Latest / Radio Davos / Demand for electricity is soaring: how can our power systems keep up?

Transcript

Demand for electricity is soaring: how can our power systems keep up? · Radio Davos

0:01
There really is currently this moment for electricity.
0:05
Welcome to Radio Davos, the podcast from the World Economic Forum
0:08
that looks at the biggest challenges and how we might solve them.
0:11
This week the global demand for electricity is rising
0:14
fast, with wars in Ukraine and the Gulf squeezing fossil fuels.
0:18
And that then led to a pivot in the focus on electrification as an alternative
0:24
to using fossil fuels in the energy system.
0:28
And now the opportunity to use clean energy for electrification
0:32
is also more available than it was any time before.
0:36
Add to that the rising demand from artificial intelligence.
0:39
AI data centre globally grew by 50% in their energy consumption in just over a year.
0:44
By 2030, they expect demand from data centres to double.
0:48
That will still be just around 3% of total electricity consumption,
0:53
but the pace and scale of it is, is quite remarkable.
0:56
We look at the pressure on electricity systems
0:58
and speak to the head of a startup who says there is a solution to the data
1:02
centre challenge.
1:02
Think of the grid as a big, massive super highway.
1:05
It's got eight lanes, right? And most of the year, it's quite empty.
1:08
You could add a ninth lane to that highway.
1:10
It'll take you 10 years, and it's very expensive, and everybody pays for it.
1:14
Or that data centre can just reduce its consumption during that peak moment
1:18
so you can use your eight lane super highway,
1:20
which is available almost all of the year.
1:22
Follow Radio Davos wherever you get podcasts, or visit
1:25
wef.ch/podcasts.
1:26
I’m Robin Pomeroy at the World Economic Forum,
1:29
and asking how we can meet surging demand for electricity...
1:31
You've avoided building that expensive ninth lane.
1:34
This is Radio Davos
1:36
Welcome to Radio Davos, where we're talking about energy,
1:38
specifically the massive increase in demand for electricity
1:42
and how we might meet that demand. Later in the episode,
1:45
we'll hear from a tech startup that says it has a solution
1:49
to the challenge posed by the rise of AI data centres.
1:53
But first, to set out the scale of the issue and
1:56
to talk about the whole range of issues, I'm joined by my colleague
2:00
Divya Reddy, Manager of Energy Initiatives at the World Economic Forum.
2:03
Hi Divya, how are you? Good, thanks. How are you, Robin?
2:06
Very well. Thank you.
2:06
What do you do at the World Economic Forum?
2:08
I'm the manager specifically on the future of power systems initiative
2:12
and we bring together stakeholders
2:13
from across the electricity value chain to really decide
2:16
what are the key priorities facing the electricity
2:19
industry and arrive at some joint solutions for addressing them.
2:23
Electricity, the demand for electricity,
2:26
whenever we do a radio Davos about energy, at the moment
2:30
we can talk about oil and gas, we can talk about all kinds of things,
2:33
but this massive increase in demand for electricity,
2:36
when did that really start and what's driving it?
2:40
So it's been over the last 10 years, and it's a combination of factors,
2:44
but it really is currently this moment for electricity.
2:48
And part of the more recent driver for this
2:51
is the number of energy crises that we faced.
2:55
First, it was the Russia-Ukraine situation that resulted in gas supply shortages.
3:00
And then more recently,
3:02
the Straits of Hormuz being closed off, again, resulting in oil and gas shortages.
3:06
And that then led to a pivot in the focus on electrification
3:10
as an alternative to using fossil fuels in the energy system.
3:15
And now the opportunity to use clean energy for electrification
3:19
is also more available than it was any time before.
3:24
So it's partly that transition to electrification
3:29
for things like cars and in industry as well.
3:33
Exactly. So it's across the board.
3:35
So the three sort of end uses would be transportation,
3:39
it would be industry, and it would buildings as well.
3:41
Also though, add to that, in the last what, three years, the rise of
3:47
artificial intelligence as a thing we're all using now,
3:50
which probably we didn't realise three years ago,
3:52
the massive energy demand that would create.
3:54
Tell us something about that.
3:55
Yeah, exactly.
3:56
So in addition to just general electrification of end uses,
4:01
there is this massive new entry of data centres, specifically
4:05
AI data centres that is coming into the system.
4:09
Now, it's still a relatively small part of the overall electricity picture.
4:13
And it probably will remain so, but it's the pace and scale of the change
4:18
that's really having a dramatic impact over the last few years.
4:22
So just to put some numbers on this, the IEA estimates that last year alone,
4:27
there was a 17% increase in data centre electricity demand.
4:31
And that's compared to just 3% for overall electricity demand growth.
4:36
And by 2030, they expect demand from data centres to, to double
4:40
that will still be just around 3% of total electricity consumption,
4:45
but the pace and scale of it is, is quite remarkable.
4:48
And AI data centres in particular draw around
4:52
10% more electricity than even standard data centres.
4:56
And they also have a very kind of volatile and more
5:01
unpredictable demand profile than traditional loads do.
5:06
So it's sort of a combination of factors.
5:08
It's the pace and scale of growth.
5:11
It's also the nature of the demand
5:14
and it's also location because. What happens
5:17
is that when data centres come into the electricity system,
5:21
they typically connect on what we call firm connections,
5:25
which means that they basically are guaranteed service at any time.
5:31
And so that requires that
5:34
this network needs to accommodate them at all times and
5:38
that it can exacerbate peaks in supply and exacerbate tightness that then leads
5:43
to higher costs that then need to be paid by all consumers.
5:47
So there's also a locational issue if they're entering into the system
5:51
at congested areas.
5:53
Later in this episode, we're going to hear an interview
5:55
I did recently with the head of a startup company
5:59
who says he has
6:02
a solution, at least in part, to some of this,
6:07
and it does go to that thing you're saying about peak demand,
6:10
which is a problem with electricity in general, right?
6:12
Maybe we can talk a little bit about that.
6:14
We'll get back to data centres in a minute,
6:16
but in general, you know, we've got wind and solar energy,
6:19
which over the last, what, 10 years, a bit more
6:22
maybe, have suddenly become this very cheap way of generating electricity.
6:29
But what they don't have is that kind of constant,
6:32
you can't, if the sun's not shining and the wind's not blowing,
6:34
you can just switch it on and you need that baseline power.
6:37
That's a real challenge, isn't it, for this massive increase in demand.
6:42
Yeah, exactly.
6:43
So essentially, the way that electricity systems
6:46
have traditionally operated and been planned for
6:50
is that you have these steady sources of demand and steady sources of supply.
6:54
And it's sort of
6:55
been this traditional model of balancing the electricity system at all times.
7:00
With the advent of more variable sources of generation like wind and solar,
7:04
you have more volatility in the supply profile.
7:08
And now with the advent of things like data centres, you also have
7:12
this more volatile demand coming into the picture.
7:15
Now there are a lot of tools that can address these.
7:17
So there are now increasingly options for flexibility solutions
7:23
that can help address these volatile supply and demand profiles,
7:28
but they need to actually be enabled
7:30
to participate in the system the way that they are able to.
7:35
And where's the low-hanging fruit?
7:36
What are the things we can do quickly to improve our electricity systems?
7:42
So given the pace and scale of what
7:44
the electricity system needs and that it does need to grow,
7:48
we also do need to enable the building of new infrastructure.
7:52
But more immediately, there is a lot of infrastructure currently
7:57
that is underutilised.
7:58
And so the low hanging fruit really is to enable the optimisation
8:02
of existing infrastructure and assets that we already have.
8:07
And there are a lot of tools,
8:08
particularly flexibility tools, but there are also digital tools,
8:12
there are AI tools that can support this.
8:15
So those are the kinds of solutions that really are the low-hanging fruit.
8:18
Where can people get more information?
8:20
You publish useful reports.
8:23
Where can people go? What resources have you got?
8:26
So we currently have a tool on our website,
8:29
which is called the Innovation Playbook for Future Power Systems.
8:32
And it includes, it's a sort of database of over a hundred solutions
8:36
backed by case studies.
8:38
And it really is aggregating solutions
8:41
that are already in the market that then can be deployed
8:45
and scaled to address these challenges of modern power systems.
8:49
And it covers both technology type of solutions, but also things like
8:54
financial models, business solutions, policy and regulatory settings
8:58
that are all part of what we call solutions.
9:02
We also have an upcoming white paper that will be released in early November
9:06
and that is really looking at three focus areas that are kind of
9:11
most pressing for today's power systems.
9:15
The first is optimising existing infrastructure.
9:18
The second is how to quickly enable and finance new infrastructure.
9:23
And the third is how to connect large loads such as data centres.
9:28
Let's get back to data centres then.
9:31
Just remind us again what the problem is
9:34
here and of the work you're doing on that
9:37
and then we'll introduce this interview.
9:40
Yeah,
9:40
so the problem, again, is that the pace and scale of demand is greater
9:47
than the electricity system in many locations can accommodate.
9:52
And so it's creating this tension.
9:54
It's requiring network expansions.
9:57
It's increasing costs.
9:59
There are some solutions that can help alleviate this challenge.
10:03
One is from the demand side that data centres are actually able
10:08
to operate more flexibly. So they can compute at different periods,
10:13
they can shift jobs to different hours, they can cut their computing in certain
10:17
peak moments, and they can also redirect processing to a different location.
10:22
And all these have significant benefits to the system.
10:25
They can free up more network capacity.
10:28
They can lower costs or unlock savings because it avoids peak charges. It
10:33
avoids new infrastructure builds and it sort of optimises the existing grid
10:38
more efficiently.
10:39
And they can also connect new sources of demand faster.
10:42
Currently we hear a lot about connection queues
10:45
where large demand is just not able to connect to the existing grids.
10:49
So it alleviates all these types of problems.
10:51
And then on the supply side, they can also offer
10:54
their own onsite storage as well as onsite generation
10:57
that also limits the amount of electricity they draw from the grid.
11:01
So that brings us to this interview that I did at an event that I believe
11:06
you were involved in here at the World Economic Forum.
11:09
His name is Varun Sivaram.
11:11
What was he doing in Geneva?
11:13
The World Economic Forum
11:14
hosted an energy forum at our headquarters in Geneva in September.
11:18
Varun Sivaram served as one of the co-chairs for this event.
11:22
And what it did is it brought together stakeholders from across the
11:26
energy spectrum to discuss challenges with current energy systems,
11:31
to discuss joint solutions, and to kind of facilitate partnerships
11:37
across the energy space that can take forward these types of solutions.
11:42
He's a very interesting person.
11:44
He's the chief
11:45
executive officer of Emerald AI, and that's what he's going to talk about.
11:48
But if you look at his CV, he's had all kinds of interesting jobs
11:53
in industry and at the policy level in the United States, in India.
11:58
Let's get on,
11:59
let's hear this interview with Varun Sivaram.
12:01
I'm Varun Sivaram, CEO and founder of Emerald AI.
12:05
What is Emerald AI?
12:06
Emerald AI transforms AI data centres into flexible grid
12:11
allies, good citizens to communities and power grids alike.
12:16
With Emerald AI's technology, we can get far more
12:19
data centres connected to today's existing power grits.
12:22
We can keep rates low, energy bills low for communities,
12:26
and we can keep the grid stable and reliable.
12:29
It's a win-win-win, but the data centres have to become
12:32
good grid citizens, and that's what our technology enables them to do.
12:37
Let's remind everyone
12:38
of the challenge here. AI, generative AI,
12:42
four years ago most people weren't even aware it was a thing.
12:46
Now most people who have a computer are using it.
12:49
We weren't initially aware, some of us, of the energy requirements
12:53
of AI. Give us some idea of
12:56
how big a problem that is.
12:57
The energy consumption of AI data centres is massive. Now,
13:02
it hasn't always been massive.
13:04
The International Energy Agency reports that historically data centres
13:08
represented 1% or less of global electricity consumption.
13:12
Now, that figure is set to hit 3% by 2030.
13:16
Still doesn't sound like a big number,
13:17
but it's important to note the growth rate.
13:19
AI data centre globally grew by 50% in their energy consumption just over a year.
13:25
Nowhere is this more pronounced than in the United States, my country,
13:29
where the US accounts for 45% or so of global
13:33
data centre power demand, and data centres are growing rapidly.
13:37
And the big problem that we're running into in the US
13:40
is that we don't have enough capacity
13:42
on the electricity grid to plug in all the data centres we want.
13:46
Look, by 2030, the United States is gonna want
13:49
to plug-in some 70 gigawatts of data centres.
13:53
But it's only really going to be able to plug in half of that number,
13:57
and that leaves a large unmet demand in order to power innovation, life-saving
14:03
cures for diseases, and all manner of other economic productivity enhancements.
14:09
Data centres have hit the energy barrier in the U.S.
14:12
It's not true in China, where in China there's plenty of energy,
14:15
400 gigawatts of available spare peak capacity
14:20
in 2030. Their limit is chips.
14:23
But in the United States
14:24
and in most of the rest of the world,
14:26
it's energy that's gonna constrain the growth of AI.
14:28
So, what's the answer? How do you solve the problem?
14:31
That's why we founded Emerald AI.
14:33
Emerald today is a unicorn technology company.
14:35
We're proud to be a WEF technology pioneer.
14:38
And we're supported by 12 Fortune 500 global companies, such as Nvidia.
14:44
So Nvidia is not only
14:45
our partner and investor, they've set the reference architecture
14:49
for the power flexible good grid citizen data centre of the future.
14:54
And Emerald software powers this.
14:56
Our solution is, to make the data centre capable of reducing its power
15:00
use every so often during moments when the grid is stressed out.
15:04
If you can do that, you can actually unlock a hundred gigawatts of power
15:09
on today's electricity grid in the US and much more around the world.
15:14
That power is just not used today.
15:16
It sounds counterintuitive.
15:17
Let me explain why.
15:19
Think of the grid as a big, massive super highway.
15:22
It's got eight lanes, right?
15:23
And most of the year, it's quite empty.
15:26
There aren't very many cars on that road. But two times a month,
15:30
just two times month,
15:31
that highway faces peak rush hour, where the cars are clogging up that road,
15:36
that's like a hot summer day when everybody's
15:38
using their air conditioners and the grid is really stressed out.
15:41
But most of the year, the grid runs at 50% capacity utilisation.
15:45
Now imagine you want to connect a new data centre.
15:48
That data centre during those peak moments
15:51
can't be served by that grid that's facing peak rush hours.
15:55
And so what you could do is you could add a ninth lane to that highway.
15:58
It'll take you 10 years, and it's very expensive, and everybody pays for it.
16:01
Or that data centre can just reduce its consumption during that peak moment.
16:06
So you can use your eight lane super highway, which is available
16:09
almost all of the year, to power that flexible, friendly data centre.
16:13
And in so doing, the revenue that it generates
16:16
helps the local communities pay their energy bills.
16:19
You've avoided building that expensive ninth lane.
16:22
You've broadened revenue for the communities.
16:24
And the best thing is you also help them prevent blackouts
16:26
because when the grid is really stressed out, hot summer day in Arizona,
16:30
we showcased with our partners NVIDIA and Oracle and others
16:33
that that data centre is actually able to reduce its consumption.
16:37
And that's why we're really excited that we've announced this new leading
16:41
coalition and alliance of the world's leading companies
16:45
committed to flexible AI data centres that are good grid citizens.
16:49
We call it the AI Energy Management Alliance.
16:54
Walk us through how it works then.
16:57
It sounds fairly simple.
16:59
If you're a big consumer of electricity,
17:03
somehow you work out now
17:07
there's a peak demand from everyone else.
17:10
I reduce my consumption.
17:12
Is it as simple as that?
17:13
That's my first question. I'll ask you a second one.
17:16
Does that mean then those data centres will be performing at less,
17:22
doing things less
17:24
quickly, will users of those AIs realise, oh okay,
17:28
it's a hot day in Arizona,
17:29
that's why chat GPT's taking so long to reply.
17:33
How is it that a data centre can moderate its electricity
17:36
use in the moment during those peak rush hour hours, you know,
17:39
three hours during the afternoon in Phoenix on a hot summer day?
17:42
Or here's another example, Robin, that you might appreciate.
17:44
Turns out that at halftime of the soccer game, the football game in England,
17:49
everybody goes into their kitchen and turns on their tea kettle.
17:52
And the tea kettle
17:53
spike puts a gigawatt of load onto the grid.
17:56
That's a moment of peak demand on the UK grid.
17:59
So in both of these cases, in Phoenix on a hot day. In the UK,
18:03
at halftime of the soccer, Emerald has worked with
18:06
real commercial data centres to reduce their power consumption
18:09
in those moments. How does it work?
18:11
There's a few ways you can do this.
18:13
One, you can slow down some of the computing that's happening on site.
18:16
There may be many AI jobs that are running.
18:19
Some of them can be slowed down without the customer really noticing or minding.
18:23
These could be a fine-tuning run, for example, of an AI model
18:27
where that run doesn't need to complete right this moment,
18:29
and you can delay it by half an hour or two hours.
18:32
Another example is batch inference process or a range of other batch processes.
18:36
And finally,
18:37
there are things that you can move from one location to another location.
18:41
Let's say
18:42
you're talking to a chat bot, chat GPT and you want your answer right now.
18:45
Well, I can run it equally well in Virginia or Chicago.
18:48
So on the coldest day of the year, we showcased that when Virginia's grid
18:52
had a peak demand moment, everybody turned on their heaters,
18:55
we moved the load over to Chicago, which as it turned
18:58
out was better equipped to handle those cold winter days.
19:01
Now, using this technology, Emerald's conductor software platform,
19:06
we can decide how best to protect the AI workloads that need to keep running right
19:10
now, those critical workloads, how to slow down other ones,
19:13
and how to move the final workloads to other locations,
19:16
all in a way that the AI customers often don't even notice
19:20
while the grid gets the precise response it needs.
19:23
Hot summer day,
19:24
the grid says, I need help, and the data centre responds precisely
19:28
based on what the grid needed,
19:29
let's say a 25% reduction in power for those three hours.
19:33
So that's how it works, Robin.
19:35
There's been a lot, particularly in the US, a lot
19:38
of local opposition to data centres.
19:42
If you've solved the problem, no one should be worrying about this anymore.
19:44
I wonder, are there blockages still in the way of making this happen?
19:48
Where's the future? Yeah, absolutely.
19:51
We need three things to happen in order for us to solve this problem.
19:55
First, we need the proof.
19:58
Second, we the coalition.
20:01
And third, we we need compact, or the bargain.
20:05
Let me walk through each of these points.
20:06
First, we need the Proof.
20:09
Emerald with NVIDIA and our other partners like EPRI
20:11
have already done these six demonstrations around the world.
20:15
London. Iceland just happened.
20:17
Phoenix, Virginia, Chicago, Portland.
20:20
So we've showcased at real data centres that you can do this.
20:23
And at Davos, we're very excited that we will announce the world's first
20:28
100 megawatt, multi-billion dollar power-flexible AI data centre.
20:32
NVIDIA and Digital Realty are building it,
20:34
and Emerald is their partner in making it a flexible data centre that can listen
20:38
to the grid and react in a way that the grid needs.
20:42
So the proof is seeing real data centre's do this, and we are well on our way.
20:46
That's point number one.
20:48
Point number two is the coalition, or the alliance.
20:51
I'm so proud that Google, NVIDIA, and Emerald
20:54
AI founded the AI Energy Management Alliance.
20:58
We have 20 launch partners, Anthropic, for example,
21:02
Analogue Devices, another Fortune 500 semiconductor company.
21:04
Or on the energy side, Constellation, NRG, AES, Fluence,
21:09
a range of large publicly traded Fortune 500 companies.
21:12
And with this cohort, this coalition, we're going to policymakers
21:17
at the American federal level, at the state level,
21:20
and eventually we'll make this global
21:21
with the collaboration of the World Economic Forum, to say
21:25
there is a technology that if you bring it onto your grid faster
21:29
will bring great benefits.
21:31
And that brings me to my third point,
21:32
the compact or the regulatory bargain.
21:35
What's the grand bargain here?
21:36
The grand bargain is if you have a data centre willing to be flexible,
21:41
you should reward it.
21:42
The reward is faster and larger connections to the power grid,
21:46
because that's a friendly data centre. That's a good grid citizen.
21:48
It's going to lower your rates.
21:49
It's going to bring on more clean energy and stabilise the grid.
21:52
It's not a data centre
21:53
that's going to cause the problems that communities are correctly worrying about.
21:57
They're worried that their electricity rates are increasing.
22:00
They're worry that these data centres could destabilise their grid.
22:04
If you want communities to support a data centre,
22:07
that data centre needs to demonstrate that it's being a good citizen.
22:11
In a way that they may not traditionally have wanted to do, right?
22:14
Traditionally, data
22:15
centres have wanted to say, give me 24-7 access to power.
22:18
I never want to think about the power grid.
22:20
Being a good grid citizen means you have to think about power grid,
22:23
you have to be responsive. When your neighbours need that power,
22:27
if you're willing to step forward and say,
22:29
okay, I'll take a sacrifice in this moment,
22:32
that's the kind of good citizenship that should earn you a fast lane onto the grid.
22:36
When we talk to our partners, the Utility National Grid, our investor
22:39
and partner in the United Kingdom. Their chief strategy officer,
22:42
when he heard about Emerald, he said, my goodness, if data centres could do this,
22:46
I wouldn't be worried about connecting them
22:49
and I wouldn't be making them wait years on end to connect them.
22:52
I'd want to bend over backwards to connect to them as quickly as possible
22:56
because they could make my system work better.
22:57
I would compete with other utilities to connect with them faster.
23:01
That's what I want,
23:01
a virtuous race to the top where communities and utilities compete
23:05
with each other to bring the friendly, good grid citizen data centres.
23:10
And that virtuous cycle means that communities benefit truly
23:13
when the data centre enters and enables them to bring on all kinds of other
23:17
loads, like hospitals and advanced manufacturing. That's the virtuous cycle.
23:21
So when you're talking to someone like the National Grid,
23:24
and they're very excited about it, is that also because not only
23:28
are you able to say we can put on a data centre and it won't
23:32
just drag everyone else's electricity away from them and put up prices,
23:37
but there's also some kind of stabilising mechanism because all power grids,
23:43
they'll have inputs.
23:44
There's suddenly all the wind turbines are going faster than we can use
23:48
that electricity. Is there that balancing thing happening?
23:50
Yeah, that's such a good question, Robin,
23:52
and there are three important reasons,
23:54
I'm sorry, I used to be a consultant, I talk in threes.
23:56
There are three reasons why utilities absolutely love
24:00
these flexible data centres and why their regulators and their political leaders,
24:04
they love these data centres.
24:05
The first is the one you said, Robin. If a flexible data centre
24:09
comes onto the grid and it can respond to whatever the grid needs
24:12
or when it's in stress, it can stabilise the system and avoid a blackout.
24:16
That's important
24:17
whether the tea kettles are going on at halftime in the UK.
24:20
Or it's a hot summer day or lightning has struck and taken out of substation,
24:24
the data centre can respond and help to bring the grid back into balance.
24:28
So reliability and energy security is one.
24:30
The second reason is because data centres that are flexible help you avoid
24:36
building out very expensive infrastructure to meet that rare peak.
24:40
They help you avoiding building that ninth lane on the eight lane superhighway
24:43
that you basically will never use except for two times a month on rush hour,
24:46
right? Now, to be clear, utilities do like to build stuff.
24:50
They get paid when they build stuff,
24:52
but the good news
24:53
now is that the data centres come in and say, I'm willing to be flexible.
24:56
I'll use your grid when no one else is using it
24:58
during the rest of the year when the eight lane
25:00
superhighway is pretty empty. And the money that I bring in,
25:04
you can now use to go build out infrastructure to serve even more load.
25:08
You can bring the manufacturing plant, the factory, you can bring that hospital,
25:12
you can serve your neighbourhoods and everybody can put air conditioners
25:16
in their homes because there are heat waves in the UK, et cetera, et cetera.
25:19
It's the data centres that are helping you fund this
25:22
instead of the local community's energy bills.
25:24
So your grandma's bill doesn't go up,
25:26
but the data centre comes in and helps you to fund all of these upgrades.
25:30
So that's the second reason.
25:31
And the third reason is that it avoids the utility being disintermediated.
25:36
That's a long and painful word. What does that mean?
25:38
The utility likes to have a relationship with its customer, right?
25:41
It likes its customer to be part of the public power grid.
25:45
The public power
25:45
grid works because so many different diverse customers are all on it
25:49
and we can socialise the costs of that infrastructure.
25:52
What's about to happen though, Robin? Is
25:54
data centres are going to say, you know what, I'm fed up with this.
25:56
I don't wanna wait 12 years to connect to this grid.
25:58
I'm going off grid.
25:59
I'm building my own little utility by myself in the middle of Texas,
26:02
or I'm actually going into space.
26:04
Did you hear
26:04
Elon Musk has a $2 trillion company called SpaceX that can put me in space?
26:08
Let me do that.
26:10
If that happens, utilities in the public power grid will shrivel,
26:13
because American, European, and global GDP
26:17
will move toward AI and AI productivity.
26:21
And if the data centres are not on the public power grid
26:23
and their revenues aren't supporting local communities,
26:25
those local communities will be left with larger and larger bills.
26:29
Utilities don't want somebody else to power those data centres.
26:32
They'd like to power them themselves. And they can do it
26:35
if the datacenters are flexible, friendly, community positive data centres,
26:39
that's why we wanna make these good grid citizen data centres a reality.
26:44
Let's be clear, then. A data centre company,
26:48
a hyperscaler, could go to a town somewhere and say, right, we're about to
26:55
put spades in the ground.
26:56
We're going to build this massive thing here.
26:59
And I know what you're going say,
27:01
everyone. You're worried about the things we're talking about.
27:04
You no longer need to worry.
27:06
You'll still get the electricity you need.
27:08
Your prices won't go up.
27:09
Can they already say that today?
27:11
They can't say it unless they do this.
27:13
Unless they are flexible,
27:15
they can't in good conscience go to a community and say I'm not going
27:18
to raise your power bills.
27:20
Even if they say I am
27:21
going to put spades down, I'm going to build this data centre,
27:23
and I am not even going to connect to your grid,
27:25
I am just going to a big gas plant next to my data centre and power it,
27:29
there are various knock-on effects.
27:30
Suddenly you've raised the cost of gas turbine
27:32
components for the public power grid
27:33
or the cost to electrical switch gear implicitly.
27:36
Suddenly, there's a chance that the emissions,
27:38
the noxious fumes from your gas plant near this community are going to cause
27:42
air pollution. Unless you are willing to connect to that public-powered grid
27:46
and be flexible and respond to the community's needs,
27:50
you can't in good conscience say, I'm going to make sure that your rates
27:54
aren't increased by my data centre coming into your community. And
27:59
to your
27:59
point, Robin, this is just a conversation
28:02
that's going to happen not just in the United States.
28:04
It's going happen all over the world.
28:06
And here's what I want to happen.
28:07
What I want to happen is the data centre says, look,
28:10
I want to be a good grid citizen. I want be a citizen.
28:12
I'm going to come to town.
28:14
I'll pay my property taxes.
28:15
I'm even going to fund some improvements to parks and schools.
28:18
I'm gonna be a flexible power consumer
28:20
that will help your bills and it will help your reliability.
28:23
And the community says, well,
28:26
we would love this because it brings lots of benefits.
28:29
We can also fund the manufacturing, the hospitals, et cetera.
28:32
But there's still some of us who are worried about AI
28:34
generally. They're worried AI might take your job.
28:36
They're worried that. AI might cause safety risks,
28:38
they're
28:39
worried a bioweapon might come out of an AI lab one day.
28:42
And then they'll weigh their risks, their pros and cons,
28:45
and they'll say, well, this particular asset
28:47
could be really positive to my community, and I have to weigh that against
28:52
what my general feelings are toward AI more broadly.
28:57
And I hope communities around the world will even-handedly
29:00
assess this and determine whether they want
29:03
to invite the data centre.
29:04
But where I don't want the argument to go is, Well, I should just reject this data
29:08
centre out of hand
29:09
because it's going to raise my power bills and it's not going to be a good citizen.
29:13
I think there's so much room now
29:14
for the data centres to change that narrative and to prove,
29:18
in fact, quantitatively, I will be an excellent citizen
29:20
and you will have no economic opportunity cost.
29:23
In fact, you will be strictly better off.
29:25
If I come to your town,
29:26
everything about your economy will be better if I come town than if I don't.
29:30
Tell us about this alliance you mentioned earlier with these big companies.
29:35
Why have you formed an alliance?
29:36
What's the difference between just serving them as customers?
29:39
Exactly.
29:40
Well, so if you remember, the three components that I'm passionate
29:43
about, first is proving that this is real, that data centres can be flexible.
29:48
Second is bringing the Alliance together.
29:49
And third is getting the regulatory grand bargain.
29:53
The Alliance,
29:53
that second component, I think is critical for that bargain to happen.
29:57
The alliance brings together leading players from across the
30:00
AI and energy value chain.
30:02
As I mentioned, you have leaders. We are the founding members, Emerald, Google, and NVIDIA, but also leaders
30:06
like Anthropic, one of the world's leading AI companies, analogue devices,
30:10
but also on the energy side, leading energy producers and utilities,
30:14
National Grid I mentioned earlier, NRG, Constellation, etc.
30:18
And they've all come together and they've said, you know what,
30:21
we're going to present a unified voice.
30:23
Because governors, regulators, utilities, they may not know this is possible.
30:27
They may not know about this grand bargain.
30:29
So we'll tell them, here's the grand bargain,
30:31
if the data centre's willing to do these things,
30:34
modulate your power use during rush hour.
30:36
And we don't care how you do it. Right?
30:37
You might decide you've got a battery on site
30:40
and that's how you're going to run the battery during peak
30:43
grid rush hour.
30:45
You'll run the battery
30:46
or you'll slow down your compute, which is what Emerald helps you do.
30:49
However you do this, we don't mind.
30:52
We will just give you faster and larger power access if you can
30:56
enforceably and verifiably be a flexible power user.
31:00
That grand bargain sounds like a very wonky technical thing,
31:04
Robin.
31:04
You know, we're talking about ramping up and down your megawatts.
31:08
It's very wonky, but it couldn't be more central
31:11
to the social compact of AI and communities.
31:14
If data centres can do this one party trick, control their power,
31:19
data centres can be net positive for communities,
31:22
and the growth of AI no longer needs to see a trade-off,
31:25
Robin, between AI innovation, energy affordability, and energy reliability.
31:31
It's that rare case, and I'm a former policymaker.
31:34
It's the rare case. Where you've got a win, win, win with basically no trade-off.
31:38
What about the timing of this argument?
31:41
Because maybe a year ago, it wasn't such a hot issue,
31:44
data centres. And now, I've certainly seen in US media, it's a very big issue.
31:49
And you'll know very well, once people have made up their mind
31:52
about something, it's going to be very hard to change.
31:54
So if you go into a town and say, yeah,
31:56
we're going to build this huge data centre. Don't worry. It's fine.
31:58
That problem you thought was a problem, we fixed it now.
32:02
Either
32:03
your timing is amazing and it's perfect, because this is a problem
32:06
so many people would like to solve. Or you've kind of missed it,
32:09
and already the public is against these awful data centres.
32:13
Which do you think it is? Robin,
32:15
the point you're making here is that
32:16
we're trying to close the barn door after the horse has bolted,
32:19
right?
32:20
And let me tell you,
32:20
I've got a lot of faith in the American people that present them
32:24
with new facts and new data, in this case, the opportunity
32:28
to have economic development and prosperity and AI innovation alongside
32:32
protecting their bills and protecting their energy reliability and security.
32:37
I think a lot communities will take you up on that.
32:39
And so you're right. It is a polarised political topic today.
32:43
But just, just you wait.
32:45
We're going to get to January 2027 to Davos,
32:49
and you'll see the announcement that Jensen Huang at NVIDIA
32:52
has already previewed that NVIDIA, Emerald AI, and Digital Realty are launching
32:58
this world's first power-flexible AI data centre, or AI factory.
33:03
You know, it's important, of course, because it's 100 megawatts,
33:05
that's meaningful. But it's really important
33:08
because that's the blueprint.
33:10
That's the first time you're going to see a purpose-built AI data centre
33:14
that can perfectly respond to the local grids.
33:17
Dominion Energy is the utility.
33:18
PJM is the grid operator.
33:20
They're all our partners with EPRI.
33:22
That's the first time you'll see one built to the specification,
33:25
and that specification is then public.
33:27
And the next thousand AI factories can be built to that DSX reference architecture.
33:33
That's what I'm really excited about, because we have real proof.
33:37
And yes, we will now need to take this door to door
33:40
and convince communities that this is reality
33:43
and that they're priors, which are reasonable, right?
33:46
The American people have had reasonable reactions.
33:49
Their priors are that data centres might not be responsive, flexible entities.
33:54
We've gotta change those priors with proof and with data.
33:57
I have a lot of faith in the American people.
34:00
That was Varun Sivaram,
34:02
the CEO of Emerald AI, talking to me a few weeks ago
34:07
at this Energy Forum, which Divya, you were involved in organising here.
34:12
If people want to find out more about your work,
34:14
remind us where they can go and what was that
34:17
report that's coming out in a couple of weeks' time from now?
34:20
So everything about our initiative
34:22
can be found on the World Economic Forum's website.
34:25
If you look under the Future Power Systems Initiative,
34:27
you can find the Innovation Playbook for Future Power System there.
34:31
The upcoming white paper on enabling the electrification
34:34
era will also be published there
34:36
in the first week of November.
34:38
Have you got a busy Davos ahead?
34:40
We're already thinking about the Annual Meeting in January, 2027.
34:44
Indeed. And given the attention
34:48
that electricity is facing right now,
34:51
they're discussing at COP to set a global electrification target.
34:55
So I think the topic of electricity will be very front and centre.
35:00
The climate conference that's happening in Turkey this year.
35:03
So heading into Davos, I think the topic of electricity will definitely
35:06
be front and centre and we will be having a number of sessions
35:10
that really look at different aspects of it.
35:13
So there's loads on that, on your part of the world Economic Forum website.
35:18
Radio Davos has done many episodes about energy and electricity.
35:23
I'll be putting links to those in the show notes. Divya Reddy, you're
35:27
manager of energy initiatives. Thanks very much for joining us on Radio Davos.
35:31
Thank you, it was very nice to be here.
35:33
Radio Davos is available every week wherever you get podcasts,
35:36
or you can find it at
35:39
wef.ch/podcasts, where
35:40
you can our other two weekly podcasts,
35:43
Meet the Leader,
35:44
conversations with some of the most interesting
35:46
people in the world about how they lead their organisations, and Agenda Dialogues,
35:50
which is the full audio from sessions recorded here at the World Economic Forum.
35:56
Radio Davos will be back next week.
35:58
Thanks very much to you for listening, and goodbye.