Latest / UBS Trending / AI needs more power

Transcript

AI needs more power · UBS Trending

• 0:00
  For many years, electricity demand barely grew.
• 0:03
  Now AI data centers and electrification are changing everything.
• 0:07
  Can America's power grid keep up?
• 0:10
  We're going to talk about that
• 0:11
  in a whole lot more right now on UBS Trending.
• 0:25
  Hi everybody and welcome to the UBS Trending I'm Anthony Pastore.
• 0:28
  Thanks so much for joining us.
• 0:30
  Join me in welcoming back to the studio Paul Baiocchi from SS&C ALPS
• 0:34
  Paul, good to see you. Good to have you back.
• 0:36
  Always good to be here.
• 0:36
  Yeah. Same same.
• 0:38
  You know, this conversation about electricity demand, AI
• 0:41
  and data centers
• 0:42
  is one of the most common conversations that we're having.
• 0:44
  I'm sure it's what you're having because there are so many
• 0:48
  first of all, questions.
• 0:49
  Second of all, concerns.
• 0:51
  Third of all data sets
• 0:53
  that need to be communicated and maybe some misconceptions,
• 0:57
  you know, kind of,
• 0:58
  I don't know what the word you want to say is.
• 1:00
  It's kind of like
• 1:01
  taking some of the misconceptions out of what this all means.
• 1:04
  But I really want to start with you, because we know that AI
• 1:07
  it's been putting a lot of demand on the grid, and electricity
• 1:11
  demand continues to increase.
• 1:12
  And some of the expectations are there.
• 1:14
  We're going to see a massive increase
• 1:16
  between now in 2030 and demand.
• 1:18
  Talk about what you are talking about with that story
• 1:22
  and even some of the invest ability opportunities
• 1:25
  that are out there for people who are saying, hey,
• 1:26
  look, I've got to get on the train here.
• 1:28
  Yeah, it's important to level set
• 1:30
  because if you go back over the course of the past
• 1:33
  30 or so years,
• 1:35
  even further than that, electricity demand growth in
• 1:37
  the United States is pretty stagnant.
• 1:38
  We grew at about half a percent or point
• 1:40
  4% annually over that period of time.
• 1:43
  And now the expectations are that toward the end of the decade,
• 1:47
  we're going to be growing by, say,
• 1:49
  4% compounded annual growth rate, 7% compound annual growth rate.
• 1:53
  In fact, we talk about this theme a lot, and a lot of times
• 1:57
  we'll put a slide deck together
• 1:59
  and then the data becomes outdated.
• 2:01
  By the time we present it,
• 2:02
  because it's being updated in real time.
• 2:04
  The new CapEx figures, the new commitments.
• 2:07
  And so in many ways, at a high level,
• 2:10
  it's about this idea that we're coming off a period of stagnation
• 2:14
  on a grid that was built 100 years ago
• 2:17
  and isn't designed necessarily to accommodate the technologies
• 2:21
  that we're implementing,
• 2:22
  or in many ways, the type of load growth that we're expecting.
• 2:25
  And so it's become a hot button issue,
• 2:28
  not just from a market perspective, but also politically,
• 2:31
  because people are seeing their individual electricity bills
• 2:34
  go up.
• 2:35
  They see these headlines around these massive
• 2:37
  AI data center projects.
• 2:38
  Famously,
• 2:39
  we've seen that big project
• 2:40
  that's been transposed on the island of Manhattan,
• 2:43
  and people are wondering, who do I blame?
• 2:46
  And in most cases, they want to blame Meta,
• 2:49
  or alphabet
• 2:49
  or any of the companies for building these data center projects.
• 2:52
  And the reality is a little more complicated than that.
• 2:54
  Well, is that,
• 2:56
  you know,
• 2:56
  this conversation about,
• 2:57
  oh, if you put a data center in my backyard,
• 3:00
  my electric bill is going to go up exponentially.
• 3:04
  And I think that a lot of folks don't truly me
• 3:08
  included, I've had to really do some research
• 3:10
  on why
• 3:10
  that's the case, and it's because the demand that's required
• 3:14
  is not going to be satisfied.
• 3:16
  But the infrastructure that's already in place, do you
• 3:18
  is that one of the main reasons why those costs are going
• 3:21
  to get passed down to the average consumer, like you and I,
• 3:23
  if we have our homes,
• 3:24
  you know,
• 3:25
  in the same neighborhood as a data center,
• 3:26
  what else do you see that's causing that to happen?
• 3:29
  Yeah. So kind of like one of those boomer movies.
• 3:31
  It's complicated.
• 3:32
  And I think in many ways it's it's
• 3:36
  attributing the wrong thing to the problem.
• 3:39
  And in many ways,
• 3:40
  what caused a lot of the rising electricity bills
• 3:43
  that we're seeing now
• 3:44
  wasn't the advent of AI data centers in the coming
• 3:48
  wave of AI data center construction
• 3:50
  and their their insatiable demand for electricity?
• 3:53
  It was, in fact,
• 3:53
  when Trump was elected
• 3:54
  the first time he removed us from the Paris Accords, which meant
• 3:57
  that we decommitted from getting to net zero by 2050.
• 4:01
  The response at the state level was to adopt your own renewable
• 4:04
  portfolio standard.
• 4:05
  So 32 states
• 4:07
  plus the District of Columbia have created their own renewable
• 4:10
  portfolio standards,
• 4:11
  which vary and in some cases, like Michigan, are very aggressive.
• 4:15
  But the high level is that's for spending by utilities.
• 4:18
  You got to decommission
• 4:19
  things like coal or quote unquote,
• 4:20
  dirtier sources of electricity.
• 4:22
  You got to replace it with wind and solar.
• 4:24
  And that force spending is being borne by the consumer.
• 4:28
  What we're seeing from a lot of these data
• 4:30
  center projects is actually they're either committing
• 4:33
  to pay for their own generation and in excess of what they need.
• 4:37
  So they're actually supplying
• 4:39
  additional generation capacity to the constituents of that state.
• 4:42
  They're generating additional tax revenue
• 4:45
  in some of these locations.
• 4:46
  But the other choice
• 4:47
  some of these projects are making
• 4:48
  is just not deal with the grid at all.
• 4:50
  They're going what's called "behind the meter."
• 4:52
  So they're striking a deal
• 4:54
  with a natural gas provider,
• 4:55
  a natural gas infrastructure company.
• 4:58
  So a pipeline company.
• 4:59
  And then they're buying their own natural gas generators.
• 5:02
  So they're basically building
• 5:03
  their own substation on the facility.
• 5:07
  And they're not dealing with the commission.
• 5:08
  They're not dealing with the the local utility.
• 5:11
  And that helps circumvent this.
• 5:13
  But it's a very complex issue.
• 5:15
  And what we're seeing
• 5:16
  is that through not just the evolution of AI
• 5:20
  and AI data centers and the increasing
• 5:22
  electricity demand forecasts that we're hearing and reading about
• 5:26
  and CIO views here, talking about what we're also seeing
• 5:30
  is that the conflict in
• 5:31
  Iran is putting focus on maybe some of the segments of energy
• 5:35
  that most people aren't focused on.
• 5:37
  42% of our generation for electricity in this country
• 5:40
  comes from natural gas.
• 5:42
  Thinking about the energy infrastructure in this country,
• 5:44
  the pipelines that connect natural
• 5:45
  gas production and consumption,
• 5:47
  that has been a massive blessing because even though crude
• 5:51
  oil prices spiked early in the Iran war
• 5:54
  and have gone up and down,
• 5:55
  what we're seeing is that natural gas prices didn't budge.
• 5:58
  And the reason they didn't budge is because we weren't
• 6:00
  reliant on Qatar natural gas.
• 6:02
  We weren't reliant on Middle Eastern natural
• 6:04
  gas coming through the Strait of Hormuz.
• 6:06
  And that insulated us somewhat.
• 6:08
  But it's also created
• 6:09
  opportunities for us to not just export more of our natural gas,
• 6:13
  but for these companies
• 6:14
  to be a little bit more thoughtful
• 6:15
  in how they implement these things.
• 6:17
  Combinations of wind and solar
• 6:20
  and natural gas is a way to generate electricity, right?
• 6:22
  Renewables obviously have to stay in the conversation
• 6:25
  if we're going to meet the demand that's required.
• 6:27
  You know, interestingly, and you mentioned CIO and,
• 6:29
  you know, just to give a number around this,
• 6:31
  the expectation for global data center electricity consumption,
• 6:35
  they expect to go from 415 terawatt hours from 2024 to 945
• 6:41
  terawatt hours by 2030, which is a roughly 15% annual growth.
• 6:45
  So given that number,
• 6:47
  and I know you've got your own data and stats, by the way,
• 6:49
  some of that was cited by the IEA and other organizations
• 6:52
  that they have cited some of that data from,
• 6:56
  what's the
• 6:57
  what's the possibility
• 6:58
  that we can actually meet the demand that's required?
• 7:00
  And like you said,
• 7:00
  some of these data centers and the companies
• 7:02
  that are building them are maybe thinking
• 7:04
  about providing their own,
• 7:05
  power by using their own natural gas or other sources.
• 7:08
  But it's still going to be a pretty big,
• 7:12
  you know, demand on the grid itself.
• 7:13
  So do you expect that this is going to be an issue
• 7:16
  or can we meet it?
• 7:17
  Yeah.
• 7:17
  So we were joking before
• 7:19
  about the famous line
• 7:20
  from "Back to the Future", where it's like 1.21GW
• 7:24
  is needed to power a Delorean to go back 30 years in time.
• 7:28
  And we're talking about hundreds of gigawatts of electricity.
• 7:31
  In fact, to put it in context,
• 7:33
  one of these data center
• 7:34
  projects, a large AI data center project, requires
• 7:37
  enough electricity to generate enough electricity for, say,
• 7:42
  150,000 homes.
• 7:43
  That's the type of scale we're talking about that
• 7:45
  we're now trying to bolt on to the grid.
• 7:47
  And so you ask a question about investability earlier.
• 7:49
  I think it ties into this concept
• 7:51
  because the answer is right now, no, we're not equipped
• 7:55
  to handle this type of load growth.
• 7:56
  There are very smart people trying to solve these
• 7:59
  these challenges,
• 8:00
  but it's going to require a mix of different energy sources.
• 8:04
  And if you look at net new capacity added to the United States
• 8:08
  in terms of electricity generation, 90% of that net new capacity
• 8:11
  is wind, solar with supplemented largely by batteries.
• 8:14
  And so we're going
• 8:15
  to need technology to enable the implementation
• 8:19
  of those sources of electricity.
• 8:21
  Because you think about
• 8:22
  when the sun isn't shining and the wind is blowing.
• 8:24
  What do you do with that? Right.
• 8:25
  Well, you might
• 8:26
  store it in an enterprise level battery,
• 8:28
  but that doesn't necessarily work
• 8:30
  with the framework of the existing grid
• 8:32
  because the grid, which is 100 years
• 8:33
  old, was designed for energy to be or electricity
• 8:37
  to be generated here
• 8:38
  and then to be pushed out here to a house or to houses,
• 8:42
  to an enterprise.
• 8:43
  And what we're trying to do is have electricity.
• 8:46
  Come here, go there, maybe park here for a little while
• 8:49
  and then go there,
• 8:50
  and then come back here and come back here
• 8:51
  and that's very complex.
• 8:53
  And so there's a software layer to this where companies
• 8:56
  are designing software that optimizes residential rooftop solar.
• 9:02
  On how storage.
• 9:03
  So batteries as well as inclusive
• 9:07
  of things like wind and solar at enterprise level.
• 9:10
  So those virtual power plant setups, as they're called, are
• 9:13
  one solution, augmented in many cases by things like natural gas,
• 9:17
  which is low latency and highly reliable and abundant
• 9:21
  and fairly cheap.
• 9:23
  And then we are also seeing the advent of nuclear.
• 9:26
  And so in many ways, in order for us
• 9:28
  to get to where we want to go and realize our AI ambitions,
• 9:32
  and not just AI data centers, but electrify transport
• 9:36
  and automate facilities
• 9:37
  and have home robots and have autonomous vehicle fleets.
• 9:41
  We're going to need a lot more transmission,
• 9:43
  which means upward pressure on things like copper.
• 9:45
  And so from an investability
• 9:46
  perspective or argument to advisors that we engage with, is
• 9:50
  you really have to think about this as a comprehensive sleeve
• 9:54
  of your asset allocation dedicated to this theme.
• 9:57
  So it has to have categories like utilities and materials
• 10:00
  and energy as well as renewable portfolios.
• 10:04
  And of course some exposure to nuclear, eventually assuming,
• 10:08
  of course,
• 10:08
  that we get the type of momentum from a regulatory perspective
• 10:12
  that we need to support
• 10:13
  high scale nuclear as well as small modular reactors,
• 10:16
  which have been in the news a lot. Right?
• 10:17
  Absolutely.
• 10:18
  I mean, and regulation
• 10:19
  could potentially be a headwind here and there as well.
• 10:21
  I mean, we know that in New York State,
• 10:23
  they put a moratorium of one year
• 10:24
  on any new data center buildout.
• 10:26
  And that might be something that we might see pop up more
• 10:29
  and more as we go through this process
• 10:30
  of trying
• 10:31
  to add more data centers to supply the AI and technology
• 10:34
  that's needed to run all of this.
• 10:35
  By the way,
• 10:35
  the AI is needed to figure out how to use less electricity.
• 10:38
  So that's sort of like
• 10:39
  it's like like the cat eating its own tail, you know, no doubt.
• 10:42
  And it's going to be a bumpy road,
• 10:44
  I think
• 10:45
  as we try and improve the grid and we invest all of this money
• 10:49
  in modernizing
• 10:50
  the grid,
• 10:51
  and we try and overlay
• 10:52
  all of these emerging technologies on the grid,
• 10:56
  there's going to be a winding road to where we want to go.
• 10:59
  And I live in Colorado.
• 11:01
  The the local utility, Excel,
• 11:04
  actually was sued because of a fire that nobody could actually
• 11:08
  necessarily place blame on them. But they settled.
• 11:10
  And so they've instituted what's called a public safety power
• 11:14
  shut off, meaning
• 11:14
  they preemptively shut down parts of the grid
• 11:17
  when there's a windstorm or dry conditions.
• 11:20
  And so our house was shut down for two separate 36 hour periods.
• 11:25
  And we have rooftop solar.
• 11:26
  So what did I do?
• 11:27
  I went out and I bought a battery for the house.
• 11:29
  We're seeing that at the residential level
• 11:31
  where people are trying to take control
• 11:33
  of their own electricity, understanding that the grid is fraught.
• 11:37
  And we're expected, based on Bloomberg's
• 11:40
  forecast, to spend $100 billion a year
• 11:43
  over the course of the next 25 years on the grid,
• 11:48
  not on data centers,
• 11:49
  not on necessarily generation, just on the grid.
• 11:53
  Cell spent $115 billion in 2025,
• 11:56
  so we're almost ahead of schedule.
• 11:58
  The utilities in the United States, regulated utlities and independent
• 12:02
  power producers are going to spend upwards of 1.3
• 12:05
  1.4 trillion in CapEx over the next five years.
• 12:08
  So you just think about all of the investment
• 12:11
  that's going into the category.
• 12:13
  And from a portfolio construction perspective,
• 12:15
  what we tell people is you own so much of the companies
• 12:19
  driving rising electricity demand.
• 12:20
  Talk about these massive CapEx budgets
• 12:22
  coming from the hyperscalers and the magic seven,
• 12:25
  the Nvidia's of the world.
• 12:26
  Well,
• 12:27
  the companies
• 12:27
  that stand to benefit
• 12:28
  from rising
• 12:29
  electricity demand
• 12:30
  are typically underrepresented in an asset allocation framework
• 12:33
  utilities, materials, energy,
• 12:35
  and specifically energy infrastructure.
• 12:37
  Those three sectors add up to less than 10% of the S&P 500
• 12:41
  by weight.
• 12:41
  The sectors where the Mag seven live
• 12:43
  represent more than 50% of the S&P 500 by weight.
• 12:46
  Commodities which stand to benefit from this upward pressure
• 12:49
  on things like copper.
• 12:50
  Yeah.
• 12:51
  The key components of grid infrastructure,
• 12:53
  they're not typically
• 12:54
  well represented in an asset allocation framework.
• 12:57
  And so from an invest ability perspective, they should be it's
• 13:00
  really about
• 13:01
  the benefits of diversification from some of these categories,
• 13:04
  but also how it insulates to some degree client
• 13:09
  portfolios from some of these long term challenges.
• 13:12
  And importantly,
• 13:13
  it ties into what investors are actually experiencing
• 13:17
  because their utility bill might be going up,
• 13:19
  they might see headlines about AI data centers.
• 13:21
  They might live in New York
• 13:22
  and see there's a moratorium on data centers.
• 13:24
  And they say, well, that's clearly the bad guy.
• 13:27
  But in some cases, it's about the complexity of the challenge.
• 13:31
  And understanding how that might benefit you as an investor.
• 13:36
  Yeah.
• 13:36
  And by the way, that aligns
• 13:37
  very much with our CIO base case,
• 13:39
  which is yes, electricity and demand and data center
• 13:42
  expansion are
• 13:43
  are definitely on the, on the, the structural trends.
• 13:46
  But you've got other areas like,
• 13:48
  power equipment generation, storage materials,
• 13:51
  they're all part of our thoughts on how to diversify a portfolio
• 13:55
  using all the different allocations
• 13:57
  within the AI adjacent spaces that you just talked about, too.
• 14:01
  So very much aligned with how we're thinking here at UBS.
• 14:04
  And the way we put it simply is AI is software, and software
• 14:09
  requires hardware,
• 14:10
  which is chips and servers and racks and data centers.
• 14:13
  But hardware requires power and power requires
• 14:17
  commodities, natural resource infrastructure, infrastructure,
• 14:21
  everything. Right.
• 14:22
  And that is long term
• 14:25
  where we see a lot of the potential diversification benefits,
• 14:30
  but also relative outperformance is in the categories
• 14:33
  that people are typically underexposed to.
• 14:35
  Excellent.
• 14:36
  Paul. Thanks! Good to have you back. Always a pleasure.
• 14:38
  Good to see you. Yeah. Please come back soon.
• 14:40
  Paul Baiocchi again from SSC&Alps.
• 14:42
  And thank you all for joining us for this conversation.
• 14:45
  Lots more where this came from.
• 14:46
  You can find it at ubs.com/views.
• 14:49
  That's our InSight's website.
• 14:50
  Plus you can follow us on the various social media
• 14:53
  channels like LinkedIn and Facebook.
• 14:54
  Plus the studio has its own dedicated Instagram channel.
• 14:57
  Follow us @UBStrending
• 14:58
  and lots of content there for you.
• 15:00
  And as always,
• 15:01
  if you do have any questions
• 15:02
  about what Paul and I spoke about today,
• 15:04
  please
• 15:04
  make sure you're continuing this conversation
• 15:06
  with your UBS financial advisor.
• 15:08
  Until next time. I'm Anthony Pastore.
• 15:10
  Have a great rest of your day, everybody,
• 15:11
  and remember to keep your eyes on what's trending.
• 15:13
  We'll see you soon.