Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
1:01
So we are here and I want
1:03
to welcome Bill Bubenicek from
1:06
Generate Capital here to
1:08
our studios , into your
1:10
headphones or into your cars or
1:12
wherever you're listening to this , and
1:16
before we get into Bill's origin
1:18
story , so just to kind of set the stage
1:20
here . So Bill and I know
1:22
each other . We've done some work in the past . Stage here
1:25
. So Bill and I know each other . We've done
1:27
some work in the past . He is a resident , as I am
1:29
, in Cape Cod , assachusetts , although
1:31
he lives there full time
1:34
, so we can talk about that , but
1:41
Bill's had an interesting background in power and we were . You know , before we hit the record
1:43
button . You know , if you look at Brad Smith's talk at MWC
1:46
on AI economy , which I thought was a really
1:48
good one to go look up , he talked about
1:50
the kind of the foundations
1:52
of the AI economy and there's kind of two pieces
1:55
to the foundation One is connectivity and
1:57
the other is power , and power as
1:59
in electricity . And so
2:02
quite often when you get in discussions around edge
2:04
computing , the
2:06
issue of power , power cost , I mean , certainly
2:08
there's power discussions . When you get into low power
2:10
systems that are , you know , battery
2:12
powered , but even you know
2:14
, data centers , edge data centers , things like
2:16
that , there's just this , you
2:19
know , kind of an existential I wouldn't say crisis
2:21
, but you know , an existential challenge
2:23
about power . And so , bill
2:25
, I want to welcome you to the show to talk about
2:27
this stuff . Why don't you give folks
2:30
kind of your intro , your origin story
2:32
, let people know who you are ?
2:33
Sure , thanks for having me Pete . Yeah
2:36
, so I started out really in GSM
2:39
network buildouts , which in other words
2:41
cell phone , which in other words cell
2:43
phone infrastructure , primarily
2:46
in emerging markets kind of where I started
2:48
. So I spent a lot of time in
2:50
the Middle East , africa , parts of Latin America
2:53
, and while doing those
2:55
build outs and selling the equipment there I
3:05
came across the cell towers that were in most of those markets running on diesel generators
3:07
24-7 . And it was around 2006 , 2007
3:09
, and we called it clean tech then . But
3:12
the clean tech thing was kind of happening
3:15
and I got into
3:17
that and I was pretty excited
3:19
about a more sustainable energy
3:22
future and I thought , well , here's a good
3:24
place to start already , already in the cell
3:26
tower business , and they're all running on diesel generators
3:28
24 seven . So what can we do
3:31
? So we implemented
3:33
some of the first we called hybrid systems
3:35
, which is just battery cycling with generator
3:37
. Eventually that turned into battery plus solar
3:39
plus generator and
3:42
I launched a company called clean power
3:44
Systems in 2010 , co-founded
3:47
that and brought those solutions
3:50
into the African continent . Mostly
3:52
Uganda , Kenya and
3:55
Ghana were our primary markets . And
3:58
then , sort of fast forward , I got
4:00
a crash course in project finance when
4:02
I realized , hey , there might be a better way , as
4:04
opposed to selling these things on a
4:06
CapEx basis , right , can we sell it as a service
4:08
? And so you know , I
4:11
don't have a proper finance
4:13
or analyst background or anything like that , so it was
4:15
really learning through osmosis and we
4:17
launched our next company then that was focused on
4:19
converting our solutions
4:21
into a as a service model or
4:23
like an SQL model , and so
4:25
we did that in East Africa
4:27
and in India and
4:30
that was a lot of fun , a lot of good
4:32
experience , but mostly an off grid environment
4:35
, right . So my , my power experience
4:38
started started kind of , you know , in the off
4:40
grid sector or the call it
4:42
grid sector , where it was
4:44
like really intermittent power
4:46
. So
4:49
, anyway , got a lot of good experience
4:51
there . In fact , that's where I first met Jigar
4:53
Shah , who's a co-founder of Generate . We
4:56
tried to do a few deals in Uganda at the time
4:58
and kind of kicked around a lot of ideas
5:01
and he had a lot of kind of good advice for me
5:03
as an entrepreneur trying to make
5:05
it in this space . And when
5:08
I finally came back to the States and was
5:10
kind of looking at a change , say like , how do
5:12
I make a much bigger , broader
5:15
impact , right , like this was a great business and
5:17
we a lot of lessons learned . But it's
5:19
really hard operating in these
5:21
other markets , operating
5:28
in these other markets . And I came back and I was in Boston at the time and reached out to Jigar
5:30
, amongst others , and he called me back I think
5:32
within 30 seconds of the email and
5:35
said hey
5:37
, what's your thesis for
5:39
all things related to digital and kind of how
5:41
that intersects with energy ? And
5:44
I said , well , I got to work on that thesis
5:46
, right , it's been a while . I've got the off-grid side
5:48
, but let me have a look . And he
5:51
kind of pitched me on Generate and
5:53
what they were doing and really found
5:56
that interesting and exciting
5:58
. And so I kind of started
6:00
looking around like what's the thesis here ? What's
6:03
the play more in the US , or it could be Europe as well ? What's the thesis
6:05
here ? Like , what's the play more in the US , or it could be Europe as well , but in
6:07
the traditional grid markets . And
6:10
I ended up spending some
6:12
time I had a lot of time with the
6:14
IFC and World Bank when I was in my previous
6:16
life and so I called up my friends
6:19
and colleagues everywhere . I was trying to kind of looking around
6:21
like where's the right thesis ? And credit
6:24
to my old friend , eric Crabtree at
6:26
the IFC , who said you've got to look at data
6:28
centers , right . Everything is moving . You're in
6:30
the cell tower . We call that ICT sector
6:33
. I think TMT was another term they used
6:35
in the World Bank . But anyway , he says look
6:38
towards data centers , right . And so
6:40
I already kind of brushed into the data
6:42
center sector . I had experience with
6:44
fiber just because I was already in the
6:46
telco sector
6:49
and all those things converge so
6:51
I started kind of calling it digital infrastructure
6:54
. That term started happening around . That point
6:56
, after
7:05
a bunch of work and really looking into data centers , was there's going to be a convergence
7:08
or an inflection point , let's say , where the grid is simply unable
7:10
to keep up with the power demands of
7:12
the digital infrastructure sector . Right
7:14
, at the end of the day , that was the thesis
7:16
at its core that I brought to Generate
7:18
in 2018 , 2019 . And
7:22
they said let's give you
7:24
a chance , right , welcome
7:28
aboard . And I came on as a consultant under what they called a developer in residence
7:30
program at the time and I think that's
7:32
just maybe
7:35
a position that Jigar had invented . That said hey , you're
7:37
an entrepreneur , developer , and we don't really
7:39
know what label to put on you because you're not
7:41
a finance person , but can you come
7:43
in and kind of try to put deals together
7:45
, etc . And it really ended
7:47
up being , I guess , what you call an origination role
7:49
. I later learned in finance that's the
7:51
term right , go out and originate deals
7:53
and bring them through um . So
7:56
again , the thesis was fundamentally
7:59
on power , right , and the
8:01
lack of power , let's say , the lack of supply
8:04
or the ability of the grid to keep up with the
8:07
demand from the digital side . And
8:11
I guess it was last year 2022
8:14
, 2023 , when the AI boom
8:16
took off , right . And then suddenly the
8:18
thesis was like right now Before
8:21
that it was kind of in the future let's prepare , let's
8:23
think about what we can do to solve this . But all
8:26
of a sudden the explosion happened and here
8:28
we are now with , I think , everyone talking
8:30
about this .
8:32
Yeah .
8:32
Power is the problem .
8:35
Yeah , I mean , if you look at the latest , like the NVIDIA
8:37
GTC announcements with Blackwell
8:39
, you know , I mean each one of those chips is , you know
8:42
, peaks at about 1200 Watts , and
8:45
so you start to get a building full of those things
8:47
you know and you know . Just
8:53
as a comparison , I mean you have um , think about like packet core switching
8:56
and things like that in sort of the telco world you know , running on Intel , you
8:58
know Xeons and stuff , and it's all good , but it's not
9:00
, uh , it's not pushing the power envelope
9:02
as much as you would if you're
9:04
stuffing that field with , you know , nvidia
9:07
H200s or Blackwells . I mean that's a whole
9:09
other order of magnitude , frankly . Oh
9:11
yeah , more power , right , and
9:13
you know , people who read the newspaper know
9:16
that . You know just , public grids in general
9:19
are struggling with keeping up with
9:21
power requirements
9:23
right in Texas and other places and rolling
9:26
brownouts . So you start adding in
9:28
these data centers filled
9:30
with AI chips and then all of a sudden
9:32
there's a gap . Maybe not all of a sudden
9:34
, but I think it's kind of an obvious , you know
9:36
gap that people are trying to fill . So that's , I
9:39
think you know . Like you said , it
9:41
started out as it's
9:44
a good thesis and now it's really becoming a
9:46
reality , because the demand for AI workloads
9:48
is so high and
9:51
folks are struggling . So how are hyperscalers
9:53
and other folks figuring this out
9:55
? I mean , how are they trying to close the gap between
9:58
, you know , ai factories
10:00
and electricity
10:02
?
10:04
Yeah , so , yeah , so
10:06
, a few different ways , and
10:09
this was kind of core to the thesis as well
10:11
. It's like , all right , there's a thesis , great , but what's
10:13
going to happen , right , as a result of this Play this
10:15
forward ? And
10:17
the theory was well , if
10:20
power is so important as we know , it is to
10:22
the growth of digital , and we're really talking
10:24
about , when you say hyperscalers , we're talking about the biggest
10:27
of the big companies in the world , the biggest
10:29
balance sheets , etc . Are
10:31
they really going to let the grid's inability
10:34
to keep up stop them from deploying
10:36
? Probably not . I don't think
10:38
so . Let my bet with you know , I
11:01
know so . So the way I saw it was well , they're going to figure out . Either they're going
11:03
to start building their own power plants or they're going to figure out how to maybe
11:05
co-locate next to existing power power plants , to unlock capacity that might otherwise not
11:07
be available due to , you know , transmission , distribution issues or what have you
11:09
Right ? And so I think that those two themes kind of come out of this . It's going to be one of
11:11
one of those two , and
11:13
if you really want to fast forward long term , I think there's a really good chance that the digital
11:16
infrastructure folks are going to become energy companies themselves out
11:19
of necessity , right , there's like a forcing function
11:21
here and again we're going to bet on somebody
11:23
. Is
11:27
it going to be that highly innovative , like largest companies in the world , that's
11:29
going to do that , or is it going to be the kind of utilities that are going to
11:31
figure out how to suddenly get super innovative
11:33
and keep up ? Both are going
11:36
to happen in a way , but I think my money is on
11:38
the big guys here .
11:39
Well , so like , instead of Microsoft having an
11:41
Xbox subsidiary that's doing gaming
11:43
, maybe they'll have a power subsidiary . Then
11:46
you buy your power . It's part of your
11:48
M365 subscription , you know .
11:51
I think something like that is already
11:54
in the works . I mean , look at the press AWS
11:57
just announced recently the deal
11:59
with the nuclear facility in Pennsylvania
12:01
, for example , right , right . So
12:04
we're kind of already seeing things moving . There's
12:07
plenty of different data centers that
12:09
have adopted on-site power , like
12:11
Bloom Energy , fuel cell
12:13
technology , right
12:16
. So I think we're seeing both
12:18
of those play out now in real time
12:20
. It's starting to actually happen and
12:23
I think we're going to see more and more of that , right . Yeah
12:25
it's really interesting .
12:27
It's almost like when Apple , you know , years ago Apple
12:31
got big investments in like mining companies
12:33
to get the rare metals that they needed to
12:35
build iPhones and iPads
12:37
right . They just kind of went straight in
12:40
the supply chain and , you
12:42
know , you're seeing maybe something similar like that
12:44
with power and some of these hyperscalers .
12:46
Yeah , I would . And , by the way , are you talking supply
12:48
chain ? I would love to see them get into
12:51
the transformer supply chain right
12:53
, which is is another
12:55
major problem in bottleneck .
12:57
Yeah , that's true , that's the . Those
12:59
are the um , the boxes
13:01
that you see on the poles right that go from
13:03
the basically transmit .
13:07
Yeah , exactly , it makes the voltage usable for
13:09
the user .
13:10
Yeah , yeah , and there's a whole supply chain
13:12
issue around transformers and that's one of
13:14
the big shortages . I know Elon Musk has talked about
13:16
that too . Yeah . It's
13:19
one thing to generate the power , then you've got to move the
13:21
power through the grid . One thing to
13:23
generate the power , then you got to move the power
13:25
through the grid . And you know maybe
13:29
people don't realize this a lot of the infrastructure that you see , especially in the US , is 50 or
13:31
60 years old , right these transformers and these power lines , and there's huge
13:33
inefficiency . So you may be dropping , you
13:35
know , a third or more of the power
13:38
by the time it goes from the generator
13:40
, whatever it is to , to
13:42
your plug . You know you could be losing a huge
13:45
percentage of power just through those
13:47
transmission lines too , which is not good . In
13:50
a world where there isn't enough power , the last thing you
13:52
want to do is kind of squander
13:54
it on kind of these inefficient old
13:57
transmission lines , right . So that's
13:59
another challenge .
14:01
It's another one . There's no doubt about it , right , but
14:03
again right . Who better to step
14:05
in , perhaps , and make a difference
14:07
here ? You know , when
14:09
it comes to the players that could do this
14:11
, again , we're talking about the biggest
14:13
balance sheets in the world .
14:15
Yeah , yeah , no , that's really interesting , yeah
14:18
, and so like and you talked a little
14:20
about nuclear , I mean . So typically
14:22
it's interesting . The other paradox
14:24
I guess with power is there's not enough power to
14:26
power kind of AI factories . So that's the
14:29
working hypothesis . I think it's probably pretty
14:31
well established at this point . And
14:33
at the same time there's also a move to
14:35
move to more greener and renewable
14:38
energy right . So wind and
14:40
solar and things like that . So
14:44
you have , these things are almost , I wouldn't say competing , but they are sort
14:46
of both pulling in slightly different directions , right
14:48
, yeah .
14:51
Yeah , I think again
14:54
you're going with the AWS , is the
14:57
Microsoft's , the Google's videos ? These guys
14:59
are all making really big commitments
15:01
to clean energy , right
15:04
, and I mean , yeah , across the board . And you know , making really big commitments to clean energy , is that right ? Yeah , that's the
15:06
board and , regardless of that acronym
15:08
, like they're making commitments
15:10
to a more sustainable future
15:12
and they have the balance
15:15
sheet to actually do it , to
15:17
sort of push innovation forward and also
15:20
be the anchor customer , let's say , on
15:22
cleaner , more sustainable
15:25
builds for power
15:27
plants , right . So I think we're
15:30
already aligned in that sense Right , in
15:32
that they want the cleanest power . However
15:35
, the lack of power altogether
15:37
right now kind of puts them in a conundrum too . Right
15:40
? Hey , if there's power available already here at
15:42
this gas plant , are they going to take it ? Do
15:44
right , hey , if there's power available already here at this gas
15:46
plant , are they going to take it ? Or are they going to not spin up
15:48
their latest data center and sort of wait for a source ? I don't know how
15:50
that's going to play out . What ? I can say , is
15:53
that where we sit ? Certainly , and
15:55
what I represent on the Generate side is we are representing
15:57
for the cleanest sources that
16:00
are financeable and sustainable
16:02
today , the degree
16:04
to which we can influence the market
16:06
towards more sustainable . We're
16:08
going to do that , but I think you already
16:11
largely have a pretty big commitment
16:13
from everybody there to
16:15
do that anyway . So I think , you're kind
16:17
of already well aligned in that sense .
16:19
Yeah , well , yeah , and
16:21
the demand for power , and then that turning into
16:24
demand for investments in more green
16:26
power solutions is beneficial for everybody
16:28
. So , and then the other trend that I've
16:30
seen too is that you
16:32
know that used to be where you'd plop a data center
16:34
area of compute out in
16:37
, you know , near a kind of a higher
16:39
density population area or whatever , to
16:41
service that area , density
16:45
population area or whatever to service that area , and then the power would be somewhere way out somewhere
16:47
else . But now we're seeing you mentioned like , so almost like the co-location
16:49
of compute and power
16:51
generation , and so that's
16:54
a different kind of you . Look at a map of any country
16:56
where their power stations are , where
16:59
the power is being generated is kind of probably where the compute
17:01
is going to be happening , nearby
17:03
, right , I mean to mitigate some of these transmission
17:06
issues that we were talking about .
17:07
Yeah , again , crystal ball would say yes , I
17:09
think that's where it's trending right . I think that in
17:11
also with AI I'm sure you know and you've
17:14
talked about right like the latency is less
17:16
of an issue than what
17:18
we've traditionally thought of when it comes to the cloud et
17:20
cetera . And so you
17:26
know , I think , locationally you still probably want to be within some reasonable driving
17:28
radius of a metro area because
17:30
you need the human resources , the
17:32
talent , to be able to get there
17:34
and reasonably be able to either commute
17:37
or live nearby . But
17:39
yeah , it does seem that the trend would be
17:41
the compute is going to move
17:43
to where the power is more
17:46
so than perhaps the power moving to the
17:48
compute .
17:50
Right , right , yeah , although some
17:52
days , at some point in the near future maybe
17:54
not in the near future you'll have these kinds of fuel
17:56
cells attached to micro
17:58
data centers , like the edge data centers , right when
18:00
it's like a little hydrogen fuel cell stuck
18:03
on the side or on the top
18:05
of a box , you know , and then then you'll
18:07
be able to sort of that's like a micro grid architecture
18:10
right when you would uh , you
18:12
know more , more , you know power
18:14
is more distributed to where the compute is
18:16
. I mean then that , but that's probably like
18:18
five to 10 years down the road , I would say .
18:21
Yeah Well , I mean , yes , I want , want to . I would
18:23
think so too , but when you look at the compounded
18:26
anal growth estimates for edge
18:28
, it's astronomically high
18:30
, right ? So you say , like , if
18:32
that's true , then we're going to have to figure
18:34
this out pretty quickly , right ?
18:36
to kind of yeah , right
18:38
because , yeah , because there's now there's like , there's core
18:40
, there's regional , there's edge , you know , there's
18:42
, you know , and then there's obviously the , the
18:45
, the , the devices too that are running
18:47
AI , and so how to get
18:49
power to where one of the benefits
18:52
of kind of thinking around edge computing
18:54
is around doing
18:57
the compute , where the data is generated , so
19:00
in the vehicle , on the robot
19:02
, in the factory floor , in the retail shop
19:04
, at the Taco Bell drive-thru
19:06
itself , and
19:08
so , as that trend continues to
19:10
happen , like you said , the growth trend is huge there
19:13
, like power needs to be
19:15
there , there needs to be power there , and
19:17
so I do think we're going to , in
19:19
addition to seeing the big
19:21
hyperscaler data centers with all the blackwells in
19:23
them , will you know , start springing up
19:25
in co-location with power plants
19:28
. I think we're also going to see a lot of innovation in
19:30
bringing microgrid power to
19:32
the edge . You know , to support
19:34
some of these more .
19:35
You know local , hyper local , you
19:38
know power needs right at the same time I
19:40
agree right , and and I think it's
19:42
also important to note that those microgrids need to
19:44
be able to scale up right and it's
19:46
modular in as simple , simple way as possible
19:49
to meet the growing power demand because
19:51
, again , based on what's being projected
19:53
, we've ever recently believed that
19:56
, yes , you're going to need that initial power and then it's going to
19:58
continue to grow , right
20:00
. So , just to take
20:02
a step back on this topic , what
20:04
I do think is interesting and where
20:07
my optimism comes in , as I
20:09
kind of look out five , 10 years to
20:11
your point on how this happens , we've
20:14
always talked about I say we
20:16
, us folks in the energy , clean tech sector
20:18
, et cetera . We've always kind of theorized about Power folks
20:20
. Yeah , these power folks like us , right
20:22
. We theorize about this grid of the future that
20:24
is going to be highly distributed , right
20:27
, and it's going to be highly it's
20:29
called bidirectional , in that a
20:31
power plant can both give
20:33
back power as much as it can consume
20:35
power , et cetera . And I think data
20:37
centers actually have the architecture
20:40
and have the ability to be these
20:42
distributed energy resources or like that
20:44
, right .
20:45
And when you start wrapping in what we call
20:47
virtual power plants or the software to put
20:49
everything together and get these really
20:52
interesting balances amongst all of these distributed
20:54
resources , you could start seeing
20:56
a case where the digital infrastructure sector
20:58
could actually underpin the
21:00
build-out of this distributed
21:03
grid of the future , which I think is super exciting
21:05
from that perspective right , right , the VPPs
21:08
, as they call them and in fact
21:10
I mean before we hit the record button we were talking
21:12
about you have this
21:14
kind of whole solar build-out at the Cape
21:16
, at your house , and
21:19
I'm currently speccing out adding solar there
21:21
too , because , for
21:23
those that don't know , the New England energy
21:25
market is quite expensive , and so
21:27
you know , in as much as you can generate your own
21:29
power there , that's a good thing for the long term . And
21:32
so then there's this thing called net metering , where , as
21:35
you're generating power and folks that have solar
21:37
know this , you know in times
21:39
when you're generating more power than you're using on your house
21:41
, you're able to sell that power back to the utility
21:44
right at whatever the kilowatt rate
21:46
hour is usually whatever
21:48
, and so that is kind of essentially
21:50
like what Bill's talking about . Right , this is
21:53
kind of like the microgrid idea
21:55
is that you know , power
21:57
generation is , could become , will
21:59
become , hyperlocal to a certain extent
22:01
through these renewable
22:04
resources , and that power
22:06
can then be fed back in to
22:08
a virtual power plant or some sort of system
22:10
that then can then feed that power out
22:13
to those that need more power than they're
22:15
generating , if that makes sense , right ? So , and
22:17
that could be the Taco Bell drive-thru that's running a bunch
22:19
of NVIDIA processors to look at like you
22:22
know how many tacos are in the bag or whatever
22:24
the AI thing is that they need to do
22:26
, and so that's a really interesting
22:29
development because it's sort of mixing
22:31
what some people are familiar with today , which is
22:33
creating their own power at their house using
22:35
solar panels or whatever , and
22:38
then you know the software
22:40
and the systems to sort of manage . All
22:42
that is very , you know , very
22:44
much in line with a lot of the uh technology
22:47
that's being developed today for data centers yeah
22:49
, I'd take it one step further .
22:50
Like the solar is . Just think of it like a little
22:53
power plant that's selling back into the grid
22:55
or offsetting . That's okay , that's like
22:57
. That's pretty simple . When you get into the
22:59
battery side or we'll call microgrid
23:01
solar plus storage and maybe plus a generator
23:04
of some kind right , you have a fully
23:06
grid independent system then , which means
23:09
it could participate in grid
23:11
services programs like demand response
23:13
, which is basically the grid says hey
23:15
, we need power back , you're consuming
23:17
it . Can you go offline and stop drawing
23:19
power from the grid ? We'll pay you for that right
23:22
, and because we have an independent system
23:24
, we can just rely on our battery
23:26
sort of little microgrid system for as long
23:29
as they need us offline . This is on
23:31
a really small scale in my home , of course
23:33
, it's not going to make much of a dent , but when you're talking
23:35
about megawatts , gigawatts , et cetera
23:37
at the data center level , that's very
23:39
meaningful and very helpful to the
23:41
grid .
23:43
Well and , like Massachusetts , is incentivizing
23:45
people through , you know , 0% interest
23:47
loans and other things to kind of start
23:49
generating their own power and selling it back
23:51
to utilities . So there's a motivation
23:53
there . It makes the grid more resilient , it
23:56
can really be an economic benefit to the
23:58
consumer and then , like you said
24:01
, we're starting to build out more capabilities
24:03
on that . The other big part , as you mentioned , was batteries
24:06
, which
24:09
we haven't really talked about . But the idea of using batteries to sort of , you know , flatten out the
24:11
demand curve a little bit . So you
24:13
know , if I'm generating a bunch of power , I'm
24:15
not using all of it . That's stored
24:17
energy somewhere right , could be on-prem
24:20
or local , and then when there's
24:22
a need , there's a store , store of stuff . And so
24:24
, you know , traditional power plants don't
24:26
operate with batteries . Obviously , they just sort
24:28
of generate power and you know
24:30
, and if people need , and if the community
24:33
needs , more power than the plant
24:35
is generating , then that's a problem . That's called
24:37
like a brownout basically , or whatever . So
24:39
, but with batteries you can actually kind
24:42
of level that off a little bit , actually
24:46
kind of level that off a little bit . And we're starting to see batteries being introduced , um
24:48
, obviously for data centers , but you can , you know , going
24:50
into these kind of edge compute situations
24:52
.
24:52
Batteries are a really important part of that too , to
24:55
help make sure that there's a good store of energy
24:57
over time yes , I
24:59
actually think this is where that more of like
25:01
the edge AI compute
25:05
load is really interesting , in
25:07
that you know there's a lot like
25:10
it is quite beneficial
25:12
to be able to participate in these grid services
25:14
programs , right , especially in ERCOT , kind
25:16
of famous for this right , where you can
25:19
do well by going offline
25:21
for an hour or whatever it may be right . And
25:23
so to think that these
25:25
AI loads could be located
25:28
in distributed basis across the US and across
25:30
the world really , and be able to participate in
25:32
these programs
25:35
, not only is that a grid positive , but from an economic
25:37
perspective for the owner
25:40
they can do quite well , and
25:42
so that becomes really interesting
25:44
. In having what we call flexible load resources
25:48
as part of this deployment
25:50
, and traditionally
25:52
I think it's been quite hard to get battery
25:54
adoption . It's still
25:56
a struggle . We're not anywhere near where we would like
25:58
to be . Perhaps
26:07
the growth in digital infrastructure is the catalyst to actually achieve all that we've
26:09
been sort of pushing for , you know , for the last 15 years , at least in my
26:11
my career so far 15 years
26:14
.
26:14
Yeah , yeah Well , and also what's been happening too
26:16
, which is helping is , I mean , if you look at
26:18
going back to the NVIDIA announcements on Blackwell
26:20
, it's about the thing about 3X more
26:23
power efficient than the
26:25
, the hopper um chips
26:27
. So you know , you're
26:29
, of course , it's 1200 watts but it's generating
26:31
a lot more tops or um , you
26:34
know , teraflops , whatever , but it's
26:36
more ai horsepower per
26:38
watt than it used to be quite a bit , and so
26:40
we're seeing a lot more improvement . And then , of course , you
26:42
get into more , you know , real
26:45
edge computing platforms . I mean , look at Qualcomm
26:47
, snapdragon and what Intel's
26:49
doing , and then getting down to the NXPs
26:51
and the STs and everyone else , they're really
26:55
doing incredible amounts of horsepower
26:57
with very little electricity . So
26:59
there's always improvements going on there too . So we're
27:01
seeing power as a part
27:03
of the value prop these days for AI chips
27:05
. Right , it's not just the horsepower , but it's the power
27:08
consumption , because , of course , some of them will be
27:10
running in battery-powered systems
27:12
, whether it's your phone or
27:14
a drone or a robot or whatever , but
27:17
also even when they're on the grid , power
27:19
is an important part of the value prop
27:21
because , like you said , it's a scarce resource
27:23
, it's an expensive resource and I think anyone with an AI . It's a scarce resource
27:26
, it's an expensive resource and I think anyone with an AI strategy also
27:28
needs a power strategy . I mean they have to be
27:30
going hand in hand , right .
27:32
I think that's right . I mean also
27:34
, we can point to the you say
27:36
expensive , we can point to the per
27:39
kilowatt hour charges at least we've been seeing
27:41
and the rise in those over the
27:43
years , right . I don't think we're
27:46
going to see a time where that really comes down
27:48
materially . I think that is heading
27:50
up right and if that assumption is
27:52
true , it also makes
27:54
a really good case for some version of these microgrids
27:57
as well . A lot of them
27:59
already pencil in high power
28:02
price markets , like in California
28:04
, like in much of the Northeast PJM
28:06
territory , et cetera . Right , these are penciling
28:09
out , whereas a handful of years ago they
28:11
didn't pencil as compared to the grid . So
28:13
now you actually have an opportunity where not only is this
28:16
more resilient and providing all
28:18
of the benefits that we just discussed , but it's also
28:20
less expensive and
28:22
can be at a fixed price . So we're
28:24
not going to see this , the escalation you might
28:26
expect from the traditional
28:28
True , yeah , and at first you know going back to
28:31
solar .
28:31
I mean , the panels are more efficient than they used to be and
28:33
all that other stuff . So now it's combination
28:36
of efficiency of the tech and , unfortunately
28:39
, the cost of the power are now , you
28:41
know , meeting at a point where it
28:43
kind of makes a lot of sense to to , to get
28:45
more into these micro grids and thinking
28:47
about that . So , that that's yeah , that's part of the equation . And
28:49
then we also talked about just that
28:52
. You know the big power stories
28:54
for some of these hyperscalers and how they're solving
28:56
that by . You know , leaning into
28:58
energy resources . Let's put it that way
29:00
and we'll see how far they lean . Lean
29:02
in , but they're going to need to
29:04
. You know
29:06
, we know the hyperscalers have whole organizations dedicated to
29:09
power planning and you
29:11
know all this other stuff right , cause it's a big deal
29:13
, it's , it's a it's non-trivial
29:15
stuff . So it'll be really interesting to see how
29:17
all this stuff plays out over the next five , 10
29:19
years . I think if people are looking for , you
29:21
know , career opportunities , I think the power space
29:24
is always , it seems like there's
29:26
a lot of work to do there , right ?
29:28
Especially where it intersects with digital right
29:30
.
29:30
I mean .
29:30
that's the equation Incredible
29:33
.
29:33
Yeah , good . Well
29:35
, bill , it's great having you on the show and
29:39
you know I do . I have it in my to-do list
29:41
to go check out your solar , your solar
29:43
setup at your house .
29:44
Yes . State of
29:47
the art . I was going to say I would
29:49
also say just to the edge side
29:51
, right , I mean
29:53
to all of it . As you know , ubiquity
29:55
is a is a platform company
29:57
and generates . It's focused traditionally
29:59
on fiber , but it's really digital infrastructure
30:01
. They're now in the edge data center business , but we
30:04
look at that business and and
30:06
we think about what's the value proposition
30:09
for the market , right ? What
30:11
we're kind of landing on is you need to be able to
30:13
provide not just a hosted edge
30:15
data center , you need to be able to provide the edge data
30:17
center with the power capacity at
30:20
a given price . That whole package
30:22
has to be together , and
30:27
I think that the industry , the sector , the developers and the , the companies that are
30:29
focusing on working on this problem , need
30:31
that as a takeaway . You need to have both
30:33
. It cannot just be hey , I have a great
30:36
edge data center pod here . It has to be what's
30:38
the total solution ? Can you deliver me , you
30:40
know , x megawatts in
30:42
less than 24 months or less than 36
30:45
months ? Right , Ready to go at this
30:47
power price ? And this is
30:49
really where Ubiquity and its
30:51
edge business anyway , is spending a lot of time . How
30:53
do we deliver that full package
30:55
? Right , but I would say that's , the whole
30:57
sector needs to be thinking along those
31:00
lines .
31:00
Yeah Well , kind of going back to the Brad Smith
31:02
thing I mentioned at the start is , like you know
31:04
, the two fundamental parts
31:06
of the AI economy are power and connectivity
31:09
and they need to be thought of together so
31:11
you can't obviously have one without
31:13
the other and still have a business . But , yeah
31:17
, cool . Well , bill , again thanks , thanks
31:19
a lot for joining us , and
31:21
you know we're at the end of another
31:23
podcast , so hopefully folks
31:25
learned a little bit and really appreciate
31:28
it . Thanks for having me All right
31:32
. Thanks
31:36
for joining us today on the Edge Celsius Show
31:38
. Please subscribe and stay tuned
31:40
for more and check us out online about how
31:42
you can scale your edge compute business
31:50
.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More