HPE's AI Servers Ready as Soon Data Centers Are, Says CEO

Bloomberg

Watch HPE's AI Servers Ready as Soon Data Centers Are, Says CEO - Bloomberg on Dailymotion

Transcript

00:00We're actually getting a really good insight through HPE into what's happening in real terms

00:05in the real world of data center build-out. We had it with CoreWeave just a few weeks ago.

00:10A part of this, right, is a delay on a specific data center project. And you can't book revenue

00:17until that project goes ahead. Is this a one-off? And if you could, just explain to the Bloomberg

00:23Tech audience what is in your power to control and what is still happening out there in the

00:29supply chain for data center build-out that's holding things up?

00:33Well, good morning, Ed, and thanks for having me. Look, in Q4, we posted a record profitable

00:38quarter where we saw revenue growing 14 percent and profit growing faster at 26 percent. And that

00:45allows us to exceed both EPS and free cash flow guidance. So capping a very strong year. But on

00:51the AI front, we continue to see strong demand. We booked another $2 billion in orders. More than 60

00:58percent of our orders are in sovereign enterprise. And as you stated, there were a couple of deals

01:04that slipped into 2026. One of them was for data center readiness. It was slightly delayed. And two

01:12were because of the U.S. government shutdown. We couldn't deliver the system and get it accepted.

01:19But in general, sovereign are longer, right? Longer cell cycle, longer acceptance cycles. But we are

01:25intentional on that because, you know, we are focused on profitable growth in AI. And now we have

01:31a backlog which is over $4.7 billion. So ultimately, it can be data center. It can be also customer

01:37speaking different technologies like Vera and Rubin and maybe waiting for AMD. But the reality is that

01:43there is more back and loaded. That's why we got it a year the way we did it. But ultimately,

01:48we are reaffirming at 17 percent to 22 percent revenue growth guidance in addition to raise the

01:53EPS and free cash flow guidance. That data center lumpiness and indeed the holdup that you saw.

01:59What is it? Is it supply chain issues? Is it cooling? Is there something that tends to be the holdback?

02:07It's a combination of many things, Caroline. You know, I think sometimes, you know, just the real estate

02:14takes time. Bringing the power and cooling takes time. In some cases, you know, the equipment,

02:20right, slightly delay because you're talking about turbines for, you know, power generation. In some

02:26cases, you know, it can be low level components. But look, these are very large build outs. We're

02:32talking about, you know, tens of megawatts, sometimes hundreds of megawatts. And now we're going

02:37through gigawatts. And so we just need to recognize that it takes time to build these large build

02:44data centers. And ultimately, we will be ready to deploy as soon as these data centers are ready

02:49to accept the systems. But also the working capital takes also longer to get through.

02:56So you just referenced Vera Rubin. And I'm trying to understand what you meant there. You know,

03:02NVIDIA has this, we've discussed it, Antonio, several times, this commitment to annual update to

03:08the technology. And are you saying that there are customers out there that just want to wait

03:14for the latest gen before they commit at scale to a big project?

03:21Well, there are customers already in the current generation, the generations before that feel now,

03:26I need to get another bump in performance so that can lower my cost per token training as I go forward.

03:33So it can be a combination of both. And depending on where they are in the build out of the data

03:38center, it makes sense for them to go to the next generation. But also we see an interest, you know,

03:44in getting the choice and the flexibility to build these data centers the right way. And that's why,

03:50you know, we at HP now with the addition of Juniper, we are becoming a networking center company. And this

03:55past week, just in Barcelona, I came back yesterday, we announced a number of amazing technologies that

04:02allows customers to adopt whether it's Nvidia, or whether it is AMD. And we announced the first

04:08scale up Helios switch that allows them to adopt this technology. So it can be a combination of

04:14things, depending on where they are in that build out cycle.

04:18Our Bloomberg intelligence analyst, Woo Jin-ho, always loving the versification of the business and the

04:23leaning into the networking, Antonio. But just going back to with the data centers and the margins,

04:27that's always been the question, like AI servers, the cost is memory. You seem pretty confident.

04:32Why are you confident in just having the handle on the shortage that we've got in memory?

04:37Yeah, absolutely, Cara. Well, look, I'm very pleased that we'll return the entire server segment,

04:43inclusive of AI, to approximately 10% operating profit in Q4, which was our commitment since Q1.

04:51And regarding cost of commodities, it's first to, we need to recognize that the cost of commodity will

04:57be driven by the shortages we are going to experience in the latter part of 26, starting the

05:03second half, particularly in DRAM, and obviously, as well, NAND. And we already enacted price increases

05:11in the month of November. And we already factor what we think is going to happen, at least the best

05:18line of sight we have in our guidance. And that's why not only we reaffirmed the revenue guidance,

05:24but we also raise our non-GAAP EPS guidance, considering what we see going forward.

05:30Antonio, I'm really interested in the differences in the demand profile of, say,

05:36Vera Rubin and against Helios. You know, Helios, the pitch is its scalability, but are you getting more

05:43interest there at the edge in terms of use cases? Just trying to explain the real world differences

05:48of what you're seeing among your customers.

05:52Yeah, I think, you know, enterprises are accelerating adoption. I give you my own example

05:58here at HPE. We already have 100 plus use cases in production using AI across a number of functions

06:05and business units. And obviously, we use AI for our own products, which is one poor differentiator

06:12for us in the market. But in terms of where the work is being done, it's done everywhere. But we see

06:18now the growing in inferencing, which tells you that AI is being deployed. And there are a lot of use

06:24cases, whether it's manufacturing, transportation, healthcare, where the inferencing is cheaper

06:29at the edge, where things takes place, where the data is generated. And then, obviously,

06:35is the training aspect, which can be a combination of locations. And in Europe, for example, this concept

06:41of sovereign AI clouds is super important because they want to preserve the sovereignty from a model

06:48perspective and the data. So it's a hybrid design by design. And that's why I always say AI is the true

06:55definition of a workload.

Category

Transcript

Be the first to comment

Recommended