November 26, 2024 • 5 mins
What if the race for AI supremacy is not just about algorithms and data, but also about the massive energy demands reshaping our power infrastructure? Join us as we explore the critical intersections of AI-driven data center loads and the immense challenges faced by distribution utilities that are compelled to build and own their infrastructure. From the complexities of wholesale power markets to the sustainability of current business models, we unravel the urgent need for new generation assets. Discover whether innovations in cooling technologies and chip efficiencies can transform the industry and what this means for the future of power consumption.
Our discussion also brings to light the fascinating dynamics of the AEP Ohio rate case, an emblematic example of the high-stakes negotiations between utilities and data centers. With key players like Walmart influencing tariff proposals, learn how these negotiations could reshape infrastructure costs and the implications for ratepayers. We'll guide you through the intricacies of 10-year contracts, minimum demand charges, and the evolution of transmission technologies. This episode promises a compelling look at the future of utilities in a rapidly advancing AI-driven world, where technology, energy, and economics collide in unexpected ways.
Our discussion also brings to light the fascinating dynamics of the AEP Ohio rate case, an emblematic example of the high-stakes negotiations between utilities and data centers. With key players like Walmart influencing tariff proposals, learn how these negotiations could reshape infrastructure costs and the implications for ratepayers. We'll guide you through the intricacies of 10-year contracts, minimum demand charges, and the evolution of transmission technologies. This episode promises a compelling look at the future of utilities in a rapidly advancing AI-driven world, where technology, energy, and economics collide in unexpected ways.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Hi, in the past four sessions we've reviewed
AI-driven data central load,we've had a limited look at key
value propositions for AI andwe've discussed growth
projections, issues related totraining, the large language
models, the LLMs, and thepossible impacts on wholesale
power prices.
In this section, we'll look atthe impacts on distribution
(00:20):
utilities and, most especially,where they build and own their
own infrastructure.
Distribution utilities and,most especially, where they
build and own their owninfrastructure.
Wholesale bulk power marketsare one thing.
There, power prices may beaffected by supply and demand
balances, but if somebody buildsa generating asset and loses,
that loss is borne byshareholders With vertically
integrated utilities, whetherthey own generation or
transmission.
The financial impact of theirdecisions is passed on to rate
(00:44):
payers.
If you want to know more abouthow something like that feels,
talk to the folks who pay billsat Georgia Power.
Now that the Alvin Vottle nukesare online, sort of Unit 3 seems
to be having some trouble there.
So let's summarize thechallenge with respect to
utilities and these huge datacenter loads.
First, they are huge.
Many of them are in the manyhundred megawatt and even
(01:07):
gigawatt scale.
Second, they want the power now.
Time to power is a criticalthing in the race for AI
supremacy.
Third, that means utilities arerapidly being asked to build
supply and supportinginfrastructure such as
transmission and transformers.
Fourth, there's a great deal ofuncertainty as to where and how
AI will actually makeprofitable business and whether
(01:29):
the LLMs can keep growing at therate they are currently doing.
And fifth, there's also a greatdeal of uncertainty as to
whether they'll continue to keepconsuming electricity at
current rates.
As one example, coolingrepresents somewhere around 40
to 50 percent of total load, butthat's using air and fans,
which is a lousy way to dealwith waste heat.
(01:50):
New liquid immersiontechnologies that involve
putting servers in dielectricfluids, as one example, can
dramatically reduce coolingloads by up to 95 percent.
Then there are the chipefficiencies themselves.
Ibm in September announced abreakthrough in chip efficiency,
and market leader NVIDIAcontinues to make gains here as
well.
So the natural question If theymake gains, won't this just be
(02:13):
another example of Jeevan'sparadox, in which they'll just
use these gains to attain morecompute power?
Perhaps, but at some pointthere will be finite limits.
Thus, utilities are faced with arapidly growing industry whose
value propositions are all butcertain, with a request to build
tens of thousands of megawattsof new generation assets and
(02:34):
infrastructure that may not lineup temporally with the load
being served.
Gas plants, for example, mayhave 30 to 40 lifespans over
which they're amortized.
The crux of the matter, then Ifbusiness models change and the
demand for those supply andinfrastructure assets evaporates
, you, the rate payer, get topay for it.
This dynamic has been playingout in the AEP Ohio rate case
(02:57):
that involves only transmissionand not any new generation.
The utility served 600megawatts of data load, with
agreements to supply anadditional 4,400 megawatts
through 2030.
Transmission in their systemwas sufficient for that quantity
, but then in May of this yearanother 30,000 megawatts of
requests showed up that wouldrequire new lines, possibly the
(03:17):
big 765KV ones.
So AEP put a moratorium on newsupply and came back with a
proposal to the data centersCommit to 10-year contracts with
an out clause and a fee afterfive years and to paying us a
minimum demand charge set at 90%of contract capacity.
This was an increase of 50%over the current 60% tariff
(03:41):
requirement for large loads thatwere not data centers.
In early October, the datacenters and some suppliers, such
as Constellation, came backwith a counter quote settlement
requesting AEP to provide proofof transmission constraints and
asking for application ofimproved transmission
technologies, so-called GETs,and arguing for a lower fee of
75%.
(04:01):
Aep and other parties,including Walmart, subsequently
filed a tariff proposal with theUtilities Commission requiring
large new data centers to pay aminimum of 85% of the energy
they expect to use each month tocover infrastructure costs.
The data centers also have toshow they're financially viable
and could pay an exit fee ifprojects are canceled or
(04:23):
companies cannot meetcontractual obligations.
The terms would include afour-year ramp during
construction and be in effectfor a total of 12 years.
And why Walmart?
Well, probably because theyshare the same concern that was
voiced in a similar utility anddata center hearing on October
in Indiana that data center loadcould have the effect of
crowding out all other economicdevelopment while increasing
(04:46):
rates.
The dramas playing out in Ohioand Indiana are a show that will
soon be coming to a theaterclose to many of you, so it's
worth seeing what happens.
And how big might these rateincreases be?
Well, nobody knows for sure.
There are still too manyvariables and uncertainties
affecting this equation today,but a study out this week from
the Jack Kemp Institute suggestsan increase of as much as 70%
(05:09):
in rates over the next decade.
Again, nobody really knows.
What we do know is that thegrid has never before seen as
rapid an increase in demand asthe one expected here in the US
and globally, with the capitalinvolved and the risk to
ratepayers involved.
Utility regulators are going tohave to be informed in ways
they never have had to be in thepast, and they will likely have
(05:30):
to institute safeguards theywould have not had to in the
past either.
Tens of billions of dollars ofratepayer money is at stake.
Well, thanks for watching andlistening.
If you like this and you thinkI may be able to help educate
you and your team on this orjust about anything else related
to the electron, please justreach out to me on LinkedIn,
take care.
Hi, in the past four sessions we've reviewed
AI-driven data central load,we've had a limited look at key
value propositions for AI andwe've discussed growth
projections, issues related totraining, the large language
models, the LLMs, and thepossible impacts on wholesale
power prices.
In this section, we'll look atthe impacts on distribution
(00:20):
utilities and, most especially,where they build and own their
own infrastructure.
Distribution utilities and,most especially, where they
build and own their owninfrastructure.
Wholesale bulk power marketsare one thing.
There, power prices may beaffected by supply and demand
balances, but if somebody buildsa generating asset and loses,
that loss is borne byshareholders With vertically
integrated utilities, whetherthey own generation or
transmission.
The financial impact of theirdecisions is passed on to rate
(00:44):
payers.
If you want to know more abouthow something like that feels,
talk to the folks who pay billsat Georgia Power.
Now that the Alvin Vottle nukesare online, sort of Unit 3 seems
to be having some trouble there.
So let's summarize thechallenge with respect to
utilities and these huge datacenter loads.
First, they are huge.
Many of them are in the manyhundred megawatt and even
(01:07):
gigawatt scale.
Second, they want the power now.
Time to power is a criticalthing in the race for AI
supremacy.
Third, that means utilities arerapidly being asked to build
supply and supportinginfrastructure such as
transmission and transformers.
Fourth, there's a great deal ofuncertainty as to where and how
AI will actually makeprofitable business and whether
(01:29):
the LLMs can keep growing at therate they are currently doing.
And fifth, there's also a greatdeal of uncertainty as to
whether they'll continue to keepconsuming electricity at
current rates.
As one example, coolingrepresents somewhere around 40
to 50 percent of total load, butthat's using air and fans,
which is a lousy way to dealwith waste heat.
(01:50):
New liquid immersiontechnologies that involve
putting servers in dielectricfluids, as one example, can
dramatically reduce coolingloads by up to 95 percent.
Then there are the chipefficiencies themselves.
Ibm in September announced abreakthrough in chip efficiency,
and market leader NVIDIAcontinues to make gains here as
well.
So the natural question If theymake gains, won't this just be
(02:13):
another example of Jeevan'sparadox, in which they'll just
use these gains to attain morecompute power?
Perhaps, but at some pointthere will be finite limits.
Thus, utilities are faced with arapidly growing industry whose
value propositions are all butcertain, with a request to build
tens of thousands of megawattsof new generation assets and
(02:34):
infrastructure that may not lineup temporally with the load
being served.
Gas plants, for example, mayhave 30 to 40 lifespans over
which they're amortized.
The crux of the matter, then Ifbusiness models change and the
demand for those supply andinfrastructure assets evaporates
, you, the rate payer, get topay for it.
This dynamic has been playingout in the AEP Ohio rate case
(02:57):
that involves only transmissionand not any new generation.
The utility served 600megawatts of data load, with
agreements to supply anadditional 4,400 megawatts
through 2030.
Transmission in their systemwas sufficient for that quantity
, but then in May of this yearanother 30,000 megawatts of
requests showed up that wouldrequire new lines, possibly the
(03:17):
big 765KV ones.
So AEP put a moratorium on newsupply and came back with a
proposal to the data centersCommit to 10-year contracts with
an out clause and a fee afterfive years and to paying us a
minimum demand charge set at 90%of contract capacity.
This was an increase of 50%over the current 60% tariff
(03:41):
requirement for large loads thatwere not data centers.
In early October, the datacenters and some suppliers, such
as Constellation, came backwith a counter quote settlement
requesting AEP to provide proofof transmission constraints and
asking for application ofimproved transmission
technologies, so-called GETs,and arguing for a lower fee of
75%.
(04:01):
Aep and other parties,including Walmart, subsequently
filed a tariff proposal with theUtilities Commission requiring
large new data centers to pay aminimum of 85% of the energy
they expect to use each month tocover infrastructure costs.
The data centers also have toshow they're financially viable
and could pay an exit fee ifprojects are canceled or
(04:23):
companies cannot meetcontractual obligations.
The terms would include afour-year ramp during
construction and be in effectfor a total of 12 years.
And why Walmart?
Well, probably because theyshare the same concern that was
voiced in a similar utility anddata center hearing on October
in Indiana that data center loadcould have the effect of
crowding out all other economicdevelopment while increasing
(04:46):
rates.
The dramas playing out in Ohioand Indiana are a show that will
soon be coming to a theaterclose to many of you, so it's
worth seeing what happens.
And how big might these rateincreases be?
Well, nobody knows for sure.
There are still too manyvariables and uncertainties
affecting this equation today,but a study out this week from
the Jack Kemp Institute suggestsan increase of as much as 70%
(05:09):
in rates over the next decade.
Again, nobody really knows.
What we do know is that thegrid has never before seen as
rapid an increase in demand asthe one expected here in the US
and globally, with the capitalinvolved and the risk to
ratepayers involved.
Utility regulators are going tohave to be informed in ways
they never have had to be in thepast, and they will likely have
(05:30):
to institute safeguards theywould have not had to in the
past either.
Tens of billions of dollars ofratepayer money is at stake.
Well, thanks for watching andlistening.
If you like this and you thinkI may be able to help educate
you and your team on this orjust about anything else related
to the electron, please justreach out to me on LinkedIn,
take care.
Popular Podcasts
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.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.