Why AI Data Centers Are Moving to 800V DC | Schneider Electric Live from Data Center World 2026

Hi everyone, my name's Victor Buenachea, with Schneider Electric, and I'm joining you here from Data Center World in Washington, D.C. today. Joining me is

Washington, D.C. today. Joining me is Jim Simonelli and Manav Patel, two other experts in the data center space who are part of Schneider Electric. You know,

you guys have been here since the start of the show, I think actually yesterday.

Uh, how's the show going so far? What do you guys think? Oh, it's great, you know,

guys think? Oh, it's great, you know, it's What's interesting is I've been to a few Data Center Worlds, but, you know, just kind of we're looking at the data hall right now, and it is extremely packed, very vibrant at this time of

day.

Uh, everyone's here, lots of buzz and discussion, and I can tell you the talks are very very thought-provoking over the past day or so.

Yeah, it's It's interesting everything's packed cuz I've been doing some power industry stuff for a while, and we're always talking about how density is getting higher, and the show is no different.

It's super loaded, it's busy, there's a buzz in the air. Manav, what do you think? How's the show going for you?

think? How's the show going for you?

Well, Data Center World this year it's exciting, you know, compared to past couple of events, this show is extremely packed, right? So many companies, so

packed, right? So many companies, so many peers here, right? A lot of exciting conversations.

Yeah, it's It's It's been great, you know, it's I've been walking around. The

word that everyone's talking about is AI, I don't think that's news for anyone who's even remotely involved in this industry.

Um, overall in the power industry, I'm coming from the AC distribution side, backup power side. We've been doing data center for a while, but now I keep hearing about with AI going on, DC keeps

popping up, right? The old Edison versus Tesla thing, right? What's What's going on? What's this comeback of DC? What's

on? What's this comeback of DC? What's

happening? Yeah, so what's interesting is is is oftentimes, you know, I think that when we think about AC versus DC, we think about it from the supply side, from AC versus DC. The reality is that

that is absolutely not the case right now. The driver for AC versus DC is

now. The driver for AC versus DC is exclusively around the compute side. So,

what's happening in Well, the buzz is here, and even at Nvidia GTC, is the fact, look, if you believe that an AI data center, as it evolves, the most profitable thing

to do is produce tokens. The best way to produce those tokens is to use GPUs. The

best production of those GPUs is to use the latest GPUs, but they also to get them to get them to be more the most productive, you have to pack them very tight. That means denser. So, we're

tight. That means denser. So, we're

dealing with things that are used to be 9 kW per rack, then maybe 20 kW, 30 kW, then a few years ago it was 60 kW per rack. Today, we are producing those same

rack. Today, we are producing those same GPUs in structures using Vera Rubin that are 250 kW per rack today. Now, all of that is supported by AC distribution

today. No real change in space. However,

today. No real change in space. However,

there's another click happening, which is now when you go to Vera Rubin Ultra, the density in that rack gets so so high, close to 600 kW per rack, that

bringing AC power at 400 or 480 V just becomes untenable. There's too much stuff in that rack that really needs to be just for compute.

So, one way to solve that has been is just serving the IT rack itself is to say, "What if I doubled the voltage from 400 or 480 up to close to 800 V

such that I have the current and then that means I just use about 1/4 of the copper just to bring power to that cable. That leaves more space for

cable. That leaves more space for compute. I can put more of that tokens

compute. I can put more of that tokens in that in that more tokens production in that same footprint itself. So, it's

completely driven by the compute side, not by the supply side itself. And that

has to do with taking out a lot of my conversion equipment from having to go from AC to DC at the individual component level, is that right?

bit. Yeah, yeah. So, the logical way to means is that is that it Think of it more as a bit of a copper problem and a bit of just being able to put take converters outside of that rack and move them move them move them someplace

else. That's the main driver for that.

else. That's the main driver for that.

So. Okay. Got it. Yeah. Manav, yeah.

Yeah, so like Jim described, right? To

me, I see this as a physics problem, right? That you need to get a certain

right? That you need to get a certain amount of power within a rack, and we believe over 400 kW it's not possible to get the traditional AC whips inside

the rack. So, if you increase the

the rack. So, if you increase the voltage, you can reduce your your current, right?

And you get higher amount of You get the higher efficiencies out of it all.

Exactly, right? So, some of those kind of a big drivers why over 400 kW we think that these physics work better in favor of direct current distribution.

There's multiple ways to address some of those distributions which we can talk about. Now, I I hear the Nvidia name

about. Now, I I hear the Nvidia name linked to 800 V a lot. What's that

connection between Nvidia and 800? So,

it's not so much Nvidia per se. They are

the leading driver of the GPUs going to that space. So, it really is around

that space. So, it really is around accelerated computing of which Nvidia is mainly the drivers. AMD has their own chips in that space. So, really what happening is here is that as you pack

the the the the as their evolutions evolve from their from their Blackwalls to Rubins, the power dissipation per chip increases. That

power dissipation puts more density as Monah was talking about and getting power into those chips with a reasonable amount of copper means that you have to change something. The thing to change is

change something. The thing to change is actually the power architecture to to increase the voltage to to decrease the copper.

And that's that's kind of where we sit and I know that you know, it's easy to say but there's a lot of things that have to change around that to to make that happen. And there's

a leaders going into it basically.

That's right. Can you tell me a little bit more about what these architectures look like? What do these systems look

look like? What do these systems look like? How is it changing things? Yeah,

like? How is it changing things? Yeah,

yeah. So, a few things here. So, one is you know, as we talk a lot about power right now, right? You know, which is usually because you know, things that aren't changing is these are all when you get up to this power level is these

are all still liquid cooled racks. It's

still the same type of liquid cooling heat rejection schemes in and of itself.

But how you bring power to the rack is very different. It's not typically an AC

very different. It's not typically an AC busway with bringing whips down into into racks. It's actually around doing

into racks. It's actually around doing the feeding and so one is step one is we call it a sidecar or power rack that we recently announced which allows Yeah, it allows

us to bring AC at a megawatt level into a rack right next to the critical IT load and that converts from AC right to internal DC with a small cable or small cable that runs right into the IT rack.

That is a very scalable and effective means to deploy AC without completely disrupting the data center architecture in in itself. So that's the first evolution that we have that that Schneider has been has been leading. The

second thing is that we see the next step is we call it centralized internal DC which is Now this gets closer to your your Edison perspective. We actually

have a distributing internals from the gray space or back room right out into the data hall. That is another type of architecture that also is evolving and it has other implications beyond just the converters

uh of of different sizes. Those can be anywhere from 4 to even 9 MW to that medium voltage low voltage fed type system. That's a big architectural

system. That's a big architectural bricks that Schneider is driving uh in those spaces as well, but it's not just around the conversion. It also is about the distribution and protection as

well which is all this stuff has to work as a system. Uh and that's partly where Manav and I kind of also are trying to collaborate. It's not just on the one

collaborate. It's not just on the one thing, actually about how this whole thing comes together. So we're thinking about the conversion, the protection and the distribution both from a localized sidecar perspective, but also how it's

going to evolve over the next few years to a centralized perspective which is the buzz of year here. A lot of buzz about the centralized perspective, but it's not here yet. Everyone is preparing

for that to prepare for when those chips arrive or those racks arrive which is you know projected to be late '27 or early '28 in that time frame. And and

you mentioned so many different systems like the distribution, the protection, all that.

Over the years AC has had so much evolution there, so much focus. DC has

always been kind of so much lower closer to the load. So, I think there's a lot of catch-up to be happening around that at the DC side. Is that why there's so much work happening around it, too?

Yeah, and and I think you know, between Manav and I, if you think about the the the notion of that like like we know how to do it. Technically, we know how to do this close breaking and protecting on DC versus AC. We do this in many systems

versus AC. We do this in many systems well, but to do it at scale and at a safe and scale manner that that that uh ordinary electricians and

licensed electricians and HJs can support that, that's really is the the increasing challenge in that space in that space as well. Which is really where I think the the the next evolution

comes is is maybe less around the things and more about how do you get the approvals to get the things in the market at scale.

Yeah, absolutely. I agree with Jim. So,

some of the constraints, right, which we the way we like to think is not about a component, right? You can have singular

component, right? You can have singular components with which work fine, but our approach is to solve this at an architectural level. Right, so we want

architectural level. Right, so we want to validate from from conversion, protection, distribution down to the racks, the system works in a harmony.

Right?

Or the key constraint which we are solving for is we want to deploy these systems at scale. Right? What data

centers care about is speed, time to market, scalability. Right? So, we want

market, scalability. Right? So, we want people want to deploy these at a gigawatt scale. We need to be ready to

gigawatt scale. We need to be ready to help them with deploying these. And Jim

also attributed it's not about one architecture, right? There's multiple

architecture, right? There's multiple variations of architectures which are possible, right? Minimum we see at least

possible, right? Minimum we see at least four to six variations of different type of architecture depending on what customer is trying to solve for, right?

A customer may have different constraint for the resiliency, backup, right? And and some of those

backup, right? And and some of those other constraints, right? Which is going to dictate what type of architecture a customer is going to choose for. Yeah,

and I hate to keep bringing it back to AC, but that's kind of something everyone understands. It's a good

everyone understands. It's a good comparative piece. AC evolved little by

comparative piece. AC evolved little by little over time, right? Since the

advent of electricity, right? Became our

standard. Now, the DC piece, it's going so fast, right? We need the capacity. We

need to figure out some of the solutions at scale, like you guys said, to quickly deploy some of these data centers. Are

those bringing some challenges that to the market overall, and how are we addressing those challenges? What's

going on to to take care of some of that? No, because that's a great point,

that? No, because that's a great point, right? So, some of the things, right,

right? So, some of the things, right, which are evolving is what we kind of attribute in the last conversation as well, is what we call as an indirect ecosystem. So, in in in addition to

ecosystem. So, in in in addition to having the technology to fill the pieces, we need to get the whole ecosystem ready.

And by ecosystem, let's first start with the codes and standards, right? The

safety comes first. We want to deploy systems which are which are scalable safely. Right? So, the codes and

safely. Right? So, the codes and standards, we are working very closely along with our industry peers to uplift the UL IEEE NFPA codes, to

name a few, right? We want to accelerate this now.

What we need to do is historically these codes and standards, they move slowly.

We're working making our best efforts to solve as an industry, how can we scale faster? How can we move faster,

scale faster? How can we move faster, right? To do this some of these

right? To do this some of these topics.

In addition to the codes and standards, think about contractors, you think about electricians, right? There is a skill

electricians, right? There is a skill set and understanding of this DC system.

In in general, it's it's on the lower side.

So, we need to work with consultants, contractors, electricians, even AJs, to help them understand these systems at a at a much better space.

And and you mentioned some of the our codes and standards involvement and things like that to drive the industry as a whole. Can you share a little bit more on what kind of involvement we have going on there? What What types of changes, what standards are we trying to

drive exactly? Yeah, absolutely. So, I

drive exactly? Yeah, absolutely. So, I

think we have been part of of what is called the current OS community, which was a pioneer for the DC protection system. Since then, we have joined Open

system. Since then, we have joined Open Compute Project. And our team, along

Compute Project. And our team, along with the industry, has been working very closely with with our peers and and the OCP community to create

framework around the white paper, right?

Which helps create some awareness what were the challenges to solve for. In addition, we are working with you and identifying the gaps, what are the poses and standards to be solved for, and we are working

with different bodies to accelerate some of those conversations. How are you finding the adoption? Because education

is such a big piece of this, I got to imagine, right? How's that going so far?

imagine, right? How's that going so far?

Everyone's on board, everyone's fighting in the same direction? Everyone in

general is trying to fight in the same direction, right? At the end of the day,

direction, right? At the end of the day, all of us want to deploy these systems as fast as we can and with what the market needs are to help us with this accelerated AI curve, right? So,

everyone is generally in the same direction, but a lot of work to follow.

Yeah, and I think the only thing I would I would add is that the governor of all of this is wow, we have good aligned interests between the compute providers, the data center builders, the equipment providers, safety is at the paramount.

So, at the end of the day, it's not just around dollars and cents, it's about doing it in a safe manner, which is why it does have to move at a deliberate pace. I can see you say that through

pace. I can see you say that through these bodies, which used to take years to kind of just collaborate to decide on what dinner might want to be had at the meeting, is that everyone's very very

focused to try to drive to some degree of kind of commonality to allow for scale, uh but also ensure that we are dealing with with safety and reliability

in the in that manner as well. If

anything, that will be the the governing headwind for deployment.

As you guys say that, thing that comes to my mind is kind of some of the NASA dynamic, right? We want to go fast, but

dynamic, right? We want to go fast, but we want to do everything safely, right?

We want to accomplish that mission together right?

So so the whole 800 V is coming from that power density, but that's just the power side of the story, right? That's

linked to the entire ecosystem of the data center with the evolution of that.

Are there other pieces connected into this other than just the density at the rack level? What else is involved?

rack level? What else is involved?

Yeah, so I think you know, just trying to wrap this up, you know, a bit into just kind of put a wrap around the whole system, right? Which is the fact that we

system, right? Which is the fact that we really think about this as an end-to-end problem. So 800 V is just one piece of

problem. So 800 V is just one piece of the of the power problem or power challenge. I don't want to call it a problem. But yeah, the ability kind of how do I provide 800 V? Well,

remember, building a data center is about building a system holistically. Think about it from a design, simulate, build, operate, maintain. Those things don't change AC

maintain. Those things don't change AC versus DC. They're just happening at

versus DC. They're just happening at scale. That is one of the things we have

scale. That is one of the things we have to keep our eyes on right now. Right now

there is zero 800 V racks in the system today, but it still is growing very fast. That's just going to add another

fast. That's just going to add another complexity probably in the late '27 or late '26.

So those but but the idea is that that's why I talked earlier about the fact that some things that that we've actually solved some problems very quickly incredibly and quickly have been on the heat rejection liquid cooling dimension.

We kind of have that problems solved technically. It's now about scaling that

technically. It's now about scaling that problem. We understand that if you want

problem. We understand that if you want to build these things with certainty, you better have the right software and the right simulation tools to be able to kind of ensure that when you build, you're building with a certainty

that that's required. And also now in terms of operations that you can actually visualize and understand what's happening at these sites. And

ironically, we're using AI to get more effective at operating these AI data centers as well, which is a big We're creating our own demand there.

Creating our own demand.

It's a good loop. I like that. So that

is kind of I added back up and say, "Hey, let's be clear that we don't over index or some interval conversation." A

lot of the fundamentals are the same.

Intervals is a is a critical area.

Primarily to enable the planet to begin to use this higher end computers, higher end tokens. It won't be every token is

end tokens. It won't be every token is is is produced by these ultra high density, but where it is is pretty important. So we better prepare to do

important. So we better prepare to do that. But that whole end-to-end system

that. But that whole end-to-end system from grid to chip and chip to chiller, exclusive of the software and services to maintain it across design, simulate, build, operate, maintain is really kind

of where the crux has to maintain and that's really what we have to keep moving moving forward.

So I think a lot of people are interested to learn a lot more about this topic. It's changing every day.

this topic. It's changing every day.

Anything you recommend for people to kind of keep up with what's going on?

Yeah, so I got to say that this this this topic started around white paper around 80 to volts. White paper 213 is the paper to read. If you want to get an overview of that of that topic and how

to think holistically about that. And

that's available on the Schneider Electric website. You got it.

Electric website. You got it.

So I mean, thanks guys. I guess one last question. Like, what's the most exciting

question. Like, what's the most exciting thing that you guys are looking forward to in this evolution?

Oh, I'll tell I'll tell you straight up.

I'll tell you the most exciting thing is how we can is is kind of what is what's what's next beyond? How do we use AI to actually

beyond? How do we use AI to actually self-improve ourselves? We touched upon

self-improve ourselves? We touched upon that right now. We're so focused on producing tokens. I'm really saying is

producing tokens. I'm really saying is how we use those tokens to really impact that industry moving forward.

Yeah, absolutely. And and for me it is I like to think about this is how do you solve problems quickly at an unprecedented rate at scale, right? So

this is an exciting time for us to kind of go ahead get ahead of the curve and solve all these problems at scale. Those

are much more insightful answers than mine. I just can't wait to have chat GPT

mine. I just can't wait to have chat GPT walk into the house and it does everything for me, right? So that's

really what I'm looking forward to it.

We're all driving toward that goal together right?

No, so the show's been great so far. I

think we still got some more time at the show over the next few days. I think we're going to invite

few days. I think we're going to invite everyone to come out to Data Center World, visit us [laughter] here at the booth. Thank you and we look forward to

booth. Thank you and we look forward to continuing this journey with you guys together and evolving the market. Thanks

for the conversation, guys, and and the education.