May 24, 2025 • 12 mins
Do you know how much energy you're really using when you're generating those cute videos and photos on AI platforms like ChatGPT, Copilot or Gemini? It's likely a lot more than you think. With so many people utilizing these platforms, entering all kinds of prompts, it's very quickly starting to put stress on our power grid. Matthew Carrara of Doble Engineering in Marlborough talks with Nichole about what's really happening behind the scenes, and what needs to be done to ensure our infrastructure is able to keep up.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:07):
From WBZ News Radio in Boston. This is New England
Weekend where each and every week we come together right
here we talk about all the topics important to you
and the place where you live. Great to have you
back with us on this Memorial Day weekend. I'm Nicole Davis.
Perhaps you're working on a project, or you want to
send somebody a cute little video and you only wanted
a few seconds long, and you figure, you know, AI
(00:30):
can do this really quick, why not? So you open
up Chat, GPT or maybe Gemini or co Pilot, and
you say, give me a video of a little dog
on a beach, and I want it to be five
seconds long. And when that video is generating, you don't
realize just how much energy it's actually using. But and
(00:50):
you report this week from the MIT Technology Review really
actually astounded me because it says every time we use
a newer AI model to generate a five second video
using enough energy that could run a microwave NonStop for
more than an hour. You can imagine with everybody generating
huge videos on AI these days, this is very quickly
becoming a problem when it comes to our power grid.
(01:12):
Matthew Carra knows about that. Firsthand. He is the president
of Doble Engineering. This is something he really cares about.
So he's here to help us talk about this and
give us a bit of an education, help us break
this down a bit more. Matthew, thank you so much
for your time and for being here. So I'd love
to hear a bit more insight about this.
Speaker 2 (01:30):
So the challenge around AI is I think the first
factor put on the table is a chat GPT search
uses ten times the amount of energy than a Google search.
Oh wow, So that's always the one that I like
to start with because you're talking constant twenty four hours,
seven days a week power. And one of the challenges
(01:53):
around AI is you don't actually know when the use
is going to be. So if someone launches a new
data side and they put a new version of Gemini
on it, let's say, and all of a sudden people
start start using it, the utilities will see a major
spike in use, and that usually happens around the time
that they that they launch something. If you compare that
(02:14):
to changes in climate seasonality, you know, utility can very
easily understanding the weather forecast say oh it's going to
get cold, We're going to see people using heat, we're
going to see increased demand. Right, think of the same
thing of it's getting warmer and people are going to
use AC. We're going to use increased demand. The problem
on the AI front is you never know when the
usage is going to be and it can happen at
(02:35):
any time.
Speaker 1 (02:37):
Now, hear a lot about water cooling situations. People talk about, Oh,
when you do a Google search, it's a certain amount
of water, but when you do an AI search, you're
using way more water. Where does the water come in here?
Speaker 2 (02:48):
The water cooling actually happens around inside the power systems
within the data center. Okay, it's it's not as prevalent
that as one would think. That's not the major the
major draw The major draw are is really just the
absolute energy demand it needs to run. And you know,
if you think about your home computer for example, So
(03:09):
you have a laptop sitting on your house and it's
drawing power. Take a data center, it's it's one hundred
thousand of those, okay, sitting sitting in a room somewhere,
drawing power. So it's this year magnitude of computing power
that is needed to drive that type of intelligence, which
is what is causing part of the demand on the grid.
Speaker 1 (03:29):
And of course AI is only getting more and more comprehensive.
We keep seeing these new platforms that can generate entire
videos and photos and this and that. It all takes
so much power I can't even imagine.
Speaker 2 (03:42):
Yeah, part of the challenge is so in order to
build a data center, you can build an average data
center and think, you know, Google, Amazon, Microsoft, that could
be done in twelve months. The problem that we're seeing
right now is it takes three years to connect it
to the grid, partly because of backline, partly because of
i'll say weak infrastructure. So the boom that you're seeing
(04:05):
right now, believe it or not, is probably half of
what's really really going on. I mean, I think the
estimates are right now data center use is about four
percent of national load of the draw on the grid.
The rexpectation is that in ten years it will be
fifteen to eighteen. Wow.
Speaker 1 (04:22):
And I mean our grid is not exactly young either.
That's the most important thing to mention here, I feel
is that we have an aging grid, and before AI
really came on the scene, it was already struggling. I
think of all the power outages in Texas, for example,
over the past couple of years, we already have a
system that is struggling to keep up. And this is
not doing us any favors.
Speaker 2 (04:43):
No, I mean the average age of let's say all
the large power transformers that handle the load, so ninety
percent of them currently in place are greater than forty
years old. So you have one outdated equipment. You then
layer on top of that just talk extreme weather events
and extreme weather events being something that can impact power
(05:04):
supply to a city. Those are ravaging around one hundred
a year for the last five years. That's two times
the prior five years. So climate change is having an impact. Again,
people that didn't need air conditioning before now do people
that didn't you keep before now do go back to
what happened in Texas was a couple of years ago
when they had a freeze right first time. Ever, then
(05:28):
put on top of that EV charges and the fact
that you know EV sales are expected to double in
the next the next twenty years. So we keep layering
in things that are draining on the grid, and then
you put on what I call the final straw, which
is AI and yet getting it's definitely struggling and getting overworked.
Speaker 1 (05:47):
I mean, do these companies realize that this is happening
or is this just not a factor when it comes
to this new technology. Because I don't want to be
somebody who says, look, we can't have new technology. I
think it's important. And you know, electrifying a lot of
our services is probably the better idea for climate. But
I mean, how do you keep up here?
Speaker 2 (06:07):
Well? I think the utilities, I think are aware of it.
What their struggle is right now is when does everything
come online? And just because a Microsoft or a Google
or an Amazon says, hey, we're going to build a
data center in New York City, they could turn around
a year from now and not turn it on, and
all of a sudden they've done all this infrastructure. So
(06:29):
I think the challenges are the speed at which they're
being built and how much truly comes online or not.
But if you take, for example, just go to the
big ones like Facebook, aama is on Microsoft, they're actually
looking at developing and buying their own renewable sites, so
their own solar and wind farm generation, so they can
generate and supply their own power. Microsoft, I think it
(06:51):
was last year began leasing what used to be three
mile island. They can generate their own power. So I
think the companies are aware that they need it, but
we're limited in our ability to expand the infrastructure quickly
and get it to where it needs to be to
handle all the volume well.
Speaker 1 (07:07):
And not to mention too, we're dealing with all these
tariffs and we've got trade issues, and we don't manufacture
a lot of these parts and supplies that we need
here in the United States. How do you think that's contributing.
Speaker 2 (07:19):
Yeah, there's two factors. If you look at transformers being
the predominant piece of asset that you need to step
down power to deliver it to a data center, which
or were even a home. If you were to go
back even three years ago, the delivery time of your
basic transformer was like eight to ten months. It's now
three to five years. Oh wow. Because so many aspects
(07:43):
of the grid are getting updated and there's such a
demand for more power right now, the industry is struggling
with getting equipment. You then add on top of that
that any equipment that's being built overseas now gets whacked
with a tariff. Costs of going up and delivery times
are going getting longer.
Speaker 1 (08:01):
So how do we handle this? I mean, clearly, this
is a lot of problems being thrown at this situation.
At the same time, I mean, how do we get
ahead of this? Or can we get ahead of this situation?
Speaker 2 (08:12):
Well, I always joke when I get asked that question,
I'm like, well, you can just turn most of your
lights off in your house.
Speaker 1 (08:16):
But oh sure, Okay, that's not really the answer for.
Speaker 2 (08:20):
The average person. But a part of it is is
better forecasting, a better understanding of where the load is
going to be and what's expected. You know, I was
actually reading an article the other day. It's about so
Virginia is considered the highest growth rate for data centers.
Has to do with tax breaks, has to do with
a lot of land. They believe that the investment in
(08:43):
Virginia Loan over the next ten years is going to
be a trillion dollars. Like this, think about the massiveness
of that and the amount of energy that the data
centers in Virginia Loan generate is the equivalent to four
hundred thousand evs on an annual basis. What we need
to do is we just need to better understand what's
(09:04):
coming down from an investment when are these things going
to go on? And we just need to really start
doing a lot of heavy lifting on investing in more
equipment and more infrastructure in the power grid. I mean,
it sounds simple, but the problems with that are, again,
are available money, are people to do the work the
equipment that you need to install to make a new substation.
(09:26):
So there's so many aspects that are going on. Do
I think we're at a point where the grid will break? No.
Do I think we're at a point where we'll overload it. No,
But I think, well, we could start seeing what we
see in the past, like in Texas and California, where
you start seeing rowland blackouts. So I think what the
challenge will be is are there ways that we can
get a forecast so we know how to redistribute power
(09:48):
when needed.
Speaker 1 (09:49):
Do you think that we have the technology, Do you
think that we have the resources we need to get
to that point or how do you think that we
need to build up to make sure that we can
get that in place and make this happen.
Speaker 2 (10:02):
I think we have I think we have the resources.
I think we have the knowledge. But one of the
challenges the sheer magnitude of the cost is what limits
a lot of the utilities from doing it, because anything
that they do gets passed down to the consumer at
the end of the day. And I think i'd seen
the data point the other day that they believe if
the data center rate continues, the consumer will see a
(10:26):
one percent increase in their bill every year for the
next fifteen years. Wow. So the problem is who's going
to pay for it? Now, there are studies being done
where the utilities are saying to the Googles and the
Amazons of the world, will connect you and will run it,
but you're paying for it, which is happening a lot.
(10:46):
But in some cases, if the utility makes the investment
and the data center underestimates its load or the data
center doesn't open, yeah, the consumer is going to pay.
Speaker 1 (10:56):
And speaking of the consumer, you know somebody listening to
this saying, oh my gosh, I use Chap GPT all
the time, or I use co pilot, or I have
a whole smart house. I'm contributing to this. I mean,
first and foremost, one individual is not going to change things.
But if somebody wants to make a change and try
to help the situation, what do you advise them to do.
Speaker 2 (11:17):
I mean, at the consumer level, it would be solar.
I mean, let's be honest, if you have the ability
to put solar panels in your home to capture and
store that power, that would dramatically help. And also in
some cases if you're creating extra power, you can push
it back into the grid in that manner too. That's
really from the consumer level all you can do. I mean,
(11:39):
I joked earlier, Yeah, you could have people start turning
lights off, but I don't know how practical that is.
But energy efficiency in appliances, energy efficiency in your home,
even commercial, not just residential, but industrials, are there ways
that they can help you drive energy used down. And
that's where you start getting into decobtization and clean energy.
(12:00):
So there are ways to get it. But you're right,
the problem is one person can't do well.
Speaker 1 (12:05):
No, they can't. But together we can get it done.
And you know, these are some great ideas you bring
to the table, and with any luck, we'll start to
see them play out sooner than later, I hope.
Speaker 2 (12:13):
So, but it does keep Doble in business, so I
can't complain fair.
Speaker 1 (12:18):
That is fair. That is fair, all right, Well, Matthew,
thank you so much for this information. This has been
a really good educational segment for me and for I
think all our listeners. So I appreciate the time and
all you're doing at double to try to get ahead
of it.
Speaker 2 (12:31):
No, appreciate the time, and if you have any other
questions of Cole, just reach out and my problem.
Speaker 1 (12:35):
Have a safe and healthy weekend. Please join me again
next week for another edition of the show. I'm Nicole
Davis from WBZ News Radio on iHeartRadio.
From WBZ News Radio in Boston. This is New England
Weekend where each and every week we come together right
here we talk about all the topics important to you
and the place where you live. Great to have you
back with us on this Memorial Day weekend. I'm Nicole Davis.
Perhaps you're working on a project, or you want to
send somebody a cute little video and you only wanted
a few seconds long, and you figure, you know, AI
(00:30):
can do this really quick, why not? So you open
up Chat, GPT or maybe Gemini or co Pilot, and
you say, give me a video of a little dog
on a beach, and I want it to be five
seconds long. And when that video is generating, you don't
realize just how much energy it's actually using. But and
(00:50):
you report this week from the MIT Technology Review really
actually astounded me because it says every time we use
a newer AI model to generate a five second video
using enough energy that could run a microwave NonStop for
more than an hour. You can imagine with everybody generating
huge videos on AI these days, this is very quickly
becoming a problem when it comes to our power grid.
(01:12):
Matthew Carra knows about that. Firsthand. He is the president
of Doble Engineering. This is something he really cares about.
So he's here to help us talk about this and
give us a bit of an education, help us break
this down a bit more. Matthew, thank you so much
for your time and for being here. So I'd love
to hear a bit more insight about this.
Speaker 2 (01:30):
So the challenge around AI is I think the first
factor put on the table is a chat GPT search
uses ten times the amount of energy than a Google search.
Oh wow, So that's always the one that I like
to start with because you're talking constant twenty four hours,
seven days a week power. And one of the challenges
(01:53):
around AI is you don't actually know when the use
is going to be. So if someone launches a new
data side and they put a new version of Gemini
on it, let's say, and all of a sudden people
start start using it, the utilities will see a major
spike in use, and that usually happens around the time
that they that they launch something. If you compare that
(02:14):
to changes in climate seasonality, you know, utility can very
easily understanding the weather forecast say oh it's going to
get cold, We're going to see people using heat, we're
going to see increased demand. Right, think of the same
thing of it's getting warmer and people are going to
use AC. We're going to use increased demand. The problem
on the AI front is you never know when the
usage is going to be and it can happen at
(02:35):
any time.
Speaker 1 (02:37):
Now, hear a lot about water cooling situations. People talk about, Oh,
when you do a Google search, it's a certain amount
of water, but when you do an AI search, you're
using way more water. Where does the water come in here?
Speaker 2 (02:48):
The water cooling actually happens around inside the power systems
within the data center. Okay, it's it's not as prevalent
that as one would think. That's not the major the
major draw The major draw are is really just the
absolute energy demand it needs to run. And you know,
if you think about your home computer for example, So
(03:09):
you have a laptop sitting on your house and it's
drawing power. Take a data center, it's it's one hundred
thousand of those, okay, sitting sitting in a room somewhere,
drawing power. So it's this year magnitude of computing power
that is needed to drive that type of intelligence, which
is what is causing part of the demand on the grid.
Speaker 1 (03:29):
And of course AI is only getting more and more comprehensive.
We keep seeing these new platforms that can generate entire
videos and photos and this and that. It all takes
so much power I can't even imagine.
Speaker 2 (03:42):
Yeah, part of the challenge is so in order to
build a data center, you can build an average data
center and think, you know, Google, Amazon, Microsoft, that could
be done in twelve months. The problem that we're seeing
right now is it takes three years to connect it
to the grid, partly because of backline, partly because of
i'll say weak infrastructure. So the boom that you're seeing
(04:05):
right now, believe it or not, is probably half of
what's really really going on. I mean, I think the
estimates are right now data center use is about four
percent of national load of the draw on the grid.
The rexpectation is that in ten years it will be
fifteen to eighteen. Wow.
Speaker 1 (04:22):
And I mean our grid is not exactly young either.
That's the most important thing to mention here, I feel
is that we have an aging grid, and before AI
really came on the scene, it was already struggling. I
think of all the power outages in Texas, for example,
over the past couple of years, we already have a
system that is struggling to keep up. And this is
not doing us any favors.
Speaker 2 (04:43):
No, I mean the average age of let's say all
the large power transformers that handle the load, so ninety
percent of them currently in place are greater than forty
years old. So you have one outdated equipment. You then
layer on top of that just talk extreme weather events
and extreme weather events being something that can impact power
(05:04):
supply to a city. Those are ravaging around one hundred
a year for the last five years. That's two times
the prior five years. So climate change is having an impact. Again,
people that didn't need air conditioning before now do people
that didn't you keep before now do go back to
what happened in Texas was a couple of years ago
when they had a freeze right first time. Ever, then
(05:28):
put on top of that EV charges and the fact
that you know EV sales are expected to double in
the next the next twenty years. So we keep layering
in things that are draining on the grid, and then
you put on what I call the final straw, which
is AI and yet getting it's definitely struggling and getting overworked.
Speaker 1 (05:47):
I mean, do these companies realize that this is happening
or is this just not a factor when it comes
to this new technology. Because I don't want to be
somebody who says, look, we can't have new technology. I
think it's important. And you know, electrifying a lot of
our services is probably the better idea for climate. But
I mean, how do you keep up here?
Speaker 2 (06:07):
Well? I think the utilities, I think are aware of it.
What their struggle is right now is when does everything
come online? And just because a Microsoft or a Google
or an Amazon says, hey, we're going to build a
data center in New York City, they could turn around
a year from now and not turn it on, and
all of a sudden they've done all this infrastructure. So
(06:29):
I think the challenges are the speed at which they're
being built and how much truly comes online or not.
But if you take, for example, just go to the
big ones like Facebook, aama is on Microsoft, they're actually
looking at developing and buying their own renewable sites, so
their own solar and wind farm generation, so they can
generate and supply their own power. Microsoft, I think it
(06:51):
was last year began leasing what used to be three
mile island. They can generate their own power. So I
think the companies are aware that they need it, but
we're limited in our ability to expand the infrastructure quickly
and get it to where it needs to be to
handle all the volume well.
Speaker 1 (07:07):
And not to mention too, we're dealing with all these
tariffs and we've got trade issues, and we don't manufacture
a lot of these parts and supplies that we need
here in the United States. How do you think that's contributing.
Speaker 2 (07:19):
Yeah, there's two factors. If you look at transformers being
the predominant piece of asset that you need to step
down power to deliver it to a data center, which
or were even a home. If you were to go
back even three years ago, the delivery time of your
basic transformer was like eight to ten months. It's now
three to five years. Oh wow. Because so many aspects
(07:43):
of the grid are getting updated and there's such a
demand for more power right now, the industry is struggling
with getting equipment. You then add on top of that
that any equipment that's being built overseas now gets whacked
with a tariff. Costs of going up and delivery times
are going getting longer.
Speaker 1 (08:01):
So how do we handle this? I mean, clearly, this
is a lot of problems being thrown at this situation.
At the same time, I mean, how do we get
ahead of this? Or can we get ahead of this situation?
Speaker 2 (08:12):
Well, I always joke when I get asked that question,
I'm like, well, you can just turn most of your
lights off in your house.
Speaker 1 (08:16):
But oh sure, Okay, that's not really the answer for.
Speaker 2 (08:20):
The average person. But a part of it is is
better forecasting, a better understanding of where the load is
going to be and what's expected. You know, I was
actually reading an article the other day. It's about so
Virginia is considered the highest growth rate for data centers.
Has to do with tax breaks, has to do with
a lot of land. They believe that the investment in
(08:43):
Virginia Loan over the next ten years is going to
be a trillion dollars. Like this, think about the massiveness
of that and the amount of energy that the data
centers in Virginia Loan generate is the equivalent to four
hundred thousand evs on an annual basis. What we need
to do is we just need to better understand what's
(09:04):
coming down from an investment when are these things going
to go on? And we just need to really start
doing a lot of heavy lifting on investing in more
equipment and more infrastructure in the power grid. I mean,
it sounds simple, but the problems with that are, again,
are available money, are people to do the work the
equipment that you need to install to make a new substation.
(09:26):
So there's so many aspects that are going on. Do
I think we're at a point where the grid will break? No.
Do I think we're at a point where we'll overload it. No,
But I think, well, we could start seeing what we
see in the past, like in Texas and California, where
you start seeing rowland blackouts. So I think what the
challenge will be is are there ways that we can
get a forecast so we know how to redistribute power
(09:48):
when needed.
Speaker 1 (09:49):
Do you think that we have the technology, Do you
think that we have the resources we need to get
to that point or how do you think that we
need to build up to make sure that we can
get that in place and make this happen.
Speaker 2 (10:02):
I think we have I think we have the resources.
I think we have the knowledge. But one of the
challenges the sheer magnitude of the cost is what limits
a lot of the utilities from doing it, because anything
that they do gets passed down to the consumer at
the end of the day. And I think i'd seen
the data point the other day that they believe if
the data center rate continues, the consumer will see a
(10:26):
one percent increase in their bill every year for the
next fifteen years. Wow. So the problem is who's going
to pay for it? Now, there are studies being done
where the utilities are saying to the Googles and the
Amazons of the world, will connect you and will run it,
but you're paying for it, which is happening a lot.
(10:46):
But in some cases, if the utility makes the investment
and the data center underestimates its load or the data
center doesn't open, yeah, the consumer is going to pay.
Speaker 1 (10:56):
And speaking of the consumer, you know somebody listening to
this saying, oh my gosh, I use Chap GPT all
the time, or I use co pilot, or I have
a whole smart house. I'm contributing to this. I mean,
first and foremost, one individual is not going to change things.
But if somebody wants to make a change and try
to help the situation, what do you advise them to do.
Speaker 2 (11:17):
I mean, at the consumer level, it would be solar.
I mean, let's be honest, if you have the ability
to put solar panels in your home to capture and
store that power, that would dramatically help. And also in
some cases if you're creating extra power, you can push
it back into the grid in that manner too. That's
really from the consumer level all you can do. I mean,
(11:39):
I joked earlier, Yeah, you could have people start turning
lights off, but I don't know how practical that is.
But energy efficiency in appliances, energy efficiency in your home,
even commercial, not just residential, but industrials, are there ways
that they can help you drive energy used down. And
that's where you start getting into decobtization and clean energy.
(12:00):
So there are ways to get it. But you're right,
the problem is one person can't do well.
Speaker 1 (12:05):
No, they can't. But together we can get it done.
And you know, these are some great ideas you bring
to the table, and with any luck, we'll start to
see them play out sooner than later, I hope.
Speaker 2 (12:13):
So, but it does keep Doble in business, so I
can't complain fair.
Speaker 1 (12:18):
That is fair. That is fair, all right, Well, Matthew,
thank you so much for this information. This has been
a really good educational segment for me and for I
think all our listeners. So I appreciate the time and
all you're doing at double to try to get ahead
of it.
Speaker 2 (12:31):
No, appreciate the time, and if you have any other
questions of Cole, just reach out and my problem.
Speaker 1 (12:35):
Have a safe and healthy weekend. Please join me again
next week for another edition of the show. I'm Nicole
Davis from WBZ News Radio on iHeartRadio.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.