The Quiet Giant Behind AI

Bloomberg

12 hours ago

Transcript

00:00Well, when I started with Flex in 2019, we're a traditional contract manufacturing company,

00:05which means we make a little bit of everything for every end market, everything from health

00:09to automotive to industrial, maybe some data centers. And what we were doing, Matt, was,

00:16of course, making compute racks for data centers. But we had this little business that was making

00:22power that powered the chip itself, like the NVIDIA chips or any chip manufacturer.

00:27It was a small little business. And my whole legacy is with electrical industries.

00:32I came in and I thought, well, compute is going to get power hungry. And this is before the days,

00:37you know, everybody was talking about NVIDIA and OpenAI and thought, you know, why are they doing

00:42all this power stuff? It's going to get intensely power hungry. So let's buy more things that puts

00:48compute and power and cooling together. So we ended up cobbling together a whole bunch of acquisitions.

00:54And one of the first companies that thought about compute is going to be power hungry.

01:00Power is going to generate a lot of heat, which means you're going to need a ton of cooling.

01:05Let's try to bring all these three things together and provide a value proposition to the customer

01:11because it's so fragmented in how they're thinking about it. So I think it's just my history of being

01:17in the electrical space that helped this. And so we invested in that and built a business that today

01:23is around six and a half billion dollars, growing at 35 percent. And it's 25 percent of our portfolio.

01:28It's interesting because at the time, I imagine, you know, Bitcoin miners would have wanted to buy

01:34your products as well. So were they an early first big customer and then transformed into a data center

01:40play? No, we're always with hyperscalers and colos. So Bitcoin miners being co-location

01:47co-location companies. So those were the two major players. We've got a lot of neoclouds today that we

01:54work with. But traditionally, if you think about hyperscalers and colos, they're the ones designing

02:01the compute. And compute, by the way, is going from being 10 kilowatt racks to one megawatt racks,

02:08generating a ton of heat. And you have to cool it down and you have to make sure power is migrating

02:13safely across this value chain. And that's what we do. By the way, you essentially control the last

02:20mile or the last, let's say, 500 feet of the U.S. power grid. Right. What do you think about the

02:27problems we hear about in the rest of the big grid, you know, that the utilities are feeding into?

02:32Yeah. So, Matt, I've spent my last couple of decades in the energy space. And I recently say this often

02:40is, you know, as to go and see all these utility CEOs. And, you know, when I ran Eaton's electrical

02:46business and begged them to spend money to upgrade the grid and you could you wouldn't hear anything

02:51from them. Now, I think it's super cool that the data center acceleration is forcing us to talk about

02:59upgrading the U.S. grid because it's so important. Our transmission systems are outdated. Our distribution

03:05systems are outdated. So I'm really excited that this is the chance for the U.S. to upgrade its

03:11infrastructure for energy because without good, cheap energy, you know, and I know there isn't good

03:17economic growth. So we need to make this happen. And the spending that you're seeing that is being

03:23announced now, which is not just in distribution, but also in transmission, will help this pretty

03:28significantly. I mean, we're starting to understand your role in the energy to compute sector is the

03:37administration reaching out to you as an expert in this area and asking what can be done to improve

03:43things? Well, I think the administration has put a lot of focus on providing the right energy policy for data

03:52centers and industries to be able to invest in this kind of an accelerated spending. Without the

03:59administration support, that wouldn't happen. I think that's one big thing where we play a role. And then

04:05the second is being one of the world's largest manufacturing companies. Everybody is talking about U.S.

04:11manufacturing. We definitely play a role in having the conversation about what does it take? How do you make

04:18sure you have good cost effective manufacturing systems and capability in the U.S. and how we can

04:24support? So we're very excited to be able to help play a role in this. So you're building a lot here

04:29in the U.S., but you are present in, I think, 25 or 30 other countries, right? Obviously, this AI data

04:37center build out is something global. Outside of the U.S., where do you see the most demands, the biggest

04:44investment, the highest capex? I would say our biggest presence is in the U.S. and in Europe. So we have a very

04:50significant presence in Europe, in Ireland, in the U.K., in Poland, where we kind of supply to almost all of

04:58Europe for data center infrastructure there. And we don't have a big Asia-Pacific presence just because of the

05:05geopolitics issues associated with it. And I would say European infrastructure spend is okay, but the U.S.

05:13is now basically consuming most of the infrastructure spend for most hyperscalers or colo. Everybody's

05:20focused on the U.S. is going to go from what used to be 20 gigawatt of data center power to 100 gigawatt in

05:27the next few years. So we're all focused on making sure the U.S. has the right resources and capability

05:34today. I'd say European spending is happening, but not to the scale that we're all allocating capital to here.

05:40By the way, 100 gigawatts is like 10 Manhattans. Correct. Are we going to eventually have as much

05:47compute power as we currently use to power the nation? I mean, that's what it seems like. And even

05:54that doesn't seem like it's enough, right? We have some eight trillion dollars of capital spending being

05:59announced in this space. I think it's going to be needed. And as a manufacturer, Matt, it's important to

06:06see what I see, which is that the A.I. journey has barely started because we're talking about consumer

06:12use of A.I. today. You look at manufacturers like us, productivity in the manufacturing space has

06:19stalled for the last decade. When we use A.I. in our factories, it's going to be a game changer for

06:24productivity that we're going to see. So I'm a huge fan of we're going to get more and more compute

06:31intensive. And that's going to happen not just in consumer behavior, but in factories like ours.

06:37And that journey is really just starting. So 100 gigawatt is definitely happening because

06:43industries like mine need it. By the way, Ravati, so when we talk about data centers,

06:47it's power that's the scarcity we're focused on because you don't really need a lot of people

06:52working in data centers. But you do need a lot of people to manufacture your products. How are you

06:56finding the scarcity of labor? I'd say labor is scarce. It is difficult, not straightforward. One,

07:02because, you know, I have to say it's not that everybody wants to apply for manufacturing jobs.

07:08It's not the coolest, sexiest thing in people's heads. But in the cities that we are in, I would say

07:15they're a very important part of the local economy. And we provide good jobs that takes the skill sets

07:21that people has and really build on it. So, you know, we have a lot of open jobs. We're competing

07:27with many different industries for it. I think availability of labor is the biggest issue in the

07:33United States today. But I like the fact that the in the cities that we are in, we provide employment

07:41that people want to be in. And I think manufacturing is cool and sexy. I'm not sure what everybody else

07:46thinks about it. So, you know, I've grown up in factories all my life. And so it's a tough fight,

07:51but I feel like we're making progress on that.

Category

Transcript

Be the first to comment

Recommended