Ways AI Data Center Capacity Will Reach 90GW by 2027 & How AMD, WDC Benefit

Nicole (0:00)

It's time now for our watchlist panel, and we're discussing the latest in data centers and the build out, and we know how much has been contributing to GDP. Lisa Martin, Business and Technology Analyst, the Lisa Martin Media Group, and Ron Westfall, VP, Networking and Infrastructure, Hyperframe Research. Glad to see you both with us. Goldman Sachs has been making some comments pertaining to this build out. Lisa, I'll start with you. How did you interpret it? What do you think it means?

Lisa Martin (0:26)

Great to be with you, Nicole. Thank you so much for having me. I think the Goldman Sachs news tells us a couple of things. The data center boom, it's becoming bigger, we know that. It's also broadening and it's also becoming a lot more physical. On the size front, we're moving from smaller to AI campuses that are measured, not in square footage, but hundreds of megawatts, eventually soon gigawatts.

On the red side, this is no longer just a GPU story, right? Data centers pull in around them this entire ecosystem, right? Compute, storage, networking, also cooling, power, et cetera. The constraints, those are changing as well.

Demand and capital for sure are there. We've known that for a while, but we do know that on the constraint side, power grid connections, we're also seeing water, and we can't forget the community acceptance. All of those are impacting really the speed of the build-out. What we're seeing is that this is really becoming a massive industrial infrastructure build-out. For investors, that both broadens the opportunities and the risks.

Nicole (1:30)

The massive capital expenditures is something you're watching as well, Ron. Tell me more about this as you think about this. The communities are pushing back to a certain extent, but these data centers are getting bigger and bigger, aren't they?

Ron Westfall (1:44)

Yeah, I think the momentum is already established and it will continue. To Lisa's point, I think this is something that's going to result in these five gigawatts of capacity into the end of this year. That is being fueled primarily by just the hyperscaler capital expenditure on data centers. That will probably reach $750 billion by the end of this year. We're really looking at what is driving this. Well, part of it is the fact that the high-density AI infrastructure is going to be required to continue the training for the frontier models out there.

Also, inferencing is becoming more and more prevalent. That will actually continue to leverage what's going on at the data centers, but also more distributed throughout the edge. That also gives more opportunity for enabling AI capabilities throughout the ecosystem. I think it's also important to add that when it comes to the US data center market specifically, that we're looking at a cage of 12 percent over the next three years through 2030 That I think aligns with the Goldman Sachs estimations. I think that the reasons for that is despite some hiccups, and despite some say local community concerns, there'll be plenty of offsets. We're seeing already that the data center bills are impacting election cycles, for example.

I think this is something that will continue to play out, where the growth is so inherent that even if there are some, I would say, minor opposition factors going into this and other constraints, this is not going to stop the overall momentum.

Nicole (3:21)

Right. Because we could see some slowing from community opposition or whatever, but you both seem to be on the camp there. We'll see that growth that's inherent and continues here. When we think of naming some names here, Lisa, what comes to your mind in the leadership and market share?

Lisa Martin (3:38)

A couple of them I mentioned, this is no longer just the GPU race. AMD and Western Digital. AMD makes this much more interesting because it isn't just a number two GPU alternate story anymore. I think AMD's data center business hit close to 7 billion last quarter.

Then within the last year, we've seen Meta, OpenAI, Anthropic. They've all committed up to, I think it's 14 gigawatts of AMD GPUs combined.

We know that power efficiencies also become part of the economics here. But I think the customer question is, how much AI can we produce from every watt, from every rack? And I don't think AMD has to win to replace NVIDIA. It just needs a bigger piece of a rapidly expanding market. On the Western Digital front, AI doesn't just have a compute problem, right? We're talking data economy. And so, of course, it's only going to continue, which means storage might be a kind of underrated, underappreciated component of the ecosystem. Training, as Ron was talking about, training inference, they get the spotlight, right? They're still compute heavy. But as we get deeper into inference agents, what's going to happen is this data flywheel, right? Continuously being created. And then, so we're not just creating new data. I think what's going to happen is historic enterprise data is going to be pulled back in for its value. And that's a tailwind for Western Digital that I think is there.