Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:02
Latitude Media, podcast at the
0:04
frontier of climate technology. So
0:08
last week we featured an interview with Brian Janis,
0:10
a former Microsoft VP who had a front row
0:12
seat to the AI energy boom and all the
0:14
grid constraints that are coming with it. It
0:17
was one of a few conversations we recorded from
0:19
Distribute Tech on the artificial intelligence theme. And
0:22
look, if you've been listening to this show
0:24
for a while, it's no secret that we
0:26
have AI on the brain at Latitude Media
0:29
because there's legitimately real commercial activity happening. In
0:31
fact, just this week we reported on NVIDIA
0:33
and Utilidata's partnership with Meter Maker Eclera to
0:35
roll out embedded AI to Smart Meters. Also
0:39
because the pathways to getting it embedded across
0:41
the power system are not simple. A
0:43
smart meter with 100 times the processing power
0:46
is a very cool technology, but utilities have
0:48
to make the case to pay for it.
0:51
And their track record for making good use
0:53
of the previous generation of smart meters is
0:55
spotty. There are a lot of
0:57
things that could hold back AI. The
0:59
biggie is such a common problem, it's almost
1:02
become cliche. It comes back to the way
1:04
data is managed or shared across a utility
1:06
or the lack of sharing. If
1:08
you talk to any vendor, this is one of
1:11
the biggest sources of frustration. And
1:13
so for a better understanding of how to change that,
1:15
I turned to a guy who spends a
1:17
lot of his time in the so-called data
1:19
cloud, Titian Palazzi, the head of power and
1:21
utilities at Snowflake. Any kind
1:24
of data naturally ends up in
1:26
different boxes, in different silos. And
1:28
when you then want to ask questions of the
1:31
data, it becomes really hard. You can't ask questions
1:33
across the enterprise. And Snowflake
1:35
is a data cloud platform. And
1:37
we sit on top of Amazon,
1:39
Microsoft, and Google, helping companies to
1:41
bring all their data together to
1:43
then create value out of it.
1:48
Snowflake is one of the hottest tech companies
1:50
that you may not have heard of. It
1:52
has a $53 billion market cap with thousands
1:54
of customers. Titian arrived there
1:56
after his company, Mist AI, was acquired by
1:59
Snowflake last year. year. In 2018, Titian
2:01
co-founded Myth with Peter Verhoeven, who built
2:03
some critical demand response applications for the
2:06
NEST thermostat. And Myth was focused on
2:08
time series forecasting for the grid. Over
2:10
the last 30 years, linear regression
2:13
or regression models were one of the
2:16
main ways in which forecasting was done.
2:18
So you would actually specify specific weights
2:20
for every factor that determined the forecast.
2:22
Let's say you want to anticipate the
2:25
output of a wind Under
2:27
linear regression, you apply weights to things
2:29
like wind speed, temperature, blade performance, and
2:31
then do some statistical analysis to make
2:34
a prediction. But time is
2:36
also an important factor. And
2:38
machine learning made it easier to
2:40
integrate temporal factors into forecasts. What
2:42
you saw was that a lot
2:44
of the progress which was made
2:46
in natural language processing, so the
2:48
ability for an AI algorithm to
2:50
understand words, applied to time series
2:53
data, because in the same way
2:55
that the words in a sentence,
2:57
their order matters, so
2:59
too, timestamps matter when you're doing a
3:01
forecast of something like energy demand. So
3:03
these were models in which you would not
3:06
specify any hard coded weights, but
3:08
you would basically tell the model, here are
3:10
all the things that will have an influence
3:12
on the thing we're trying to predict, such
3:14
as the output of that 200 megawatt wind
3:16
park come up with a prediction. What
3:18
they found was that these AI driven
3:20
time series models could improve accuracy by
3:23
30 to 50 percent. And that caught
3:25
the attention of Snowflake, where Titian and
3:27
his co-founder are now working on product
3:29
development and go-to-market strategies in energy. I
3:32
think that's a big shift that has taken
3:34
place in production by companies that are serving
3:37
real customers. AI is used much
3:39
more commonly, so a shift from hard
3:41
coded, pre-set up, fully visible
3:43
models into more machine learning
3:45
and AI. But
3:50
these models do have an Achilles heel.
3:52
You need access to lots of clean
3:54
data, And we still have a long
3:56
way to go to unlock their full potential. The
4:00
Industry: There is a lot of
4:02
time series data coming from the
4:04
grid or power generation or electric
4:06
cars or even from electricity. Nothing's
4:09
at the same time. Using a
4:11
hard for forecasting is quite challenging
4:13
because every time you need to
4:15
create a new prediction unique to
4:17
have the latest data. and so
4:19
from an engineering perspective it was
4:22
quite complicated to do. This
4:26
is the carbon copy and Steven lazy. And.
4:30
This week, a conversation with snowflakes titian
4:32
policy on busting data silos some early
4:34
wins for a I in the power
4:37
sector, and what size of the transition
4:39
would. I
4:44
want to take a brief moment to talk
4:46
about the new season of the Big Switch
4:48
podcast. We've been working on this for the
4:51
last six months for so excited to bring
4:53
it to you Are production team at Latitude
4:55
Media has been working for years, the Doctor
4:57
Melissa Law in the team at Columbia University
5:00
Center on Global Energy Policy or To Make
5:02
The Big Switch it's a narrative, show them
5:04
how to rebuild our energy systems and we're
5:06
back with a five part series exploring the
5:09
supply chains behind lithium ion batteries and a
5:11
very complicated economic and political forces that com
5:13
as batteries. Take Over the World. So
5:15
in this season we break batteries apart,
5:17
go to mining operations, manufacturing facilities, recycling
5:19
plants, and talk to some of the
5:22
most prominent experts about the pitfalls and
5:24
promise of are expanding battery based energy
5:26
economy and you'll hear the trailer a
5:28
bit later in the shell. So if
5:30
this sounds like something you wanna listen
5:32
to, find a big switch Anywhere You
5:34
get your podcasts. So
5:41
where are we today in the
5:43
advancement in adoption of a I
5:46
like what's your read on the
5:48
technological moment we're in, right? I.
5:50
Would say at there are. A. Few real
5:52
changes that are happening. One
5:55
of them is. That
5:57
we're seeing. A new type
5:59
of model Generative A I was really
6:01
perform extremely well. And what that means
6:04
is that some of the previous ways
6:06
in which the I am now we're
6:08
done my at disappear become less relevant.
6:11
So for example, it's quite possible. but
6:13
in the not too distant future you
6:15
might actually do forecasting by asking another
6:18
lamps can you predict the next. Few
6:20
weeks of data for as to date
6:22
you would use very different algorithms for
6:24
net. I think the other big thing
6:26
that's happening is that previously whenever you
6:29
wanted to use an I mobile you
6:31
have to write code or maybe you
6:33
could use a user interface but now
6:35
we're able to use the I would
6:37
just natural language and I think because
6:39
of death and broadening of the aperture
6:41
that ability for people who. In
6:43
plain English or any other language of
6:46
their choice and ask questions. I think
6:48
you're seeing an opening up of the
6:50
user base and that actually leads to
6:52
the models becoming better since his kind
6:54
of cycle and cyclical effect and I
6:56
think that's quite powerful and I think
6:58
it will accelerate development. So.
7:00
Now you're a snowflake after Miss Day
7:02
I was acquire last year out what
7:04
is snowflakes approach to building out that
7:06
the data cloud And how does energy
7:08
actually fit into that? Yes. So. In.
7:11
Companies of all types data
7:13
get silos as the company
7:15
cross. It's a bit like
7:17
entropy. It happens naturally unless
7:19
you try to prevent it's
7:22
and snowflakes. Aim and vision
7:24
is to mobilize the data
7:26
so that business leaders can
7:28
ask questions that span across
7:30
the business. for example which.
7:33
Parts. Of our grid should we do
7:35
maintenance first to minimise a chance of
7:37
there's an outage, or how much capital
7:39
should we allocate to make sure that
7:41
we meet the risk requirements of a
7:43
market like are called. The
7:45
sofa data club sits on top
7:47
of the three. Cloud. Hyper
7:49
Skaters so we don't own
7:51
our own data centers and
7:53
as companies. Used as data
7:56
Cloud and Store or that the
7:58
clouds that and unlocks workloads. Like.
8:00
A I am machine learning
8:02
or applications that run on
8:05
their data or visualizations would
8:07
be really. A lock
8:09
is the ability to combine I
8:11
T date also data around at
8:13
unit sales or supply chain data
8:15
with Ot data which is data
8:17
from devices. so data from wind
8:19
turbines are data from the grid.
8:22
And by combining those two things
8:24
that really unlocks a lot of
8:26
these valuable insights for executives. the
8:28
take action on. City. Utility
8:30
industry is one of the greatest data generators
8:32
have any industry. If I'm sitting inside a
8:34
toilet, Sam and engineer a program manager and
8:37
I need access to certain data stream to
8:39
make sense of it. Like one of the
8:41
current limitations today. it's what do I see
8:43
where I have access to one of the
8:45
limits and dance? What is A Data clouds
8:48
functionally deliver this difference. Yes, I
8:50
totally agree. There's. So much
8:52
daytime the power utility sector to give
8:54
two examples. One is when it comes
8:56
to have to see market every five
8:58
minutes, every values time stamps or then
9:00
when you look at things like the
9:02
grid or at power generation. You
9:05
might have at data for every second
9:07
or millisecond that something is an operation
9:09
so completely agreed. Huge volumes of data
9:11
and debts or so I think were
9:13
really sometimes a challenge. Nice. So. Let's
9:16
say that you are Rico Energy provider and you
9:18
want to forecast electricity demand for your customers for
9:21
the next few weeks to mixture of a do.
9:23
At. By adequate supply of for
9:25
your customers would I've seen is at
9:28
often it can take three months, six
9:30
months or nine months to get access
9:32
to all the right data such as
9:34
smart meters the for your customer and
9:37
then only a couple of weeks or
9:39
couple of months to actually build and
9:41
deploy a predictive model. So would you
9:43
see that utilities and other companies in
9:46
the power sector of money to spend
9:48
a much larger amount of time on
9:50
data collection flaming then on the actual
9:52
a I'm at. Work. So.
9:55
i think one of the real challenges
9:57
is to bring all the data into
9:59
one place establish a strong foundation
10:01
on top of which you can build
10:04
predictive models and other AI. Why
10:06
is this a problem for AI?
10:08
This lack of clean data or lack
10:10
of access to data? I mean, it's
10:13
very obvious on its face. In
10:15
order to build powerful models, you need a lot
10:17
of information. But what are some examples of how
10:20
this holds AI applications back? Well,
10:22
I think the real problem is that as
10:25
part of the energy transition, there's so many challenges
10:27
that we need to solve really fast. For
10:30
example, I work with a
10:32
number of distribution system
10:34
operators and transmission system operators
10:37
that need to deal with enormous
10:40
amounts of new
10:42
interconnection requests for solar, wind, and
10:44
battery storage. And the reality
10:46
is that many of them don't really
10:48
have visibility into their network in a
10:50
way that allows them to dynamically manage
10:53
both new resources being added and
10:55
then to ensure that the grid
10:57
operates in a reliable manner. And
10:59
then as a result, what they have to do
11:01
right now is they have to essentially limit
11:04
with all kinds of levels of safety,
11:07
how much can be added. So
11:09
in some ways, the energy transition
11:12
is being blocked by the
11:14
ability for energy companies to have access
11:16
to that data. I work with a
11:18
number of grid operators around the world.
11:21
Many of these companies are trying to
11:23
add more solar, wind, battery storage, and
11:25
other renewables to their grids,
11:27
as well as more flexible demand. And
11:30
one of the key challenges they face is that
11:32
they don't have good visibility into what's actually happening
11:34
on their grid. So that's just
11:37
one example of where access to data
11:39
is holding us back. And
11:41
these companies are typically engaged in
11:44
enormous programs with thousands of people just to make
11:46
sure that there are sensors on the grid and
11:48
that the data from those sensors is captured somewhere
11:50
in the cloud so that they can then run
11:53
analytics on top of that. I'm
12:01
Dr. Melissa Lott and I'm the host of
12:03
The Big Switch, a show about how to
12:06
rebuild our energy systems. Batteries
12:10
are finding their way into everything, from
12:12
cars and heavy equipment to the electric grid.
12:15
But scaling up production to meet
12:17
the demands of a net zero
12:19
economy is complicated and it's contentious.
12:22
If every country says we need to own the entire
12:24
supply chain because we want all of those economic benefits,
12:26
it's going to make the clean energy transition so much
12:28
harder. In a new five-part series,
12:31
we're digging into the global battery
12:33
supply chain, from mining to manufacturing.
12:35
And we're asking what gets mined,
12:37
traded and consumed on the road
12:39
to decarbonization. If we think
12:42
climate change is the existential threat facing our
12:44
planet, we have to be having a broad
12:46
conversation about where we want to get the
12:48
minerals that fill these products.
12:51
Listen to The Big Switch from Columbia
12:53
University's SIPA Center on Global Energy Policy,
12:55
available on February 28th, wherever you get
12:57
your podcasts. Well,
13:05
let's go to a couple of examples.
13:07
You've worked with a range of utilities
13:09
and retail energy providers and you said
13:11
grid operators on using AI for asset
13:13
management, for forecasting. Where
13:16
are the most compelling applications you're seeing today? When
13:19
I think of all these use
13:21
cases, I typically think of three areas. The
13:24
first is assets and operations. So
13:26
that has everything to do with
13:28
the physical infrastructure, grid, power generation,
13:30
et cetera. The next is finance
13:32
and markets. So that has everything
13:34
to do around power markets. And
13:37
the third is everything around customer, customer
13:39
360. So making sure that
13:41
every individual who is part of this energy
13:43
transition is treated in
13:46
the right way. So first, in assets and
13:48
operations, two examples come to mind. One
13:51
is we work with a retail energy
13:54
provider in the Midwest that serves more than a
13:56
million customers with electricity, gas, and energy. And the
13:58
second is the cost of the company. and
14:00
distributed energy resources. So they have an
14:02
offering, for example, for rooftop solar. And
14:05
I believe about 20,000 of their customers have
14:08
rooftop solar through the energy retailer.
14:11
Now, sometimes issues crop up
14:13
with rooftop solar. There might
14:15
be soiling on the panel, so the
14:17
panel might be dirty, or maybe the
14:19
wires weren't connected properly and the panel isn't
14:21
actually producing, or maybe a tree grew
14:23
and now there's more shading. And historically,
14:26
this company basically had to wait until a
14:28
customer called and said, I'm looking
14:30
at my bill and it doesn't look like the solar is working
14:33
as it used to be. Can you
14:35
come and investigate? And they built, based
14:37
on the smart meter data coming from
14:40
the solar panels, a predictive model that
14:42
basically says, here's how much
14:44
solar we would expect based
14:47
on the location
14:49
of the solar system, the tilt of the
14:51
roof, these kinds of things. Here's
14:53
what we're actually seeing. And if there is
14:55
a big discrepancy, then the customer service team
14:58
will get an alert and they will either
15:00
call the customer or send a crew to
15:02
go and check. So now often,
15:04
before the customer has even noticed that something
15:06
is going on with their bill, the
15:09
retail energy provider has already contacted them. Another
15:12
example in the same group is, we work
15:14
with one of the biggest renewable asset owners
15:16
in the world, LightSource, which is a part
15:18
of BP. They
15:21
operate more than five gigawatts of
15:23
solar globally. And a big
15:25
problem for utility-skilled solar
15:28
is hillstorms. So some of the
15:31
bulk of insurance claims for solar asset
15:33
owners actually hill damage to panels, just
15:36
breaking the glass. And
15:38
so they developed a predictive
15:40
system where they're incorporating hill
15:42
warnings from a variety of weather sources.
15:45
And when hill is expected to
15:48
come near a solar system, a
15:51
utility-skilled solar system, they will turn
15:54
the panels, because many panels can be turned
15:57
On a single-axis tracker, so that they
15:59
are... Position frantically so that the
16:01
hill doesn't damage to panels and thereby the
16:03
stay. Avoid Huge. Out the
16:05
just as well as financial cost
16:07
of replacing the panels. Another example
16:10
is more around the wholesale market
16:12
side so. A utility we
16:14
work with closely that has about
16:16
a million an electric and gas
16:18
customers. They collect the smart meter
16:21
data or am I data for
16:23
all their customers in snowflakes and
16:25
with that allows him to do
16:27
is that day Now. Create.
16:30
Electricity demand forecasts for the next
16:32
few weeks based on all the
16:34
data would is amazing is that
16:36
this data science team the typically
16:38
has very little interaction with the
16:40
business suddenly got a call from
16:42
the chief financial officer of the
16:45
company because apparently by improving their
16:47
predictions they save more than five
16:49
million dollars in a single month
16:51
by avoiding exposure to a bit
16:53
real time prospect. It's a
16:55
good less cause it illustrates just
16:57
how why the applications are and if
16:59
you think about where adoption is
17:01
in the power sector broadly enough we
17:04
think on on one end of the
17:06
spectrum where any enhancements sais and
17:08
then maybe at the other end of
17:10
the spectrum it's full automation, human out
17:13
of the loop. like where are
17:15
we in that Sais and and what
17:17
is the eventual say? Steve, These
17:19
things by the robots are not yet
17:21
taking over and I air. Utilities
17:25
are not known to be the
17:27
fastest adopters, so. Although. There
17:29
are studies that will say that a very
17:31
high percentage of utilities are. Already
17:34
or see the the importance of a
17:36
I I think the A Monkey Timothy
17:39
said have fully production I use Cases
17:41
for a I Live is relatively small,
17:43
and I think that that is the
17:46
reflection of a few things. First is,
17:48
as we said earlier, it's sometimes hard
17:50
to get all the data. Second is
17:53
that utilities don't always have the most
17:55
sophisticated difficulties. asserting might
17:57
also be culture what i see in
18:00
other industries is that what's really
18:02
powerful for an organization is to
18:05
have the culture and the technical systems
18:08
that allow staff to experiment and to
18:10
make small prototypes for things. Because
18:13
it's now so easy to put things together with
18:16
LLMs and existing tools, that
18:19
it is quite powerful if your teams can
18:21
explore and try new things and see what
18:23
sticks. I think that there's a long way
18:25
to go. And I think that what
18:28
we will see is that AI
18:31
will drive efficiency throughout the
18:33
process. It will be used in all kinds
18:35
of business divisions. It will be used
18:37
in the form of standalone
18:39
applications sold by software companies,
18:42
as well as homegrown solutions
18:44
built on their own cloud
18:46
platforms. And I'm excited to see what's
18:48
coming. So let's go deeper about what
18:50
needs to happen inside a utility
18:52
to grapple with that transition
18:54
and actually make appropriate investments.
18:58
Many utilities are partnering with outside
19:00
teams. Some are hiring their own
19:03
in-house data science teams, AI experts,
19:05
some of whom don't have experience
19:07
in the power sector. They may
19:11
be AI specific experts who are just now
19:13
figuring out how to apply it to the
19:15
power sector. And so you have a number
19:17
of different approaches inside of utilities.
19:19
What are the most common approaches now? And
19:22
are there any that you think are particularly
19:24
effective when it comes to team
19:26
building and then eventually like testing out
19:29
products and collaborating and building this sandbox
19:31
approach so that they can figure out
19:33
how to actually scale an application? Yes,
19:36
I think that there are a few key components.
19:39
One is, as you mentioned, unlocking
19:41
a data foundation in which all the data is there
19:43
so that you can actually run
19:46
models. The second
19:48
one is around experimentation.
19:51
So something specific
19:53
I've seen work very well is to
19:55
embed teams of data
19:58
scientists and software engineers with the... lines
20:00
of business. So to make sure that somebody
20:02
who is a data scientist is actually sitting with
20:04
a trader for a week, or sitting with the
20:07
grid management team for a week, or
20:09
sitting with line crew for a
20:11
week, and maybe even going with them, to get
20:13
a clearer sense of the issues
20:15
that they grapple with. And then part two of
20:17
that is setting
20:19
up those technical teams to experiment
20:22
quickly. So for example, creating sandbox
20:24
environments that do not have the
20:27
same requirements as the fully skilled
20:29
production models that power the utilities
20:31
and customers, in which
20:34
they can, within a matter of days,
20:36
deploy new applications that are initially
20:38
used by just a small subset of users
20:40
to see what works and what doesn't. The
20:43
final thing I'll say on accelerating AI adoption
20:45
is that I think
20:47
sharing data in the utility industry
20:51
is not yet delivering on its full
20:53
promise. So one of the amazing things
20:55
about some parts of utilities is that
20:57
they're actually not competitive. And
21:00
as an example, in California, the
21:02
smart meter data that the three
21:04
big investor-owned utilities have actually
21:07
should be shared with a variety
21:09
of different players. So for example,
21:11
the distributed energy companies like Sunrun
21:13
or Sunpower, they
21:16
can really benefit from having smart meter
21:18
data for their customers so that
21:20
they can make them better offerings. And
21:22
there are actually technical ways also through
21:24
Snowflake to make it very easy to
21:26
share that data from one organization to
21:29
the other while maintaining all
21:31
the right levels of security so that you
21:33
don't need to duplicate data
21:35
and send it around. What today
21:37
still sometimes happens through email or
21:40
FTP so that you
21:42
can actually solve these use cases that are
21:44
not within the bounds of a single company,
21:46
but really within the bounds of the industry.
21:49
We're here at Distribute Tech where
21:51
there are like 17,000 people
21:54
here and hundreds of companies
21:56
and if you walk around the floor,
21:58
you suddenly see... AI slaps
22:00
on everything and like I'm sure
22:03
there's you know, some companies it's more
22:05
of a cynical play But like in reality a lot of
22:07
these companies are Starting to build
22:09
out AI offerings as an extension of what
22:11
they were already doing I mean when you
22:13
look at the tech platforms of
22:15
some of these third-party Service
22:17
providers, how much are you
22:19
seeing AI getting integrated into
22:21
their products and teams? I
22:24
think broadly AI is being
22:26
infused in all kinds of solutions for all
22:28
business units and I think that they will
22:30
only accelerate You see it
22:32
show up in surprising ways. Maybe so for
22:36
example utilities that might
22:38
have handwritten maintenance reports
22:40
for things like substations or lines
22:42
can now actually digitize all the
22:45
data using the power of llms
22:48
and now they have a queryable
22:50
so questionable database that they
22:52
might ask Which
22:54
what are the most common fault mechanisms
22:56
that we see in our substations and
22:58
which substations should we do preventative maintenance
23:00
on next? so Even
23:03
areas that historically seemed oh hi is
23:05
never going to get there are now
23:07
increasingly accessible to AI if
23:10
you think about Let's say the next
23:12
decade of advancement in AI,
23:14
which is a very long time I mean
23:16
we see such radical improvements in the technology in a
23:19
12 month time frame So a decade is quite long
23:21
But if you think about how it will play out
23:23
in the power sector where you see more conservative
23:26
adoption What will
23:28
hold it back and what will accelerate it? Yeah,
23:31
well one thing that could hold it
23:33
back is certain regulatory changes if those occur at
23:36
the at the you know National super national
23:38
level on what we can do with AI
23:41
One thing specific to energy that will hold it
23:43
back is what we've seen is that in the
23:45
last? 15 20 years while the amount
23:50
of compute and data storage has increased
23:52
a lot in some cases more than
23:54
10x depending on the geography and time
23:56
frame The energy consumption has actually
23:58
barely budged. It has barely And
24:01
that's mostly thanks to the increasing efficiency, the energy
24:03
efficiency of data centers. And
24:05
it looks like that trend may end. So
24:08
it looks like if we look at the coming
24:10
years, actually,
24:13
data center energy efficiency might not increase that much, while
24:16
our use of data centers is rapidly
24:18
growing, in part because of the
24:21
generative AI hype. And so I think one
24:23
thing that could hold it back is, can
24:25
we build enough data centers with access to
24:27
data centers? With access to
24:30
power, ideally zero carbon power, that
24:32
would be one thing that could hold it back. In
24:35
terms of acceleration, I think a
24:37
lot of that is already happening. And I
24:40
certainly am frequently
24:42
amazed by how quickly new
24:45
models, new
24:47
types of models are coming out and what
24:49
it can be, whether it's now most recently,
24:52
OpenAI's Sora model, in which
24:54
you can generate one-minute videos
24:57
with just a few lines of text to
25:00
all the things that we will see next. So
25:02
if we look at what could hold back AI, it's
25:05
chip availability and power infrastructure. And
25:07
there are a lot of questions
25:09
about what the power demands
25:12
of data centers look like as AI use
25:15
expands. And so that
25:17
has a lot of people hand-wringing about the
25:19
energy intensity of the data
25:21
center industry and whether it will cause us to run
25:23
in place and whether we'll need a lot more clean
25:25
resources just to make up
25:28
for AI computational infrastructure. But
25:30
then there are all these other great benefits that AI
25:33
can unlock for the grid
25:35
and even unlock for the benefit of data centers.
25:38
What do you think about the net impact of
25:40
AI in the energy space, both
25:43
as a energy consumer,
25:46
potentially an exponential energy consumer, and
25:48
as an unlock for clean resources? So
25:51
I think we might see an
25:53
increase in energy consumption from AI. I've
25:55
seen some utilities in their integrated resource
25:58
plans Mention... Wrote
26:00
have to x five x, even ten
26:02
x in the next decade. Driven primarily
26:04
by it's more data center bill that
26:06
first, I'm not sure if that's what
26:09
will happen. History tells us
26:11
that we are not always very good
26:13
at forecasting what's happening in the long
26:15
term, so I think maybe will not
26:18
be that much increase energy consumption. But
26:20
then the other thing is, I think
26:22
Ai has so many applications to drive
26:25
energy efficiency and reduce emissions. Whether it's
26:27
about how we manage global supply chains
26:29
a hobby routes planes and ships, or
26:32
whether it is on all the opportunities
26:34
within the entered system to run things
26:36
more efficiently. So. I generally
26:39
think that. We should
26:41
proceed with their technology innovations. That
26:43
proceedings. And in fact remain of
26:45
really have the full control for not doing
26:47
so and that. Their advances
26:50
that we will get in
26:52
emission reduction in energy consumption
26:54
reduction will far outweigh the
26:56
increased energy consumption of data
26:58
centers running. At. To
27:01
some policy. Thank you So much strength! Even a
27:03
pleasure to be heard. That
27:11
is it for Zone and Copy. The
27:13
production of Latitude Media is produced and
27:15
written by me and Sean. Mark one
27:18
is our technical director. He also mixes
27:20
the show and road are theme song.
27:22
Got a lot of to media.com to
27:24
get all of our stories there and
27:26
if you want to get some your
27:28
stories on the go subscribe to our
27:31
companion podcast the Latitude. You can find
27:33
that anywhere you game shows. We also
27:35
have transcripts of this podcast and catalyst
27:37
that Latitude Media So you can go
27:39
over there subscribe to our newsletter. You
27:41
get it all in your inbox. And
27:43
ah thanks to Prelude Ventures for being
27:46
a supporter of Lights Would media Prelude
27:48
that visionaries accelerating climate innovation that will
27:50
receive the global economy for the betterment
27:52
of people and the planet and you
27:54
from their more about their portfolio a
27:56
Prelude ventures.com And if you have feedback
27:58
on the show and you. The under
28:00
some ideas are just have thought you
28:02
want to share Send us a note
28:05
that editorial at Latitude media.com and we
28:07
will catch you next week and they
28:09
see this record and.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More