Transcript
Folks, NAND costs have doubled in the past six months. Energy costs are up 10% or more at two-thirds of enterprises, and the AI arms race is consuming both simultaneously. Many enterprise buyers have no idea what's actually happening to their infrastructure economics right now and are looking for some of the straight truth and some transparency about where we are and where we're going next. Well, today I'm joined by Woojin Ho from Bloomberg. Woojin, why don't you introduce yourself to our audience? Hey, Sean, thanks for having me on the show. Yeah, I'm Senior Hardware Analyst at Bloomberg Intelligence. Been with Bloomberg for the past 10 years, cover hardware and networking. I also cover the enterprise storage space, and hopefully I can help your audience in guiding and navigating through this crisis. Yeah, you know, I mean, we've seen these kind of cycles before, these super cycles before. This one feels different, but why don't you start off by walking us through this NAND cycle? How'd we get here? How bad is it? And, you know, the question on everybody's mind, how long does history say it's gonna last? Yeah, so I don't know if history's gonna be precedence this time around, right? If you think about the NAND cycle from trough to peak, we're talking about four to six quarters, right? We're already in quarter two. In this cycle with prices doubling as you started off the show. What we're starting to see over the past couple of months, we've seen sequential pricing increases since the beginning of the year. There's very little sign of that slowing down. We talk about double digit price increases on a per gig basis. And I got off the SanDisk call the other night, and essentially what's happening is is that the hyperscale cloud providers are really eating up a lot of the capacity in NAND capacity that's out there right now to help support their AI models and the AI infrastructure. This time truly is different. We've heard this multiple times ago already in the past, but this time truly is different because the way they're signing these long-term agreements, we're talking about five-year terms, right? And up from purchase commitments. So this is gonna be an elongated cycle with the capacity not coming back online or more capacity not coming online until late 2027, 2028 timeframe. So let's talk about these hyperscalers for a moment because I don't think we've seen this dynamic before, right? We're talking about 650 billion in capital infrastructure outlay last time I added it all up. And in essence, we're talking about the spending coming from hyperscalers who are generating so much cash and growing at such a fast rate that ultimately if they wanted to, they could continue to buy at this level indefinitely. So if that's the demand curve, first of all, is that a normal or dependable demand curve? Do you think that will continue? And second of all, the supply coming online. I mean, all estimates are that's kind of the second half of next year, but by that point, is that gonna be enough? Yeah. So the $650 billion in CapEx for we're talking about only five or six major hyperscalers. We're not including some of the Neo clouds that are out there. Plus we're not talking about some of the sovereigns who need some of the storage capacity as well, right? By our estimates, we're still looking around 15 to 20% CapEx growth over the next couple of years. There's a lot of side to a trillion dollars in CapEx, right? By exiting out the 29 to 30. One of the reasons why they're spending this much is because they want to power their large language models. And what we're seeing from the hyperscalers right now is these gigawatt scale data centers, right? They've broken ground a couple of years back and we're starting to see the construction of those. And now they're starting to fill those buildings today. That's your $650 billion. We're just starting to fill it with the infrastructure. So the demand over the next couple of years and we've heard some of the major projects, whether it's Prometheus by Meta, they're going to be filling those up over the next couple of years with some multi gigawatt scale data centers down the line. Okay, and then when we talk about supply coming online mid 2027, we've talked to some of the big semis. Do you think that's enough? Do you think that's really going to resolve the problem or is that just going to be barely allowing us to continue to stay afloat? Yeah, and that's one thing that's unclear, right? Because if we think about how we got into this crisis in the first place, we were coming off of the supply chain, supply glut off of the COVID period. So all aspects of the supply chain, whether it is the storage or the memory or any widget maker, they still feel that, they still remember that pain, right? And they were late in building up the new capacity because again, they didn't know if this time was different. Right, there is going to be some skepticism in building up the necessary capacity to help support the storage for not only for AI demand, but also enterprise storage, as well as device demand as well. So I think there's going to be some slack in terms of capacity, meaning the overall demand picture going out to 28, 29, because we still don't know, because we're still building for the core. Yeah. And we haven't even started talking about the edge yet. Well, I heard you say a couple of interesting things there. So first of all, I heard you say that hyperscalers were building for their large language models. And yes, Prometheus and Meta is at the top of that, but let's not forget, Microsoft's also got their language model. Amazon's got their language model. Google's got their language model. And we always think that those are about hyperscalers in terms of selling capacity, but they're also their own foundation or anchor tenant for the infrastructure clouds that they're building. But in addition, phones and PCs got really expensive. So people may not refresh those this year. And that demand pushes forward, which essentially causes a super cycle of PC or phone refresh when supply and pricing comes back online. So it's kind of this perpetual circle. And I think when you start talking about 28, 29, that's absolutely there. Now, if we take a Wall Street view, are analysts starting to differentiate infrastructure vendors based on their supply chain resilience? I mean, this is always a subtext in the annual report, but right now, is it more about who can actually ship than what they're charging or is it something else? No, I think they are starting to differentiate vendors in terms of how strong or how well they manage the supply chain, whether it's on the CPU side, whether it's on the GPU side and also on the storage side, right? That there are winners and losers on this. Well, we all know that the hyperscale cloud providers will go direct to the supplier themselves, but from an enterprise standpoint, the companies that are going to stand out are those that are able to get the supply number one, but also get it at a reasonable price, right? Even if the prices have gone up, the vendors prefer supply or the suppliers prefer vendors who are able to meet the long-term commitments to that supply, right? So they wanna make sure that they'll be able to fill that demand going forward. And you're gonna win it with the margin on the long end, on the long term. Yeah, it's interesting, right? Because ultimately, I think what gets all the airplay today is pricing and price changes and how gear has gotten more expensive and component inflation, et cetera. But I think the real story here is who's actually shipping product and how quickly can they ship product? Because at the end of the day, once customers kind of can absorb or get their heads around the fact that this has gotten more expensive, they actually have to get back to business and get it deployed. So that'll be an interesting part of this quarter's earning cycle in terms of ability to deliver. I wanna switch gears a little bit, Woojin. I wanna move to this energy reckoning because as if supplies and components weren't enough, we've also got the fact that we've got energy cost increases going up, right? Even before some of the military action that's in place today, we were already seeing electricity go way up. We're now seeing oil go way up. But the reality is everybody's kind of sitting in this bridge to running out of power. You talked about gigawatt data centers. Most enterprises aren't in a position to be breaking ground on new gigawatt data centers. That's a hyperscaler statement. Most are trying to figure out how do I put more gear? How do I even get these GPUs into my data center? And so I wanna shift to kind of this increasing and looming crisis of A, running out of power and having your budget in power as you try to stand this all up. How are you looking at this situation? I mean, is this really as dire as we say it is? And if you look at the companies you're talking to at the board level, are they as concerned about running out of electricity and access to power as what we're seeing? So we have to think of this holistically, Sean. Number one, if it's an issue for the hyperscale clouds, it's also an issue for the colos as well, right? And the first in line in terms of getting access to the power from the grid is going to be those with the deeper pockets and it's hyperscale cloud companies, right? The colos may have a tougher time in terms of getting that power, right? So what that does create over time is this capacity crunch for the enterprise or for the colocation devices, for the colocation infrastructure. Number two, if energy prices go up for everybody, including the colocation data centers, it is going to be at the end of the day an issue for the enterprises in the sense that their energy prices are going to go up, number one. And also on the hybrid cloud side, if you're using capacity on the cloud side, your bills are going to go up regardless, right? Because one way or another, those energy prices are going to trickle into your cloud bill somehow or another. So I don't know if the enterprises have really fully thought through this energy crisis yet, but sooner or later, it is going to pinch in to the OPEX because a cloud model at the end of the day was supposed to be an OPEX model versus a CapEx model. And if their OPEX starts creeping up, it's going to affect their margins going forward. But we haven't seen that yet, but that is going to be a problem going forward. Well, and it's interesting because we're seeing some very creative solutions in the marketplace, right? So some out there are saying, hey, I'll move the low end of my disk back to tape and I'll move the low end of my flash back to disk. We're seeing this trickling effect back to these legacy technologies that were frankly replaced because they were energy inefficient. How big a deal do you see this? I mean, we were all kind of seeing that flash was the de facto way forward, but are we kind of taking a step backwards here and is that as big an impact as we think it might be? Part of it has to do with the supply aspect of it, Sean, as well, because you're not getting the capacity on the NAND side as one had previously thought, right? And look, it's not that much better on the hard disk drive side as well, right? But at the end of the day, you have to start thinking about it from a total cost of ownership standpoint as well, right? Because you can only put so much infrastructure to your older legacy technologies, primarily because retrieval times is actually a cost to business operations, right? Something like an SSD or NAND driven, flash driven data is a lot more immediate than having some guy with a ponytail in the data center, retrieving the tape in the back room and then putting up online, right? There's a reason why tape is called cold storage and NAND is called hot storage. Hey, I remember those days where you would ask for something to be mounted on the mainframe and you might wait two or three hours and they'd send you a message saying it's mounted, you can go get what you want now. I just see there's some short-term decision-making and just tied back to this energy consumption. I think everybody's talking about the cost of infrastructure and the amount that hyperscalers are consuming, but I just, I can't help but think we're underplaying the fact that there still is a fixed envelope of electricity in the world. And as far as I know, we're not turning on any new sources in the next 12 to 24 months that are gonna dramatically change supply. So it's, I don't know, do you see any panacea to solve this energy crisis or do you think this is with us for three, five, 10 years? Maybe it's the next era of concern. Yeah, so again, that's the big bottleneck in data center build-outs, right? There is no quick panacea available. I do think that this consumption-based model will help, right? It's at the end of the day, it's cloud-based. And if you can add capacity on a piecemeal basis, that will help bridge the gap using this consumption-based model to at least bridge the gap in the near term until the number one, the storage capacity issue is resolved and number two, some of the energy. But quite frankly, I think the consumption-based model, this actually might help the consumption-based model, which has been a little bit slower uptake than one had hoped, but this actually might elevate the growth profile for that business. So let's kind of draw a parallel here because I think when people think about consumption models and storage as a service, they're kind of like, ah, this isn't the way I bought infrastructure. But if we think of other industries, like let's just think of the futures market, right? When we had unpredictability, if I take airlines, for instance, with jet fuel, the way that you sort of smooth out that unpredictability is you buy a futures contract and you know that you can now sell tickets and you know what your dependable cost is gonna be and you have a long-term agreement on those fuel costs. Is and do you view this new consumption model for infrastructure? And is that a fair way to put it? Yeah, I think data in general will be viewed more as a utility, right? You are going to see more and more of that. We're already seeing futures contracts for GPU capacity. And I don't see any difference in terms of storage as well, right? In terms of capacity utilization and how much data is being stored and utilized at any given moment. Because what we're starting to see is that data is becoming very fungible going from one area to another area. And given this whole hybrid cloud domain, it might not necessarily be stored in the cloud at all times but at peak areas, right? Much like the internet, you might want to increase that capacity for seasonal reasons, for let's just say compliance reasons until you can bridge that gap. So more sophisticated companies, I believe, will start thinking about data in more of a consumption model and a utilization model going forward. Yeah, I like how you put it there as a utility, right? Because one of the things you said was this fungibility and this mobility story that says it might live in the cloud, it might live on-prem, it might live at the edge, it might live in this cloud or that cloud. But that's kind of where a lot of the hyperbole breaks down, Woojin, is that telling me that I'm going to get a fixed asset, let's say a box that supports X amount of terabytes of data or petabytes of data, and fixing me to that box, that's really a glorified lease. That's not really an as-a-service consumption model. So do you have a way in your mind that you kind of really separate who's driving utility from who's essentially just giving me a operating lease on a existing bunch of hardware? Yeah, and that's the pushback I get from a lot of industry analysts, quite frankly. But at the end of the day, organizations and institutions really have to change their frame of thinking, right? From a physical asset, right? As in the storage, as a storage physical asset, but to the data itself, right? What you're seeing or what you're going to see is an exponential increase in the storage, right? The data that's being produced and that need to be stored, and it's going to get worse. It's going to compound in the AI era for the enterprises. We haven't seen that yet, right? What the companies like yours, as well as others can do, is take the data, compress it, and leverage the existing physical asset, right? And try to store more in the physical asset itself. So if organizations can really start thinking ahead or out of the box, literally out of the box, and think of it from more of a data layer, and I believe you're going to hear more and more of that going forward. And outside of the box itself, I think you'll start seeing more uptake and more embrace of this consumption-based model going forward. Yeah, and I'm going to get back to your utility model because you can really see the strength and robustness of the model when you start to see it play out. So once again, I'll use an analogy of internet access to my house. If I can go from 500 megabit to a gig, and I can go from a gig to two gig or even five gig, and I can do it seamlessly just by going on and clicking a button and saying, I'd like a higher level of service, that's a very different experience than, well, hold on, I got to send someone to your house, and I have to change what's plugged in, and I have to ship you a new router, and you have to go and install and change your wireless configuration. I mean, I honestly think that what clients want is that completely seamless view of, I want to update my service, I want to upgrade my service, and I want it to be completely seamless and non-disruptive. But I'm not sure that's what the industry has given them today. And I would absolutely encourage customers to take a look at what is being offered and really say, is this truly utility? Because I think that's the right lens to look through it. Yeah, I agree with you, Sean. And I do think that if the customers start asking for it, the vendors will start providing it. Well, more than that, if customers start to say, I won't accept your current model, then it forces others to innovate and change their model. But giving me the same thing with a different skin is way less expensive. And hey, if you'll buy it, I'll sell it to you. But I think the market is where this transformation takes place. And it's customers and enterprises saying, no, we don't want that, we want this. And at that point, things will move forward. So we talked about the supply crisis, we talked about NAND, we talked about DRAM, we talked about this impending kind of price increase and probably some sort of stability at 28, 29 and whatever the next super cycle is. We also talked about energy and the looming crisis of, where are we gonna get the next generation of power? And we sort of huddled down to this, if I'm an enterprise, my best opportunity here is to shift that risk. So maybe I don't buy infrastructure and I don't worry about how to power it. I actually find a vendor who's going to give me the infrastructure and even get a vendor who's gonna pay for that increased power cost for the data that's sitting in my data center. But more specifically, what would be your guidance to folks out there who are in this situation and trying to figure out, do I sweat my assets? Do I downscale my requirements? Or do I just change the way I'm buying infrastructure? I mean, what guidance would you give to folks out there? Yeah, look, there's always gonna be two camps, the wait and see camp and then the innovator camp, right? One of the things that I've learned as 25 years as an analyst is that companies that tend to wait in a period of destructive change, which is AI, right? They're the last ones to really catch up. So they potentially could be in a strategic bind over the next five to 10 years, right? Because they didn't make the investments that they needed to make, right? I think the companies that are forward thinking, making the necessary investments now and investing in AI right now are going to be the ones that could potentially be the innovative leaders and take market share for those who were stagnant and afraid of change, right? This has happened time and time again throughout history, especially in the tech space, but it doesn't necessarily mean, this doesn't necessarily have to be just a tech space, but in all industries. So, John Chambers, who I speak to quite a bit, so just a name drop here, disruption creates opportunity, right? And new ways of doing things. And there are new ways of doing things right now. And this is the time to explore new ways and adopting new things. Yeah, I think what I heard you say was, those who look at this situation and say, actually execution at this point is a huge opportunity to gain share, to disrupt existing markets, to disrupt existing players. And you only get these opportunities every 10, 15 years to actually take an industry and take a market and turn it on its head. We saw that in .com, right? A lot of companies disappeared, a lot of companies de-invested from internet. Those that kind of doubled down, let's just say Amazon.com as an example, took significant chunks out of the market that was there. Obviously you talked about Chambers, Cisco did it back in the rise of the internet. Let's bring it back to Wall Street. So if Wall Street looks at the situation, do you think they look at this price increase and this energy crisis and say, people are gonna batten down the hatches, not spend and there's demand destruction? Or do you think Wall Street's gonna say, hey, there's gonna be a ton of investment here. And those that invest have a tremendous amount of upside to gain customers that others will simply not be able to get to. Yeah, so if we look at the shares from a stock perspective, right? If we look at the shares of Dell, if we look at the shares of HPE and Everpure and NetApp, Dell's benefited from the AI boom. But all those shares on the traditional enterprise side, they've actually stayed stable, right? That's contrary to what one would think from higher pricing. Because higher pricing means gross margin erosion. If I think about it in the sense of what investors are thinking, what they're thinking is that for thinking enterprises are investing in their infrastructure right now, right? In what form, I don't know. But they're still waiting on who the winners and losers are in terms of how enterprises are investing and in terms of in what form they're investing. AI is the big driver right now, but at the end of the day, investors understand that the enterprise infrastructure is going to be the lifeline of what corporate America or global corporate IT is going to be. Yeah, it's interesting. We used to say that IT people who were not close to the business would never survive. It feels more and more that now IT is the lifeblood of every business. And those in business who are not closest to IT and the IT projects may not be in the best position. Two quick questions. So look, as we close this off, let's talk about a decision right now you think some customers might be making that they'll regret the most in 18 months. You go first and then I'll take a throw at it as well. I think you asked the question already, but those who waited for the memory crisis to blow over and did not make the investments that they need to make, they're gonna regret it a lot because they're gonna see their pricing for their infrastructure costs increase whether it's going to double or triple from when the first purchase order came in. So there is going to be some 50-50 hindsight to regret in terms of not making the necessary investments today. Yeah, I'm gonna take your investment piece and I'm gonna say, we're at a time right now where the best analogy I can come up with is it's kind of like when we had early cars, right? When the first cars were hitting the streets and there was no traffic lights and there was no stop signs. You look at folks like Henry Ford and others who took advantage of that opportunity and said, hey, look, we will go create the roads and we'll work to create the ability to license drivers. AI is a positive and it is also one of the most scary things out there. But organizations that are not empowering their teams in light of everything else that's going on, empowering their teams to see how could we do things differently? What would this look like for us? What would the company have to do to have this make sense? Could find themselves five years behind when these tools become mainstream and it's all happening so fast. It feels a little bit like the horse-driven carriage to automobile revolution. I'm old, but not quite that old. Okay, last question for you, Huji. So five years from now, we're gonna look back at 2026 and we're gonna talk about this moment, a little bit like how we look at COVID now, years later. What are we gonna say? I mean, who got it right? What happened here that we're gonna look at and say that was the pivot point that ultimately led to all of this future growth? Well, we're gonna rely more on the cloud, right? Because the cost of bearing IT inside the enterprise is going to get tougher and tougher. That's what I think is gonna happen. And I do believe that consumption-based models are going to win out eventually, because like I said, data is going to be considered more of a utility than actually only the hardware. Yeah, makes perfect sense. I always bring people back to the fact that it seemed very normal when I was in college to have a book of CDs that traveled with me everywhere I went. And now we buy music as a utility. We pay for a streaming service and whatever that happens to be. And that is by far and wide, the most efficient way to listen to music. So folks, Woojin, I wanna thank you for your time today. Thank you for your participation. Your insights are incredibly valuable. And for our listeners, I'd encourage you to please subscribe, like this episode, share it with your friends, but also visit us at thedatawire.com. And you'll find all of our episodes on Spotify and YouTube. And with that, for Woojin, I'm Sean Rosemaryn for Beyond the IT Headlines, and we'll look forward to speaking with you soon.
