasses devices like AI models agents data center all at the same time mostly from their ad business. This is how incredible their ad business is guys. Imagine they just stopped doing all this. They're it's insane how much money they will have in the bank but you as a company you cannot be doing that because in a world where AI thrives you constantly need to reiterate your your business plan and invest accordingly and that's what meta is doing. And there was a surprise in there though meta announced a new non-voting share class so it's actually two to four one where shareholders get like non-voting shares. Why does that matter because it gives Mark another way to issue stock for things like acquisitions compensation without diluting his voting control. Some people love it some people hate from a governance point of view but it's worth knowing about it's basically that he's going to go and get more money to fund this capex. Now the big bear case is always there with meta right like you have to think about the negative cash flow now which is fine capex is going to potentially go up reality Labs is losing money and just today meta just a judge just ordered meta to pay 25 billion after a jury found it you know of consumers about data privacy. So there's a lot of moving pieces with meta but they honestly when I look at this quarter, I see a company growing 26% 26% growing ad business that is praying for a product that people clearly want, and only thing that's holding them back is compute. >> Thanks for listening to Milk Road. If you enjoyed the show, make sure you like and subscribe. >> [music] >> And if you're struggling to find winners in the market, that's exactly what Milk Road Pro is built for. Our analysts have called some of the biggest winners early, and [music] Pro lets you see what they're buying next, every trade they make, and the research behind every position. Check out Milk Road Pro at the link below. Everything you hear on Milk Road is for informational purposes only. These are our personal opinions, not financial advice, and we may own some of the investments we talk about. Always do your own research and make the decisions that are right for you. See you next time.