Michael Dell on the 3,000-Pound Liquid-Cooled Racks Powering AI Data Centers — Transcript
Full transcript
- 0:00Elon is probably the most demanding
- 0:02customer. No, he's definitely He's
- 0:03definitely the most demanding customer
- 0:05there there is.
- 0:06>> Dell is best known as the king of
- 0:07direct-to-consumer PCs in the '90s and
- 0:10early 2000s.
- 0:11>> Dude, you're getting a Dell.
- 0:13>> The computer giant is now one of the
- 0:14most important manufacturers of the
- 0:16infrastructure behind the AI boom.
- 0:19>> Dell entered the server business in
- 0:211989.
- 0:22>> Wow.
- 0:23>> When I was 24 years old. So, we've been
- 0:25at it It didn't actually work the first
- 0:27time. Okay, it didn't It started working
- 0:29around the mid-'90s. You know, of course
- 0:31we got into storage shortly thereafter.
- 0:34Then we acquired EMC. And so, we've
- 0:37actually been in this business a long
- 0:38time and we are number one at it, right?
- 0:41And so,
- 0:42>> You You mean by
- 0:44like revenues, by volume?
- 0:46>> Exactly. And so, it shouldn't be too
- 0:47surprising that we're also number one in
- 0:50AI infrastructure.
- 0:51>> Dell did half of the servers in the
- 0:54Colossus one AI supercomputer for
- 0:57SpaceX. And you did it really fast,
- 0:59right? I think that was a record-fast
- 1:01build. It was like 6 weeks or something.
- 1:03>> Elon is probably the most demanding
- 1:05customer. No, he's definitely [laughter]
- 1:06He's definitely the most demanding
- 1:07customer there there is.
- 1:09>> Yeah. Was he like Was it Was there a
- 1:11bonus involved? Get this done in 6 weeks
- 1:13and
- 1:14>> I don't know if there was a bonus
- 1:15involved, but they have
- 1:18rewarded us with additional business
- 1:20because we've done a great job and
- 1:22that's pretty much the pattern.
- 1:24>> Dell has about 40% of the market for
- 1:26sales of powerful servers to companies
- 1:27in North America, but it also has stiff
- 1:30competition from number two Supermicro
- 1:32as well as HPE and Lenovo. And AI
- 1:36hyperscalers like Meta, Microsoft,
- 1:38Amazon, and Google design and build many
- 1:40of their own servers.
- 1:42>> A substantial portion of your business
- 1:44remains sales to I mean, companies like
- 1:47Dow Jones, where we are, right? Like We
- 1:49walked We toured the newsroom. You saw
- 1:50the Dell monitors, [laughter] the the
- 1:52Dell PCs, all the rest. That is still
- 1:54tens of billions in revenue for you
- 1:56every year. But the the growth driver is
- 2:00like to me this very peculiar almost
- 2:02alien object which is the I call it the
- 2:04AI supercomputer. 6 ft tall, it's 2 ft
- 2:08wide, it's 3 ft deep, but it weighs
- 2:093,000 lb and it draws as much power as a
- 2:14hundred typical US households or maybe
- 2:16two or three hundred if it's like a
- 2:18really high-end. And this is the brains
- 2:21of these new AI data centers.
- 2:23>> These systems are incredibly
- 2:24complicated. It's not just a rack with a
- 2:27bunch of chips in it, right?
- 2:29>> No, it's full of plumbing now.
- 2:31>> Well, it's you know, literal plumbing.
- 2:32Millions of miles of fiber optic cable
- 2:36and engineering the liquid cooling to be
- 2:40super power efficient, the storage, the
- 2:42networking, the deployment,
- 2:45you know, teams that come in and make
- 2:47these things all work and then keep them
- 2:48running all the time.
- 2:50>> Right. Because each one of these boxes
- 2:51is like six figures.
- 2:54>> They're they're they're pretty
- 2:55expensive, right? And and and uh we we
- 2:58have learned a lot about how to quickly
- 3:01deploy these because we have this
- 3:04demand.
- 3:05>> Let's talk about lessons from your
- 3:07[clears throat] big like neo cloud like
- 3:09hyper scalar customers, you know, the
- 3:11the folks that are standing up these
- 3:12these big data centers that we read
- 3:13often about in the news. You're
- 3:15transferring these lessons now because
- 3:17companies are coming to you and saying
- 3:19we want our own AI which means we have
- 3:22to have our own AI infrastructure,
- 3:24right?
- 3:24>> Companies want control of their data,
- 3:27their their economics, their kind of
- 3:30privacy, their security. So, let's say
- 3:32you have a hundred factories. You're a
- 3:34manufacturer.
- 3:36You actually want to have the AI inside
- 3:39each factory, right? Doing inference to
- 3:41help you be very effective. And so, yes,
- 3:44we're we're building these out. It took
- 3:46us eight quarters to have 3,200
- 3:49customers building these Dell AI
- 3:51factories.
- 3:52And in the last three quarters we've
- 3:54added 3,300 customers. So,
- 3:56>> So, it's doubled. It's it's more than
- 3:57doubled 6,500.
- 3:59>> And what's also interesting is when you
- 4:01look inside these companies and actually
- 4:03what they're doing with the
- 4:06with their data. They're just kind of at
- 4:08the beginning of figuring out how to do
- 4:10this.
- 4:11>> You're one of the few CEOs who was also
- 4:15a CEO of well, the same company
- 4:17obviously during the dot-com bubble.
- 4:20We're in the middle of an infrastructure
- 4:22buildout that according to economic
- 4:24historians is bigger than any in
- 4:26history. Like as a percentage of GDP
- 4:28it's bigger than the railroad boom. Do
- 4:30you have any concerns about
- 4:32overcapacity? You know, having seen a
- 4:34lot of these waves over the last 42
- 4:37years, we take a pretty measured view of
- 4:40this.
- 4:41And we think about the risks and the
- 4:44challenges and how we balance all that.
- 4:46You know, having seen the the dot-com
- 4:50period
- 4:51>> Which Dell survived, not just intact but
- 4:53pretty well.
- 4:54>> Sure. And and and you know, our plan is
- 4:56to survive this as well and and to come
- 4:58out stronger
- 4:59>> Are there similar strategies? I mean
- 5:01Dell famously you've always kept
- 5:02inventories low, you were building on
- 5:03demand. Are you still doing that?
- 5:05>> Sure. I mean we we're we're doing it to
- 5:07the best we can, right? There are some
- 5:09new challenges that this presents.
- 5:12But if you look at our cash conversion
- 5:14cycle, it's still negative and we, you
- 5:16know, have an incredible focus on a lot
- 5:19of the fundamentals.
- 5:21>> Well, let's talk about those challenges.
- 5:22I think on earnings calls some of the
- 5:24other executives at Dell have said, you
- 5:26know, our number one challenge is DRAM
- 5:28DRAM DRAM. is memory, memory chips.
- 5:31>> Right.
- 5:31>> And of course this is a problem that
- 5:33everybody has. Like if you're trying to
- 5:34buy a laptop, this is a problem.
- 5:36If you're trying to buy a PlayStation,
- 5:38this is a problem. If you're trying to
- 5:39buy phones, the price of those are going
- 5:40up.
- 5:41>> Cars, right? Because they need memory.
- 5:43Do you see that bottleneck easing at
- 5:45all?
- 5:45>> I think it's going to get worse in 2027
- 5:48based on what I see right now.
- 5:49>> crunch.
- 5:50>> Yes. And and the reason is that it takes
- 5:54about 4 years to build a new memory
- 5:56factory. There are a number of them
- 5:58under construction. You know, Micron for
- 6:00example, we're a big Micron customer.
- 6:02>> Right. Like they've pledged $30 billion
- 6:03dollars a
- 6:05construction here in the US.
- 6:06>> Exactly. And And those factories will
- 6:09start producing in 2029.
- 6:12Okay.
- 6:12>> And who knows where [laughter]
- 6:13everything else will be then.
- 6:14>> And so, maybe maybe 2029 everything will
- 6:17be fine, right?
- 6:19What I see is the way people are using
- 6:23this technology now is very different
- 6:25than it was even a year or two ago.
- 6:27>> With AI, you mean?
- 6:28>> With agents. Yes, exactly. And so, the
- 6:32demand is growing faster than supply. It
- 6:35takes 4 years to build these factories,
- 6:38so I don't think that the shortage is
- 6:41going to quickly resolve itself.
About this transcript
This page contains the full transcript of Michael Dell on the 3,000-Pound Liquid-Cooled Racks Powering AI Data Centers by WSJ Podcasts, generated from the public captions YouTube serves with the video. The transcript has 1,229 words across 198 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.