Ep. 015 - DG Matrix Explains 800V DC vs Legacy AC Distribution (Datacenter, Energy)

Hello everyone. Welcome back to

SemiAnalysis Weekly. I'm Jordan today.

I'm joined by Nico, Jeremy, and Haroun,

our first ever guest on the podcast. Up

until this point, it's been all

SemiAnalysis people. Uh but now we're

bringing on a guest because 800 V DC

adoption is

too important. Uh we need to bring in

the experts. Um

Yeah, Haroun, welcome to the show. Do

you mind starting by uh introducing

yourself to the audience and and what

you guys do?

>> Sure. Um so, Haroun, co-founder and CEO

of DG Matrix.

And I got [clears throat] into power

electronics when it was extremely

unpopular and a very uncool thing to do.

And and so now it's become much more

popular. I've had a chance of of working

on anything from computer room power,

which was a precursor to data centers.

I've done solar inverters before the sun

started shining on that industry. I've

done transmission power flow control at

hundreds of megawatt level using power

electronics. And even had a chance to

pioneer some of the electronic jet

engine starters. Uh all power

electronics based on the Dreamliner and

the Joint Strike Fighter. So, very and

and and now I think we're we're sitting

on an extremely exciting era for

humanity, where the 800 V DC

architecture is helping us propel propel

the human race to to superhuman

intelligence.

>> Awesome. Nico, why don't you uh kick

things off with like one of you know, a

few questions. I I know you guys

collaborated a little bit um to create

the

>> Yeah.

>> part one of the 800 V DC Revolution

article.

>> Yeah, sure. I mean, I think Haroun

already mentioned that the recent main

topic for this conversation is going to

be 800 V. I believe that it's been one

of the main trends that we've been

hearing all around during this 2026. We

go to conferences and nowadays 800 V

pretty much everywhere. All companies

are, you know, showcasing their side

cars, their prototypes, pretty much

everything. So, you know the obvious

question and before we get into the

solid state transformers and all the

cool stuff that we're going to cover

today is, Haroon, like, why are we

discussing a 100 V

in 2026 today and why are we discussing

a 100 V when we think about those 1 MW 1

MW racks?

>> I think the the compute power required

for GPUs and synchronicity is increasing

to a point where legacy AC architecture

and standard strands of AC cables are

unable to carry the power. And and and

so the question is, how high can you go

in in in voltage that you can you can

lower the the the cost and and and

remove the constraint of the copper

delivery system? And so taking a 240 V

AC single phase * 3 phases to to 800 V,

effectively root mean squared and root

mean squared,

you you you're going to get almost

triple the power on on on the same

copper cable provided you can handle the

distribution. And I think those

economics are are what's taking it to

800. So, the second question would be,

well, why not 1200? Why why why is it

800? Why is it not 950? And and I'm

going to venture a a guess, an educated

guess, that that one, it has to do with

EVs that that developed a lot of 800 V

architecture, but it also has to do with

the fact that semiconductors, the most

popular semiconductor, is the 1200 V

device that that that is used in motor

drives all over the world. And so when

silicon carbide and wide band gap came

out, they sort of came out for a 1200 V

architecture and and that when the

device can take that much, then 800 V is

a good safe voltage to to settle on. And

maybe that's where some of the genesis

of 800 V is the power density, the

ratings.

>> Okay, so I mean to put it simply for for

the audience, when you say enable to get

the power required for these 600 kW 1 MW

MW rack, is it a matter of you know, we

hear a lot about the weight of the bus

bars because of the copper amount they

will need to you know, distribute all

the current required

to to to get into that power levels. Is

it a weight matter? Is it a cost uh

discussion because of you know, we know

about the price of the copper cost is

just going up like crazy. Is it a matter

of efficiency to lower down current and

therefore I square I square R losses? Is

it a bit of everything? In your opinion,

what's the main you know, driver for

this whole uh revolution we we'd like to

call it?

>> I think you you guys are very good at

understanding the the physics and the

economics, right? Very impressive work

that Semi now says does. So so you've

hit all the points and and I think it's

it's

it's how much current can you get and

and generally when you raise the

voltage,

you know, as long as you have the

separation, the creepage and clearance,

raising voltage to get more power is far

cheaper than raising current to get more

power. And I think you're you're right.

So you're getting far more effective use

out of the same um out of the same

copper.

>> That's great.

Um okay, I I I think you know, the the

foundation of why we are talking about

that about this today, it's it's now

clear. Um of course like we are like

being asked a lot on timings like oh, is

this something that's already happening?

Is something that this is going to you

know, start kicking off in two years? So

just to put simply, in your opinion,

when does 800 V become a necessity? when

a 100-V is more of a, let's say, still

in the face of proof of concept?

Um of course, we have this news the the

article with all these faces.

It's not I I know it's not a short

answer, but in your opinion, when does

the 800-V revolution really start to to

kick off?

>> I think it's it's uh started. The The

question is

when does the right architecture from

Nvidia come out that starts driving the

demand?

And And when that comes out, you can't

say, "Well, I'm going to invent 800-V

architecture now." You You've got to do

it up front. So, in a way, that that

that maturation of technology, the

maturation of the manufacturing

approach, the build-up of the supply

chain, and or or or migration of the

supply chain from from the EV side to to

the to the to the side of data centers

has started already.

And And uh And but there's a big

question. And And it's interesting. We

get this question a lot. And And And And

this is where we've come up with with a

very unique solution using our

multi-port transformers. The question in

customers' minds is, "Well, how much

percentage will DC be, and how much

percentage will be AC?" If I go

exclusively DC, and adoption rate is

less than we want,

are we going to be left with stranded

power? Are we going to be left with a

stranded investment? And if we ignore

AC, and and those new companies that are

coming out with chips that are going to

run, let's say, less power or run on an

AC architecture, then what do we do? Now

Now are we going to be in the same

situation? And I think the answer we're

gravitating towards is it shouldn't

matter. You should have an architecture

that's highly flexible that can do AC

and DC in any percentage that you want

right from the same uh same product

line. And I think that's why people are

so interested. So, we just want to

immunize the financial risk for the

developers, the neo clouds, and anybody

else on on how much is DC, how much is

AC,

and uh reduce the risk on the timing.

>> Yeah, because I mean and I think you're

you know, your last comment is a

wonderful

uh

you know, opportunity to introduce your

your multi-part products and how the the

value proposition of your multi-part

solutions. So, just for the audience

that's not maybe that familiar with

Digitmetrics and the multi-part

solutions, how do your solutions uh work

and how do your take like different

inputs, different voltages, different

frequencies?

>> So, one of the things that that I did

back in 2011 to 2013, when I was working

for an SST company, is we were doing we

were we were developing an SST, a

solid-state transformer, to do AC

conversion to AC conversion. And we were

pitching like, "Oh, it'll clean up the

power. It'll do this. It'll do that."

But we realized after 2 years of work

and and spending millions and millions

of dollars, that that was probably one

of the dumbest things we could have

done. Why? It's because you're taking a

a hunk of iron and a hunk of copper

wound around it with some insulation

that's going to last 40 50 years. You

know, and and it's going to last in heat

and it's going to last in thermal

cycles. Why the hell would anybody in

their right mind try to replace that

with a bunch of electronics that are

going to be more delicate and and are

going to cost a lot more and going to be

less reliable. Why would anybody do

that? And and so, as we started to ask

that question, the answer was, "Well, it

might be a great science experiment, but

that's where SSTs are going to stop."

However, the answer came that if you

were look if you look at what happens to

that AC after you transform it, do you

do variable AC with it? Do you do a

motor drive at the end of it? Do you

take a medium voltage and convert it to

a low voltage? What do you do? And and

so the answer is, well, if you combine,

for example, the rectification function

after the AC and and you put it all in

an SST, holy moly, now you've got a

balance of system that's actually

cheaper, it's more reliable because it's

integrated, it has less margin stacking

cuz it's coming from one company, and lo

and behold, you found the first value

proposition for for the SST, but it's AC

to DC conversion. So we started thinking

further about it. Well, we're like,

well, anybody can do that, you know,

what what do How do you differentiate

that? And so we came up with this crazy

idea

that if you're adding a port that does

DC, you're adding much more value cuz

you're collapsing a lot of the system

that happens afterwards. So why not look

at more ports? So we said, what if we

added more AC ports? What if we added

more DC ports? What if we could make

every port bidirectional? And and and

and so you should holy well, the word is

so something else, but I'll replace it

with holy moly. We said, holy moly, look

at the value that you will add here, and

and you could replace a statcom, you

could replace a UPS, a rectifier, you

could replace the energy management

system, you could replace behind the

meter energy aggregation, all with a

multiport SST. And we said, boy, that's

the holy grail. That's what we need to

develop it because the economics and the

physics are all in your favor. And and

so as we went down that path, we didn't

realize that it it the controls, the

cooling, the electromagnetic

interference, the density, that we would

have so many brick walls we would run

into. So it took us, you know, at least

700,000 engineering hours to get

multiport to a point where we could

start doing deployments all over. And

and that's how we came up with multiport

is basically the and physics driving

innovation.

Mhm.

>> Okay, that that that's fascinating.

That's That's truly fascinating.

Um Yeah, I think you you know, touch

upon incredibly

interesting points and I really don't

want to be jumping, you know, from the

very beginning of the conversation we

are having now until the that end state,

but you you mentioned like, "Oh, what if

we take all these functions that

currently UPS systems cover, all these

other part of the legacy in electrical

equipment." So, just first question

before we go back into, you know, where

we are today, but in your view, when you

think of the of the data center of

of a data center in in in 5 years, in 10

years, how does it look? How does the

electrical architecture look?

>> I I think uh clearly as densities

increase, right? The the number of um

the the intelligence goes up, the the

number of uh points that you uh compute

goes up. The the the token the the cost

of a token in kilowatt hours goes down.

The question is what is going to drive

that metric, right? Is is the cost of

the token per kilowatt hour, assuming

everything else is depreciated, it's

going to come down to power, right? So,

when it comes down to power, it's power

in tokens out. So, how do you get the

absolute lowest cost of that token? And

and and and how do you maximize that

infrastructure is the answer. And and I

think the voltages are probably going to

go up at some point. People are already

talking about 1,500 V DC. I think the

density of the racks will probably go

up. And um and uh the racks are going to

get smaller and smaller and smaller, and

the power infrastructure also has to

follow a similar similar thing. That's

where collapsing multiple systems into

one makes sense. Not only do you get rid

of a whole lot of copper and iron and

junk, but you have far better

functionality to eliminate stranded

power and supply those dynamic loads.

So, that's where I I I think it's it's

going to end up in a far denser

environments with even more integrated

cooling and

>> That's That's fascinating and yeah, it's

like you Now that we have you here with

us today, great pleasure to have you

here. It's just taking the opportunity

to just go into your pick your brains

and like know how you are envisioning

these data centers looking 5 to 10 years

out. But yeah, let's

I would say like let's go back to to to

the present. Let's go back to today. Um

today we are early early days of this

whole revolution. We're still even at

the point that we hear about 100 volts

as a whole, but you know, when we look

deeper into the systems

look

we we know about some hyperscalers

working with plus minus 400 volts. Some

others

are working directly looking into into

single-ended 800 volts. Just again, put

it simply for for the audience, for

everyone to to understand the

implications of

you know, as an industry or from the

perspective of Digi-Key Matrix, how you

approach this? What are the implications

of going plus minus 400 volts or going

directly to 800 volts?

>> So, so the interesting thing is I think

I think the the the question is what's

driving plus minus 400 volts versus 800.

And is it a balanced plus minus 400 volt

load? That's That's the first question.

And and I I tell you what, when we did

the Dreamliner,

um it's interesting. Whenever you fly

something at those altitudes of 30,

40,000 ft, the air is very different.

The ionization of of of insulation

happens in a in a way where you degrade

insulation above 300 volts. So, the

magic rule is you don't want to go above

300 volts. And and so, as density of of

power goes up in airplanes, I mean, it's

gone up considerably from the 747 to the

787 and what's coming beyond, it it was

like to to run 270 volt DC cables was

untenable. So, we came up with actually,

I think I think the guys who did the

Dreamliner came up with this. Let's run

plus 270 and minus 270 with a common

conductor in between and now you've got

the best of both worlds. And and so

you're running 540 or or or whatever,

but not really from an ionization

standpoint. And I'm wondering if the

same thing drove the plus minus 400

vision, but from a different physics,

the physics of arc flash. Was it that

that arc flash is better understood at 4

500 volts DC and and there's a bigger

perceived risk at 800? Um that that may

have been where it came from. And and so

so so do on the the competing

architecture, which is a close cousin,

is 800 volts without the third

conductor. So if you have a balanced

load, the third conductor may be very

very small, but then you get into faults

and and how do faults propagate, you get

into the grounding schemes, and it

becomes a a nightmare for non-isolated

converters. And and and I think that's

where uh it would be nice to get some

harmonization. We frankly don't care

which way it goes cuz we're a common

every one of our ports is galvanically

isolated and when it is, you can float

it anywhere you want. You can float it

at minus 800, you can float it at 800,

you can ground the center point and and

get plus minus 400, and and we can use

any grounding scheme that Nvidia's

proposing in its in its general

reference architectures.

But I think it's it's going to come down

to a conductor cost in which 800 might

be cheaper and it might come down to a

the the the the opposite of that, like

how do you solve for arc flash? And then

again, I think a detection of arc flash

and being able to quench the source from

feeding the fault, I think that's where

the magical answer will lie in setting

the unified architecture, hopefully.

>> The

the cost consideration that you

mentioned, is it just because you know,

V,

do you have one conductor less to to you

know protect and to to control? Is it

just because of that or is there any

other consideration when we think about

the cost of different systems?

>> I I'm I'm sure there are many other

considerations, but I think that that

copper cable, the third copper cable for

for for it is a significant

consideration. Um um and and there may

be many others. What I would do is is um

we we can come back to you with a more

comprehensive look of of what feeds

that, but but generally I think it's

that copper it's that copper conductor.

>> Yeah, so um cuz we mentioned cost and

complexity. Is this in complexity? Um

when we think about, you know, Nvidia

and Nvidia's partners working on

initially this sidecar that's going to

be uh single-ended 800 V,

what are the considerations when it

comes to the complexity systems? Is it

actually more difficult to implement and

design a system that's using

single-ended 800 V

compared to uh one that other

um

uh agents may may might be working on

that use plus minus 400 V? V?

>> I think that question, the the essential

question when you have plus minus 400 is

are the loads going to be balanced at at

400 and minus 400? If the load is not

balanced, it's clearly a more complex

system. And for example, you know, some

fuel cells come at close to plus minus

400. And and so that's going to always

going to be the question can we just

take power differentially. But if you

can't power differentially, you have to

treat it as two different circuits so so

that imbalance doesn't persist and it

usually causes can cause in certain

circuits a runaway condition where you

collapse one voltage versus the other.

So I I think there are reasons to favor

a unipolar 800 V

as long as you can answer the arc flash

risk reduction properly. It also gives

you a way to where you can do

standardized grounding on the return

conductor with a multiplicity of ways

rather than worry about are you going to

do grounding on on three conductors

versus versus just a return?

>> Makes sense. Um okay, so I'm I'm going

to take this

a little bit higher higher level and

talk about that like adoption curve,

let's say. So four phases, right? Um

whitespace retrofit native compute

facility wide DC and then the end state

of housing these SSDs.

Uh maybe just to start the discussion

here, let me share

uh a specific chart that you guys put in

the 800-V DC article. Do you guys

believe that this is, you know, a pretty

solid uh adoption curve that's going to

happen? Is there Is there chances that

this gets accelerated or gets pushed if

that theoretical 1-MW rack like doesn't

really come to fruition or or just the

road map just gets pushed out like

uh for those who are

uh just listening, we've got on on stage

a on on screen for the YouTube audience

a chart that shows 800-V DC adoption

going from

basically nothing in 2026

to

almost 80% of the market by 2030

um in terms of like the incremental

capacity that's being added to the data

center market every year

and um

pushing above 30 GW worth of uh

actual like adoption, which is just

unbelievable to think about.

But it it happened in phases where

initially it's going to be a side car

and then later it's going to be facility

level wide that is actually happening.

So what like what's your high-level take

when you see a chart like this?

>> High-level take is it's always very

difficult to project into the future.

And and and so while we we can't tell

you whether these numbers are right or

wrong, we don't have any um special

crystal ball. We We do agree that there

will be a market for sidecar that will

go down over time as the native

architecture for 800 V DC takes root in

AI data centers.

And And And so the question is, yeah,

how long will that sidecar last?

Especially when you have AC dominated

architectures. Um and And you're doing a

brownfield install, it's far easier to

do it with a sidecar. Or you're trying

to mitigate the risk of

of not having the DC migration happen

fast enough. You go with an AC data

center, then you need the sidecar if it

starts to happen. And so yeah, I think

we generally agree with the shape, but

very difficult to predict the numbers.

We don't have that crystal ball. And in

our case, we solved the problem both

with a sidecar that we're developing as

well and releasing through partners. But

we're also developing that multi-port

that can handle the problem without a

sidecar cuz you've got both DC and AC

coming out. So it's a different way of

solving it for the whole data center.

>> Yeah. Jordan, I think you you mentioned

a really important point, which is, you

know, the possibility that this curve

gets at least displaced into the right

for some time, let's say a year or, you

know, how many time. Um Not Not down,

just to the right.

>> It could It could be to the right. It It

could go up. It could go longer. It It

could go down faster. It could be any

one of those scenarios, but the shift to

the right may is very very possible.

You're right.

>> Yeah. It's possible in the sense that

when thinking of this adoption curve, I

mean, we need to think it of and this is

what how we started the conversation.

This is a hardware and physics driven

transition, which is driven by these,

you know, road maps of 600 kW racks.

Suddenly, soon we will have 1 MW racks.

If these systems that are extremely

complex to design and to adopt a

large-scale, you know, are delayed for a

year

or whatever like Nvidia road maps for we

know try get this place for a year. We

we we know that this happens and even

especially when thinking of this super

complex systems. Well, this adoption

curve will naturally just, you know,

follow the hardware. It's not, you know,

it's just like

>> driven by the facilities? So the concept

of a sidecar is like I'm going to

retrofit a facility that wasn't designed

from the ground up to accept it multi-c.

Something that doesn't have a sidecar,

you don't It's not like sidecar

design beneficial to do. It's just

really dependent on the site they're

going into. Is that fair to say?

>> Mhm.

>> Yeah. Um so an interesting parallel that

we saw uh earlier this year uh I guess

last year as well was with uh chillers

because cuz Nvidia was pitching, "Hey,

you can run your chillers at 45° C when

you're doing liquid cooling." And in

theory you can do it, but in practice

the share of folks running their

chillers at that temperature is

extremely low. And then the question is

why? You know, it's more efficient,

supposedly. It's more energy efficient.

You can even save on CAPEX if you do

this. The problem is that the buyers

themselves sort of don't really know

exactly what their mix is going to be.

And in fact, if you think about it,

they've actually been proven right

because, you know, you would think maybe

everything is GPUs and what you're

realizing, and you know, I think semi

also been probably the first to call it

out at the end of last year, CPUs are so

back. Right? So you're actually very

much CPU constrained now as well. And so

it actually makes sense if you have a

limited data center footprint that you

want your facilities to be able to

handle many different types of hardware.

So the probably the biggest risk to SSD

adoption here would be the uncertainty

on the hardware remains high.

Um the timeline is part of it. The

diversity of hardware is another one. In

a world that is very largely say Nvidia

and Nvidia's road map is 800 volts.

The decision is easier, but in a world

where you have, you know, many different

types of ASICs, some of them maybe don't

require 800 volts. Maybe CPUs are even

more of of a need, which we actually are

pretty bullish CPUs right now.

And storage and others,

it makes sense that, you know, you want

your hardware to and your data centers

to be able to handle multiple types of

hardware. And so

it can it also goes goes back to like,

you know, who is actually building the

data centers. And right now you have

like this very interesting moment where

the folks building the data centers for

a big portion of them aren't actually

the ones

the ones really using them. The big

users are basically OpenAI and

Anthropic.

And, you know, the folks building data

centers are Amazon and Microsoft who are

building

for OpenAI and Anthropic. And Amazon and

Microsoft, they both have this struggle,

which is that their business is very

diversified. They have a giant CPU cloud

business as well.

And so they they're they're like at the

core of this uncertainty with regards to

like what types of hardware am I going

to deploy.

A few years down the road that could

change.

If folks like, you know, OpenAI and

Anthropic self-to-self build start to

lease directly, they're going to have

different requirements. They're going to

be

probably much more AI optimized in some

of their designs. And, you know, our

institutional clients already know that

pretty well. We've talked We've talked

about this at length.

But so these the these sort of state of

the industry right now where you have

like different layers of third parties

that are not the actual end users. And

so you have this uncertainty to what

type of hardware is being deployed.

That's one of the risks to SSD adoption.

And knowing it's going to be, you know,

2028, 2029, 2030, 2031 for the very

large scale numbers.

>> So I think I think by the way, excellent

points. And and I agree with everything

you said. The only thing I'd like to add

in there

is I think multi-port SSD, even if I am

biased, solves that problem for you by

allowing you to put any load on DC and

any load on AC. So, it de-risks it for

you. However, having said that, can I

predict adoption curve of multi-port

SST? No, I can't. Because because the

hyperscalers are generally more

conservative and and they have a right

to be, right? They're building gazillion

dollar data centers and they they are

going to be a little bit more

risk-averse. But, the neo clouds and the

data center developers may be more

willing to take a risk to to make sure

their investment has a faster payback.

So, I think there's several ways to

solve that problem. Uh we have one way

that we think is very powerful. We we

also have the sidecar way and we agree

with you. That that that it's going to

be the CPUs, the GPUs, the TPUs, what

power they use, how much goes to colos,

how much goes to AC loads, how much goes

to DC loads, and and and and how much

behind the meter power do you need. So,

there there's quite a bit of flux. That

that is for sure. And and I think

certain classes of SSTs are going to be

at more risk of adoption versus other

ones.

>> Okay. And and I guess okay, one one

interesting question for you then is uh

you said the multi-port kind of solves

the issue. Uh but the complication here

is that obviously the electrical system

of a data center is very complex. Uh

things have to be decided ahead of time.

And so,

I just want to like why does multi-port

actually solve it? Because if you design

your data center for AC, if your whole

distribution, your switch gear and

whatnot is, you know, AC, then you're

going to do the sidecar regardless. And

if it's DC, then you're going to do it

DC base. So, multi-port

>> I guess that's a lot better.

>> So, a lot of folks that are looking at

it with us are doing a hybrid that they

want to do a certain amount on AC and a

certain amount on DC. And what we offer

them in that case is you can put full

load on DC or you can put full load on

AC. As long as the two loads are under

the full load rating of the machine, we

don't care. We you give you both. So, it

gives them the flexibility. We're also

saying, if you have DC today or AC today

and you want to convert it to DC, we

offer a very simple changeout for for

for our portion. You're not going to

change the copper. You're going to

change the protection, and you offer a

port switchout from AC to DC. That's

what makes it easier to do it. So,

either buy both both, and then you deal

with the distribution, especially with

the protection. Um right, the copper is

not going to change. You're going to get

much more out of your copper when you

switch from AC to DC. But, you change

the protection, possibly the connectors

and the whips and whatnot. And and so,

it leaves you with an easier path when

that transition happens.

>> And so, actually, that's I think a good

transition to Nico's next banger article

in the specials,

uh because, you know, modular data

centers is like one topic we're looking

at very closely. And I guess you could

imagine that if you if you're multi-port

and you can handle both easily, then

perhaps, you know, there's a world where

you could use modular DCs. And you know,

you have one module, whatever, 5 MW AC,

5 MW DC, 5 MW AC. And then, you you

know, you can do whatever you want,

right? Like, that could be an

interesting, I guess, future for you

guys and for me reference architectures.

Um yeah.

>> Yeah. You know what I like about about

uh

the way you guys think, right? And it

it's reflected in that 65-page article.

Uh and I think it's the most widely read

publication from what I know. A lot of

our folks that called us up and said,

"Have you read the SemiAnalysis piece?"

We're like, "Wow, you know, these guys

are really good." But, we like the way

that you systematically think about it

from the whole system perspective and

not just focus on one little doohickey.

So, my compliments to you in in looking

at all the things on the load side and

on the AI side that'll that'll cause

architectural and technology adoption

changes.

>> Always nice to hear self-promotion on

the SemiAnalysis podcast room. Thank you

for that. Uh we will uh

>> Well, in in this case

you know, because a third party was

doing it or your guest was doing it

without the offer of a free cappuccino.

I feel that it was genuine, right? So

>> Cappuccino coming your way next time,

man.

For sure.

So,

one thing that uh

it just coming from the Neo Cloud

perspective, one thing that the

hyperscalers always talk about is

fungibility. They treat this at the

fleet level, I think, where like

different data centers might be used for

different people or different things and

then they they try to solve this with

software. It seems like everything

you're saying right now is making the

case for fungibility at the power level

um in the data center itself. Can you

Can you talk about like

future-proofing even beyond 1 MW? I

mean,

like let's actually Sorry, before I ask

that question, let's take a step back

and go through the

rack-level power roadmap just for a

second um because I

I think maybe we we glossed over this a

little bit or or assumed that the

general audience is is going to

understand this. So, let let me put this

on screen just so that we we know about

this. When When I started doing

um you know, design work on like

uh compute systems for

uh GPU servers, it was in like 2016-2017

time frame and you're working on like

the V100, the Volta generation systems.

And so, like a rack, which is a standard

data center rack um that you might have

in like US-East-1 and air-cooled CPUs

for AWS is like 12 kW.

And then 2020 started COVID and we

started seeing more air-cooled, you

know, density go to 30, 40 kW per rack.

We're now shipping today somewhere

between 130 and 140 kW per rack with the

GB200 and GB300 systems.

Um

next year

or potentially at the end of this year

uh Vera Rubin and and so, what data

centers were designed for 2 years ago or

2 to 3 years ago is 360 kilowatts per

rack. And then the very ribbon by the

end of 2027 is 600 kilowatt per rack. So

for the audience like that's already a

massive we have to put this chart on a

log scale for those looking at it on

screen because

it's going up by 6X.

Without the transition to 800 volt DC

even considered, right? When we say one

one megawatt racks and what we're

considering for the 2030 or or

potentially 28, 29 time frames

is beyond

a

you know,

60 times

a multiple of of power per rack that

that has had to be contended with. But

now I'm going to ask the question which

is

uh

what's the future proofing look like

beyond this? Let's say you build a data

center

um

that's 100 megawatt scale

uh

which by the way I was in one of these

facilities a week ago. It's absolutely

unbelievable how much of the facility

itself goes towards power and cooling as

opposed to white space as opposed to

like chips and data hall space now. Like

well over 80% of the physical square

footage is just power and cooling now.

And so you know, I can't even imagine

what the future ones are going to look

like. But let's say it's a 100 megawatt

site or something like that.

Um

gigabyte site even.

Uh these sites are expected to to go for

15 years,

right? And the whole case for

fungibility on power I assume is like

you're not going to rip out systems that

you've deployed in the middle of their

life. It's just like we want to reuse

this this facility for future systems in

the future. So is there stuff beyond

um the current generation of systems if

you push this out 10, 15 years where you

think SSDs would be able to be more

capable of handling the future load at

the end of a 15-year life cycle for the

data center facility itself that was

built to handle those chips? Not only uh

yeah, I think I think so. I think I

think you have to look at a architecture

that's going to deliver far more

density. It's going to be able to

work with multiple sources behind the

meter cuz it's that you know, what when

you look at the transmission grid and

the distribution grid, even if you've

got enough generation and and then you

put a 100 megawatt data center at one

spot, you choke up all the lines around

it. That's why there's all this issue

with well, how am I going to improve my

grid to get there? So, the answer in the

in the short run is well, I've got to do

the behind the meter power until the

grid upgrades. But, if the grid upgrades

and the cost of grid goes up or or the

cost of of depreciating that asset, that

cost gets passed down in more expensive

dollars per kilowatt hour. That means

your token cost is going to go up. So,

so how do you leverage today's behind

the meter power that you've put in and

depreciated, can you still continue to

use it and leverage it and yet increase

the density of delivery towards the

racks that might go higher in power. And

and and I think I think that maybe that

maybe one one thing to look at. The

second thing to look at is it's sort of

like this movie I saw a while ago where

there's this gigantic 80-ft robot and

and when it comes to a stop, the top

opens up and a little kitty cat who's

running the whole robot jumps out.

That's how it is. It's sort of like you

got this massive power architecture and

the brain, which is the GPU stack, keeps

shrinking and shrinking and shrinking.

So, what geometry of the data center is

going to optimize that that that brain

shrinking? Is it going to be like a

bicycle wheel where you've got power

coming in from from multiple places and

boom, you pop down

you know, increasingly smaller set of

GPUs that allow you to to handle that.

And then and then what about

superconducting? At what point

Uh does superconducting kick in where

you can where you can do 5-6 MW on on a

strand of um cryogenically uh cooled uh

cables that'll bring you unprecedented

density. And and then how do you

distribute it to where any failure mode

will not give you any stranded power.

And you can route the power to wherever

the GPUs uh demand it for the cheapest

token generation. Or another way to look

at it, where you might even have an

auctioning system for selling the the

the the token generation to the highest

bidder.

And and so I think it's going to be a

tremendous amount of software-defined

uh GPU scheduling, a tremendous amount

of software-defined power routing, and

and uh power handling at every single

level. There will be a cooling fabric,

there will be a power fabric that can

adapt to all these situations. And then

there will be a a GPU job scheduling and

and and whatnot as you look at different

phases of of GPUs roll out. Now, it's

also very conceivable, right? It's easy

to brainstorm because because you you

you're just thinking out the the reality

making it real is different. But what

about all these optical interfaces and

all this optical computing that's coming

out? Is that going to reverse the power

density or will it keep power density

and make it flat at some point where the

optos kick in and reduce the amount of

uh of uh uh of power that you need for

the same amount of computation? Those

are the questions and I I I'm smart

enough to know that I'm not that smart

and I don't have those answers on when

it's going to happen or or how, but

these are some things to think through.

>> I think we're um we're big believers in

Jevons paradox for everything including

power. So even if you've got that

optical stuff, I think we're going to

still keep consuming quite a bit of

power into the future.

It's interesting to hear you say that

specifically for behind the meter power

generation, Do think this is a trend

that's going to continue? In other

words, just building more

facilities at the same site, even if you

get grid-connected, or just trying to

deploy more chips in the same site?

Um do you think that's

if if people are planning for 800 V

right now, or planning big data centers

you're working with them right now, is

this behind-the-meter trend more here to

stay than we think?

>> I think Gordon, that's an excellent

question. I think I think So, I've done

a lot of work on the distribution grid.

I've done a lot of work on the

transmission grid. And and and I've

studied the economic models of

utilities, right? And and the world all

over the world, utilities are generally

they have they have unipolar or

unidirectional flow of power, where

power goes from generators down the

transmission and distribution networks

to where it's used. And and upgrading

that infrastructure is a

multi-multi-year time frame.

And and and so you've got you need

hundreds of millions of dollars to do

it. So, so I'm going I'm going and and

now you've got this this cellular power

concept that we call, where you can add

10 20 MW blocks at a time behind the

meter, and and start to add a gigawatt

of distributed power. So, which one is

going to win out? And I think the speed

to power or the speed to compute will

win out. And for that reason, uh

distributed power generation and

behind-the-meter power generation, which

is another another word for it, is is is

going to take root. And I don't think

it's going to take root in just uh AI

data centers. I think it's it's going to

take root wherever you've got to develop

uh electrical power delivery without the

cost of a billion-dollar nuclear plant

or a ten-billion-dollar nuclear plant.

It's far easier to put a

five-million-dollar pod and give

villagers a a hospital, give them a

school, give them a chance to educate

their kids. That's that's right? So,

there's an electrification trend that's

going to drive the the need to cellular

power behind the meter power, but

there's the massive market right now at

hand that's going to drive the volume to

get make make all the infrastructure for

behind the meter power more palatable

and drive the levelized cost of energy

down. And then you adapt it to different

areas. So I think it's a disruption of a

multi-trillion dollar energy market or

or maybe not disruption, maybe that's

too bold. Maybe it's the augmentation of

a centralized generation model of

utilities with distributed generation

augmenting it because it's far more far

easy to deploy, far easy to redeploy,

and and and far more incremental

investment with far faster payback.

>> Yeah, that's that's really inspiring

honestly to hear that framed that way

where innovations that people are doing

to serve the demand from coding

assistant tokens right now

is potentially I think highly likely to

have a lot of positive downstream

effects in all sorts of other industries

that all just need a lot of power in the

future.

>> That's right, Jordan. And and and think

let's think about it, right? All of us

on this call grew up with energy. I

don't think we ever worried when we

flipped a light switch on, right? We we

had light to do our homework in. We had

light for our we had power for our

computers. We had access to the world's

resources with the internet. We could

charge our cell phones. But let's think

about the world that didn't have power,

right? Or or that part of the world that

doesn't have power. They live a life of

poverty. And the same thing is going to

happen with AI. Those that can use AI

and become really adept at it will

create a further divide. So so I think I

think certainly for today for DG Matrix

shareholders, I got to focus on AI data

centers, but there's a part of me that

also is looking out at the

electrification world and and that part

that says you want to leave the world in

a better place, you got to think of the

rest of humanity and how you can help

them in some way. So, yeah, I hope the

AI data center not only drives us to

superhuman intelligence, but makes power

cheaper for everybody around the world,

fusion or no fusion.

>> You're offering up a lot of options for

where we can take this for the last few

minutes of the podcast here.

>> [laughter]

>> Nico, Germany company over there.

>> The double espresso kick me in the

>> Yeah, just one thing I'm curious cuz you

you mentioned

initially that one of the reasons for

800 volts is because we reuse existing

supply chains, for example, from

automotive. I'm just curious like for

for you, for your supply chain, like do

you actually use automotive suppliers

and all the automotive vendors, auto

parts, or is it just something

completely different?

>> No, we use semiconductors that the

silicon carbide that was developed for

1200 volt architecture. Could some of

those be used in in EVs? Yeah, some of

those are used in EVs. And

do they give us a benefit? Yeah, I think

they do. They do. You have to look at um

when when you are running these

surges, you've got to look at the

physics of the semiconductor failure.

And and then you got to translate that

to people who drive EVs who have a lead

foot. There's a lot of commonality

between all those surges. And and so the

people who have designed the physics to

accommodate that, there's some magic

there.

>> Silicon carbide or gallium nitride for

power electronics?

>> Doesn't matter. I I think it Right now,

silicon carbide is more apt to give you

hundreds of kilowatts to to to

megawatts. Gallium nitride is coming up.

It's more suited for hundreds of watts

to kilowatts. And and and quite frankly,

as as I was discussing today in an

investor panel, it shouldn't matter to

those of us that want to deliver

economic value to customers. The

question is which one does a better job?

We're agnostic. We are actually we've

been experimenting with both for 10 plus

years. And it's just silicon carbide is

more mature at the right power levels

right now.

>> All right, another one. How how big can

your SST get? Could we see a 10 MW unit

a few years down the road?

>> Yeah, actually the medium voltage SST

that we're working on, which is 35 KVN

and and let's say 800 or 1500 V

programmable out, that is designed for

10 MW in in one container. It's going to

be one large container, but it's

designed with higher voltage

semiconductors and on the on the front

end and and and a divide down and and

then and then a lower one. And I think

that's slated for '28, but '27 we're

looking at the 6 MW SST. And today we

have, of course, 400 kW that we can

parallel to a multi-MW.

>> Where are customers

expecting to place that? Is it in going

to be in the gray space? Is it going to

be outdoors? Is it going to be

>> not going to be in the white space. And

and you know what's interesting is in

2011 I worked on a product that was

bringing medium voltage to the top of a

rack.

Um I can't talk much more about it, but

that was the first SST, one of the first

SSTs that we did. And and and really if

you want to reduce the the cable to to

copper or or get the most, you got to

bring medium voltage, but there's a lot

of safety issues and architectural

zoning issues and and whatnot in at a

national level, so it makes it tough.

Maybe China would be the one to to get

that done first. But but I think I think

raising voltages and bringing bringing

power and and and compute together

physical proximity is is a one one trend

that's taking root now.

>> Speaking of China, is there any any kind

of issues for you guys to source silicon

carbide from China?

>> We're not sourcing any silicon carbide

from China. We're just sourcing it from

the best folks we can we can find and

and and so our sources are are United

States

potentially Japan, but it's United

States and Europe right now. And and in

yeah, Europe Europe has two two very big

suppliers for us. And and then America

right right there North Carolina has a

very big supplier for us, too. And

that's what we're focusing on. We are

sourcing some, you know, non-CPU,

non-software electromechanical stuff

from China, but we have a China plus one

sourcing

strategy. And so we can get the same

parts from say Mexico or Vietnam and

whatnot. And we like everybody. We're

just trying to mitigate future risks.

All right?

And I think one one thing I would just

mention in in in going away is we think

of all these architectures, let's not

forget that the more software driven

your power becomes, the the better your

cybersecurity must become. Cuz you don't

want third parties to hack into it. So

we've developed and and deployed in the

past cybersecurity proof power solutions

on the transmission grid. And and and

that's a skill set that I think has to

expand in the industry. And if it

doesn't, uh you have the risk of uh of

um you know, miscreants coming in and

taking your data center down. So so

let's make sure that that's at some

point we cover this, too. Is how do you

how do you really make uh

uh make make this cybersecurity proof,

including background checks on every

single entity that touches the

electronics and develops the software.

>> Yeah, we don't want Stuxnet for any of

these new big uh

data centers. Seems pretty important.

That's right. Yeah. Awesome.

>> Okay.

>> Well, guys, thank you so much. This was

a

>> whirlwind tour of 800-V DC, SSTs,

all the implications on the supply

chain. Appreciate you spending the time

with us, Arun. Yeah, thanks to everybody

who took the time to listen today.

>> My compliments to SemiAnalysis again.

Thank you very much for the opportunity.

>> All right, take care, guys.

>> Bye-bye.