Dan Ives: Is the AI Infrastructure Race Heading for a Crash?

Dan Ies argues that Nvidia's rumored 15% price hike for data centers is actually a massive bullish indicator for the tech trade. Ives, managing director at Yorkville Ives, joined Bloomberg television recently to break down why this pricing power signals an accelerating AI super cycle rather than a market bubble. By the end of this video, you will understand how a hidden memory bottleneck is driving these costs and why the ultimate payoff could dwarf the massive capital expenditure keeping investors awake. Dan joins us now for more. Dan, good morning. >> Great to be here. >> What's getting more expensive? >> Look, memory. I mean, look, the reality is is that we're seeing a memory super cycle take place. And I think that's the word all around Korea. And that's why Nvidia, they're going to need to pass some of this to their customers. But but that's not that's not slowing down because the reality today, demand, the supply for chips is upwards of 15 to one. And I think that's that's what not just Nvidia, but ultimately their end users. Who needs to eat the higher costs right now? Who's eating it? >> Now through the supply chain, you'll see some of that eaten, but then ultimately it's the end user, it's the customers, it's enterprises, it's tokenization. As that all plays out, then the question will be, okay, when do you hit some sort of equilibrium? I mean, we don't think you hit core equilibrium, probably till middle to late 2028. And I think that right now is a quagmire whether it's Nvidia whether it's the memory players Micron everyone else because the hyperscalers you just like as Lisa talked like as they'll continue to raise capital and capbacks will continue to increase the AI revolution still in the third inning and I think you'll hear that more from Nvidia later this week then it comes down to like memory and Apple obviously seen that front and center those costs going to continue to go up. They're going to look for other costs to go down though. And I was looking, for example, at the story over the weekend of Nvidia potentially telling some of its customers like Microsoft and Alphabet uh and Oracle that their costs are going to go up about 15% tied to data centers because of the memory chip and how much the price is going up. And you start wondering, is this part of the reason why they're cutting back on Frontier models? Is this part of the reason why they're looking for efficiencies from say open-sourced uh AI models coming from China? >> Yeah. And that's why when you look at Apple and they looked at chips potentially from China because they're looking across the board and say okay where do we cut cost because on one side of the ledger costs are going up but it speaks to our view that you you until you really hit scale costs are going to continue to go higher. I think what the hyperscalers and what the chip players and from Nvidia what everyone's dealing with is you cannot take their foot off the gas right now because of the demand cycle and to your point when it comes to like the debt raises despite what we're seeing that's going to accelerate. I mean, you talked about it going to 220. You go to next year, it's probably double from that, you know, in terms of as it's going to continue to increase given the capex that's needed in terms of what I'll view it is building. It's building Vegas strip 1955 type movement. >> I was looking at Nvidia's earnings and just the the returns that you've seen in their shares. So far this year, the shares are up 15%. Last year, they were up 38%. The year before they were up more than 100%, the year before up more than 200%. It's been a rapid deceleration in the equity returns as multiples have gone higher and higher. At what point have they just reached a pinnacle of what multiples can look like and even if they deliver amazing results, they're not going to get that kind of pop in the stock? >> I think that's the debate. But then to some extent, Divo funded the AI party. I if you think about the memory play, look, think about Korea. Cosby is not where it is without Nvidia. You know, there is only one chip in the world fueling the AI revolution. That is Nvidia. And I think the point here is that as you see with numbers as it plays out, there's no one that has a better perch than them. And that's why I think you'll hear from Jensen because the question is is it as enterprise accelerate and we think accelerating probably 20 25% even over the last few months as that runs through the system. It's investors ultimately recognizing that cuz I still it's my view investors are still underestimating the scale and scope of the AI revolution. Now it goes back to like the bond market. You're going to need to take out more debt and then where is there a patience sort of a tugof-war that goes on but from a demand perspective it's accelerating not decelerating. >> Earnings look great. Are they as good as they look? Lisa's asked this question before. or Blimbo subscriber wrote in just moments ago and asked the following question whether this isn't just about accounting tricks. The question directly is someone's capex which gets depreciated over many years falls straight into someone else's top line recognized instantly. No one predicted this earnings boost because no one thought this through. Not one single southside analyst. What's your response to that? I my response would be look at Palunteer earnings as a good example of just purely just forget capbacks enterprises signing on the dotted line to ultimately accelerate their AI initiatives. If you look at Microsoft's quarter that was a perfect example of what's happening to that just forget capex for a second what are their install base their enterprises saying what they're basically saying is bright green light to go down the AI path. So when you go down to the capbacks and the circular financing and you know they're invested in there and and everyone needs everyone else to to ultimately be positive in terms of open AI and anthropic I think it gets away from just the core demand and also for every dollar of capex there's a5 $6 multiply that the earnings aren't there they're not that great I'm just implying that they're not as good as people might be saying they are is that fair >> I look I would say 6 n months from Now they have to continue to execute and prove themselves out and if they don't then the stocks reflect that. But I think right now that's why every it look this is all a Jenga puzzle from Palunteer to Nvidia to Microsoft to open athropic you got to put it together to understand what's ultimately happening. What I would just say is from an Asia perspective and what you've seen with demand, demand's accelerating, that's why price are going higher and that's something that's bullish for the overall tech trade. >> The immediate pressure point driving this entire conversation is the soaring cost of hardware. The host opened the discussion by asking Ives what exactly is getting more expensive in the technology supply chain. Ives pointed directly to Asia, explaining that a massive memory super cycle is currently underway, particularly in South Korea. This has created a severe supply crunch where the demand to supply ratio for high performance silicon is sitting at a staggering 15 to1. Because of this extreme imbalance, the company at the center of the global technology trade, Nvidia, ticker symbol NVDA, is being forced to pass these rising memory costs directly down to its customers. According to Ives, there is simply no sign of this supply demand bottleneck easing anytime soon. When the host pushed him on who exactly is going to eat these higher costs, Ives clarified that while some of the financial pain will be absorbed through the intermediate steps of the supply chain, the ultimate burden falls squarely on the end users. This means the major cloud providers, enterprise customers, and developers working on model tokenization will have to absorb the pricing premium. He estimated that the industry will not reach a core equilibrium between supply and demand until the middle to late part of 2028. This is a critical timeline because it suggests we are facing years of elevated capital requirements. My take on this is that while Nvidia's pricing power is undeniably robust, it highlights a fragile systemic dependency. Nvidia does not operate in a vacuum, its margins are hostage to the pricing of memory manufacturers like Micron, ticker symbol MU. If memory costs continue to climb, Nvidia's eyewatering gross margins will face a persistent headwind that even their market dominance cannot entirely ignore. It is a reminder that the hardware layer of the artificial intelligence boom is a complex web of interdependent players where a bottleneck at one node can ripple through the valuations of the entire sector. To understand the scale of a 15 to1 demand to supply ratio, we have to look at what it represents in terms of market leverage. When 15 buyers are competing for a single unit of output, traditional costbenefit analyses go out the window. companies cannot afford to negotiate on price because the alternative is getting left behind in the infrastructure race. Ives emphasized that this extreme imbalance is why Nvidia can pass these costs down without slowing down its sales momentum. In my view, this is the definition of inelastic demand, but it also creates an environment ripe for double ordering and artificial demand inflation. If buyers are ordering more than they need just to secure a spot in line, the eventual correction could be far sharper than the market is currently pricing in. This raises a vital question about how these massive buyers are reacting to the squeeze. If hyperscalers are forced to pay significantly more for hardware, where do they find the money to cover these premiums without crushing their own margins? That's the part most coverage skips right over. So, if it landed for you, like the video and subscribe to the channel. The host steered the conversation toward a major industry rumor, pointing out reports that Nvidia is telling its largest customers, including Microsoft, ticker symbol MSFT, Alphabet, ticker symbol GO, and Oracle, ticker symbol OCL, that their data center costs are going up by 15% due to memory pricing. She asked Ives if this pricing pressure is the real reason why these tech giants are suddenly cutting back on expensive frontier models and looking for cost efficiencies in open-source AI models coming from China. Ives agreed that companies across the board including Apple ticker symbol AAPL are actively looking at alternative chip suppliers and cost cutting measures to balance their ledgers. However, he argued that hyperscalers simply cannot afford to take their foot off the gas right now because of the sheer velocity of the demand cycle. Instead of scaling back, he expects the massive capital expenditure and debt raises we are seeing to accelerate even further. To put this massive buildout into perspective, Ives used a vivid historical analogy, comparing the current infrastructure phase to building the Las Vegas strip in 1955. It is an era of speculative highstakes construction where the foundation of an entirely new ecosystem is being laid down simultaneously. My take on this is that the Vegas 1955 metaphor is highly revealing, perhaps in ways Ives did not fully intend. In 1955, the Nevada desert was mostly empty, and developers were laying down massive capital on a speculative bet that the crowds would eventually show up to gamble. While some of those early casinos became incredibly lucrative, many of the original builders went bankrupt long before the tourist boom fully materialized, tech giants today are building massive data centers under the same philosophy. build the capacity now and figure out the monetization models later. It is an infrastructure land grab where the risk of overcapacity is being ignored in favor of sheer competitive speed. While this approach keeps Nvidia's order books full, it places an immense financial burden on the hyperscalers who must continuously find ways to subsidize this buildout. They are managing this margin squeeze by seeking open-source software efficiencies and squeezing costs out of other business segments. But this is a temporary fix for a long-term capital problem. But as these tech giants pour hundreds of billions of dollars into this physical infrastructure, is the stock market starting to reject the math behind these valuations? The discussion naturally pivoted to the stock market itself with the Bloomberg anchor highlighting a troubling trend in Nvidia's historical returns. The host pointed out that Nvidia's equity returns have shown a rapid deceleration. A few years ago, the stock was up more than 200%, then more than 100%, then 38%, and recently it has slowed to around 15%. She asked Ives at what point the company reaches a pinnacle where multiples can no longer expand, meaning even stellar earnings reports will fail to move the stock higher. Ives responded by arguing that the market is still fundamentally underestimating the scale and scope of this technological shift. He asserted that Nvidia has essentially funded the entire AI party and pointed out that the transition from hardware installation to software deployment is where the true exponential growth lies. Ives believes we are only in the second inning of this cycle and he estimates that for every dollar spent on an Nvidia GPU, there is a multiplier effect of $8 to$10 across the software and services ecosystem. In his view, this is not a short-term bubble like the.com era, but the early stages of a massive productivity wave that will eventually touch every sector of the global economy. Where I land on this is that the 8:1 multiplier is a beautiful theory, but the math on the ground is looking increasingly strained. If a company spends a billion dollars on hardware, it has to find $8 billion in software value just to justify that initial outlay. Right now, enterprise software companies are finding that selling AI features is a slow consultative process. Corporate clients are questioning the security, the accuracy, and the actual return on investment of these tools. We are seeing a massive gap between the physical capacity being installed and the software revenues being generated. If that gap does not close soon, the return on invested capital for these hyperscalers is going to drop off a cliff. Are we putting the cart before the horse? It certainly feels like it when you look at the physical bottlenecks. The Bloomberg host pressed Ives on this exact issue, turning the focus to a constraint that cannot be solved by writing code, the power grid. She pointed out that these advanced clusters require immense amounts of electricity and asked how the industry can possibly sustain this growth when the utility infrastructure is already operating near capacity. Ies acknowledged the bottleneck but framed it as a massive investment opportunity rather than a barrier, pointing to recent deals where tech giants are partnering with energy companies to secure dedicated nuclear power. My take is that Wall Street is vastly underestimating the friction of the physical world. You can build a software application in a weekend and you can build a data center in a year, but upgrading transmission lines, securing environmental permits, and building or restarting nuclear reactors takes a decade. The energy bottleneck is not a problem you can solve by throwing venture capital at it. If the power is not there, the chips cannot run and those expensive data centers become monumentally expensive monuments to overconfidence. We are already seeing projects delayed in key data center hubs because local grids simply cannot supply the megawatts required. This brings us to the ultimate reckoning, the timeline for monetization. Toward the end of the sitdown, the host asked Ives what it will take for the market to maintain these elevated valuations if the capital expenditure continues to climb. Ives remained steadfast, arguing that the upcoming earning season will act as a major catalyst, proving that enterprise spending is translating into real topline growth. He believes the bears are going to be proven wrong yet again as companies report accelerating cloud revenues driven by AI integration. I am highly skeptical of this optimistic timeline. The market has been incredibly patient, but that patience is directly tied to the expectation of near-term results. When companies report their earnings, investors are no longer going to be satisfied with vague promises about future productivity gains. They are going to look at the capital expenditure line, compare it to the free cash flow, and demand to see the exact margin contribution from AI services. If those margins are compressed by rising memory costs and energy premiums, the multiple contraction could be swift and severe. Ultimately, the race is no longer just about who can buy the most chips. It is about who can build a viable business model around them before the capital runs out. The next few quarters will tell us if we are indeed in 1955 building the foundation of a modern empire or if we are simply overbuilding in a desert of hype. Keep a close eye on the capital expenditure guidance of the major hyperscalers next week. That is where the real story is written. The reality is that we are about to find out exactly how much of this infrastructure spending is a strategic necessity and how much is simply a race to avoid being left behind. As that data comes in, the market will likely split between those who have built a moat and those who have just built a bill. The era of blind faith in massive spending is drawing to a close, and the era of hard-nosed margin analysis is about to begin. Thanks for sticking with me all the way through. Like the video and subscribe to the channel if you haven't yet.