The Bid Episode 250: Powering AI 2.0: Why the AI Boom Is Becoming an Energy Story
Web title: Why the AI Boom Is Becoming an Energy Story
Full episode Description:
Powering AI 2.0 is no longer just a technology story — it’s an energy and infrastructure story reshaping capital markets and the global economy. As artificial intelligence scales from training to real-world inference, electricity demand is accelerating at a pace few anticipated.
In this episode of The Bid, host Oscar Pulido is joined by Will Su from BlackRock’s Fundamental Equities Group to examine how Powering AI 2.0 is transforming utilities, natural gas markets, renewables, and nuclear power. With data centers expanding rapidly and gigawatt-scale facilities coming online, the AI build-out is driving a structural shift in U.S. electricity demand after more than a decade of stagnation.
Will explains why the energy sector sits at the center of AI investing. From the rise of bring your own power models to the growing role of natural gas as a dispatchable, scalable fuel source, the infrastructure required to support AI represents one of the largest capital investment cycles in modern history. The conversation also explores renewables, battery storage, and nuclear power — including the limits of restarts and the long timeline for new reactor construction.
Key insights from this episode:
Why natural gas has emerged as a key here and now fuel for AI infrastructure
How renewables and battery storage fit into the AI electricity mix
The long-term outlook for nuclear power and reactor construction
What bring your own power means for hyperscalers and utilities
How electrification and reshoring intersect with AI investing
Why the relationship between compute and energy is reshaping stock market trends
Keywords: Powering AI 2.0, AI investing, infrastructure, capital markets, energy transition, utilities, stock market trends, megaforces
Sources: From CES 2026 to Yottaflops: Why the AMD Keynote Highlights a Turning Point for AI Compute, AMD 2026; The Industrial Revolution, coal mining, and the Felling Colliery Disaster, Lancaster University, 2026; Bureau of Economic Analysis data 2026; Stargate's First Data Center Site is Size of Central Park, With At Least 57 Jobs, Bloomberg 2026; Energy Demand from AI, IEA 2026; Scaling bigger, faster, cheaper data centers with smarter designs, McKinsey 2025; EEI 2024 Review; Data Centers Ditching the Power Grid, Mark Carney's Viral Speech, and Some Joy, Clearview Energy; 2024 North American Energy Inventory, IER;
Written Disclosures In Episode Description:
This content is for informational purposes only and is not an offer or a solicitation. Reliance upon information in this material is at the sole discretion of the listener. Reference to any company or investment strategy mentioned is for illustrative purposes only and not investment advice. For full disclosures, visit blackrock.com/corporate/compliance/bid-disclosures.
<<TRANSCRIPT>>
Oscar Pulido: AI has been growing exponentially since Chat GPT burst onto the scene in November of 2022. The backdrop for the energy supporting AI today looks very different from two years ago. Models are growing larger data centers are scaling faster, and the demand for electricity is rising at a pace few expected. The story now goes beyond innovation in silicon. It's about the strain on global power grids, the race to build new generation, and the emerging divide between countries rich in energy and those constrained by it. AI becomes more deeply embedded in the global economy. The question isn't just what this technology can do, it's what amount of energy will it require to run?
Welcome to the Bid where we break down what's happening in the markets and explore the forces changing the economy and finance. I'm Oscar Pulido.
Today we're joined by Will Su from BlackRock's Fundamental Equities Group. Will has been closely tracking the surge in electricity demand. Driven by AI and the implications for natural gas renewables, and nuclear power. We'll discuss how the energy landscape has shifted over the last 18 months, whether the world can build fast enough to meet AI's needs, and how investors might expect to see the energy landscape evolve.
Will, thank you so much for joining us on The Bid.
Will Su: Thanks for having me back, Oscar.
Oscar Pulido: It has been almost two years since you were here. You and I spoke in May of 2024, and I actually remember that conversation because, Will, I think you were one of the first people that at least, drew some attention in my mind to the fact that the AI story is also an energy story. In fact, AI consumes a lot of energy, and you put a lot of numbers around that. AI has obviously become an even more dominant part of the conversation and ever more important part of the economy. So, can you talk to us about what changes you've seen in the energy landscape as it relates to this AI industrial revolution?
Will Su: So, a lot has changed over the last two years, and I think the biggest change has really been this deepening recognition of the critical role that energy plays, not just to power artificial intelligence, but in some ways being artificial intelligence. So, if you recall from our last conversation, the way you get these AI models to become smarter is to basically train them with more tokens and parameters. And that's just a fancy way of saying you're feeding in more data and you're building models with more and more variables, so they become increasingly intelligent.
The more tokens you put in, the more parameters you run, and the more pre and post training that you do, the smarter the models become, the closer they start to approach to human level general intelligence, and could one day surpass it. These improvements are coming with an exponential increase in power demand.
So even with some of the efficiency gains that we have made on the semiconductor side, the reality is today's AI models are more power hungry than ever to create. On our last episode, we talked about the concept of zeta flops. So let me introduce a new term. Today, we are firmly marching into the world of yottaflop compute. That's 1 trillion, trillion calculations per second. Again, just mind-bending numbers. And of course, each calculation just involves the movement of electrons between the semiconductor chips. So over time that power demand really starts to add up and grow exponentially. I like to think of it this way, since the beginning of human history, knowledge is power, but for the first time ever, power is knowledge. So, by some estimates, we are investing more than $1 trillion per year in the United States alone to enable everything around the AI build outs. Like our tech guru Tony Kim said on his episode of The Bid, this is the largest investment in human history.
Now, there has been some legitimate concerns in the markets recently that we could be overbuilding this AI capacity and the power infrastructure that's associated with it. The skeptics tend to point to a lack of clarity around AI use cases in the real world to justify the massive amounts of investment, and my pushback to that more pessimistic view is really grounded on two points.
Number one, we're still in the very early innings of powering AI, where training dominates electricity consumption. But over time, as AI goes from being a science experiment to a part of everyday life, you're going to start to see that power demand increasingly shift towards what's called inference. And inference is basically when you take the AI models that were perfected during training, put them into the real world and have them answer questions and solve problems, and do things for us in real time. And that inference demand is going to continue to ramp up over time as adoption increases, and we're already seeing real time signs of adoption increase.
And the second point really is as the technology of AI improves and as the unit cost of compute decreases, you are likely to see this long tail of inference demand really pick up over time. And this is actually a phenomenon that we have seen in the energy markets two centuries ago around coal demand during the Industrial Revolution So just like coal demand in England where the introduction of the combustion engine has made efficiencies, that much higher coal demand eventually increased by tenfold. In England, we have a blueprint for how AI demand is really poised to see exponential growth. As that technology gets better and the cost becomes cheaper. And so as more and more consumers and businesses start to adapt and embed AI into their daily activities, as the things that AI can do for us become more complex and thus more power intensive, there is a long, long runway for more power demand that will be absolutely necessary to enable this AI future of ours.
Oscar Pulido: I feel like we've already seen an inflection point in power demand, and I think you started to touch on that. Yet, despite the fact that we're still in the early innings, we have been seeing and hearing a lot about the exponential growth of data centers, which are needed to, to power this demand for AI and its development. How big of an impact is AI having on electricity demand in the world?
Will Su: Great question I think there's no mystery at this point to anyone that there's a lot of investment going into building these AI factories of the future. It's almost a cliche at this point, but we really are all living and investing in an AI driven economy today. If you look around the world today, there are about 12,000 operating data centers each with an average power capacity of 11 megawatts. That's enough to supply about 10,000 households. So, it's by no means a trivial number. But just compare that with the Stargate Project being built in Abilene, Texas. When the first phase of that project comes online later this year and energizes it will bring with it 1.2 gigawatts of compute capacity. So, for reference one gigawatt is roughly the average power capacity of the city of San Francisco. There are just going to be so many more of these things being built. And by the way, they're just impressive campuses from a physical standpoint. So, Stargate is going to be roughly the same footprint as Central Park in New York City. And not only are they the size of cities, but some of these data centers also literally consume more power than some of the biggest cities in the world. And we're currently building more than a dozen of these gigawatts scale data centers around the US and around the world.
So, the question really becomes, if you add all of these together what percent of total electricity demand do they represent? the International Energy Agency estimated that in 2024 data centers worldwide consumed about 415 terawatt hours of electricity. That's about 1.5% of the world's total power demand. And here in the US we're clearly leading the world in this AI build out that market share is higher. Data centers today represent about 5% of American electricity demand.
The growth ahead is just about to get so much bigger. Since 2023 with the introduction of ChatGPT 4, which is really like this watershed moment for the rapid expansion of AI Buildouts, we have seen a whopping 200 gigawatts of new data center capacity proposed in the United States alone. But we think at least a quarter of that, or more than 50 gigawatts, could come online by the end of this decade. And that would take market share of data centers up to as much as 12% of American power production by then. And all of this is just going to drive a step change in America's power needs.
We are coming out of a 15-year period where power demand in this country has barely grown, but the forward estimates now expect up to 3% to annual growth all the way through 2040.
And what that really just means is we're going to make a lot of more investments into the overall power infrastructure within this country and around the world. Everything from generation to transmission, to distribution to last mile logistics. And I think our friend Balfe Morrison said a best on his episode of The Bid a few months ago. If you're bullish on AI and AI adoption, you have to be bullish on power and utilities. They really are one and the same.
Oscar Pulido: Yeah. And you mentioned knowledge is power, and now for the first time, power is knowledge. we really do need a lot of power to help this industrial revolution that is going on actually continue. You, you mentioned a number of statistics. I think you said that the, to power San Francisco, that's one gigawatt. and that there are plans for, as much as 200 gigawatts of data centers in the us but even if only a quarter of those pan out, that's 50 San Franciscos in terms of power demand.
And the last time you came on here, will, as part of how we generate this power, you talked about natural gas as one of the ways that, we are going to find the power to enable this AI, industrial revolution. Can you bring us up to speed on how the sector has done and what are the current trends for gas power generation?
Will Su: So, gas has really moved front and center in this whole AI data center build out, and I think it's important to just take a step back and set the scene for what gas really brings to the table here.
If you wanted to bring, bring a new data center online today, there are basically three ways you can power that data center. You can request an interconnection into your regional grid. You can build your own power plant and feed that electricity directly into your data center. Or you can do a combination of both.
So, on the grid side, the American utilities are really doing everything that they can to help us meet this future demand growth. They're investing more than a trillion dollars over the next five years to modernize every aspect of the grid. So again, generation, transmission, distribution. And that is a 30% increase from the last five-year period. So, a significant step up. But despite their best efforts, it is becoming increasingly clear that the grid alone is going to struggle to enable this AI revolution. You now have to wait more than four years to plug a new project into the grid. That is a delay that the hyperscalers can't really afford to wait for in this very competitive AI arms race.
Another very important issue is affordability for the American consumer. So, retail power prices in America were basically flat for the decade going into COVID, but after that they have grown by more than 30% in the last five years, and the trend is for additional increases in the coming years.
So, the very loud and clear message is this: to enable our AI build out and this AI future, we can no longer rely on connecting into the grid. Increasingly, we're going to have to invest in and construct new power generating assets that are dedicated to the data centers that they serve.
And this is what the industry calls BYOP or bring your own power. The power source sits behind the meter as a separate part from the overall grid to supply that data center.
This is where natural gas enjoys some really deep structural advantages that has positioned it as the here and now fuel source for the AI revolution. Natural gas is already the largest input into power generation in America, responsible for more than 40% of our generation. They're highly efficient and they release less than half the emissions when compared with coal power. We have abundant natural gas resources in this country that will meet our current demand for the next century. And gas can be deployed relatively quickly. So, when you put all that together, gas is very uniquely advantaged in this AI build out.
The next question obviously becomes, what is the investment opportunity in energy equities?
The top six natural gas producers in the country produce one fifth of all gas Su supply. That is a level of consolidation in a commodity business that we simply haven't seen since the start of the shale revolution 20 years ago. And there are other really bullish factors happening for natural gas. the most front and center one is going to be liquified natural gas or LNG. This is basically when you take natural gas and you super cool it down to about minus 260 degrees Fahrenheit, or minus 162 degrees Celsius so that it condenses into a liquid. LNG Demand has a long future to grow because the world requires this amount of electricity for their own power and their own electrification, including the build out of their own AI infrastructure.
So, if you take a look at US natural gas balances There are about 110 billion cubic feet per day, or BCFS per day, flowing through the US every day. Just to visualize that's enough gas to fill 3000 Empire State buildings. each and every day. I think between data centers and other electrification sources like the reshoring of industrials, we will add at least five BCS per day to that demand number on the power side by the end of this decade.
So, this should raise the floor price for natural gas prices and enable the lo low-cost producers to deliver better earnings and better returns for their shareholders as we go through this AI scale up in the coming years.
Oscar Pulido: You and I work down the street from the Empire State Building, and I walk by it almost every day. So, I'm trying to picture what 3,000 of those buildings looks like and then think about the volume of natural gas that is going to fill those buildings. But you're saying that's where we are now, that doesn't even take into account what the demand for natural gas could be going forward as again, we keep moving in this direction of this AI economy.
Will, we haven't talked about renewables, we talked about natural gas, but I think you've made the point that there needed to be, or needs to be, a partnership between gas and renewables to, to fuel the growth in the AI economy. How have renewable energy companies navigated the demand for power and how are they navigating the regulatory landscape as well?
Will Su: So, I think we should really be very proud of the progress that we have made in low emission power generation in this country. at the turn of the century, less than 30% of our power generation came from low emission resources, and the vast bulk of that was really nuclear and hydroelectricity is fast forward to today and more than 40% of our power now comes from low carbon emitting resources, and that's of course driven by the fast growth in both solar and wind. Solar is now almost 8% of our power generation, and wind is about 10%, both having grown from almost nothing 25 years ago.
Now that being said, the forward path between solar and wind is really starting to diverge here. While solar still has a lot of very strong momentum and the annual installation of capacity is tracking at around 30 gigawatts per year for when that number has really taken a step back and is tracking at just about a quarter of the pace of solar. And if you think about the future with data center demand front and center, I think solar enjoys some really nice structural advantages within renewables.
We're also seeing a lot more solar being built today with what's called battery energy storage systems, or BESS. What this does is it enables you to store some of the excess solar that you generate in the middle of the day and discharge it to meet demand even after the sunsets. So even though this doesn't quite replace the truly dispatchable resources like nuclear and natural gas, it does I think, make solar a more relevant and durable part of this future of power when you have data centers added into the demand profile.
Now, on the other hand for wind, we do think that wind generation will continue to grow in the coming years. Wind does face some more unique challenges versus solar. They face more issues with permitting, with zoning, with regulatory, federal regulatory pushback, along with a host of other issues.
So, I think the bottom line on renewables is this, there is going to be more renewables in our future power generation. We think more of it is going to come from solar than from wind. We think battery backed solar has an increasingly big role to play in powering data center projects as we go forward and just as it always has been, renewables will continue to work in this really close partnership. Dispatchable resources like natural gas to ensure that we have the scalability, reliability, and affordability that we have today as we expand our overall power infrastructure.
Oscar Pulido: it really sounds like you have to look in a lot of different directions for these power sources. We've talked about natural gas, you've, you've compared solar and wind, but it sounds like both are going to be part of the equation.
We haven't talked about nuclear. Last time we spoke, you had mentioned that. There were nuclear power plants that had been closed that were starting to reopen as, as we saw this sort of underlying change in the economy causing power demand to pick up. So, what's the story on nuclear now? More recently?
Will Su: Yeah. So, Oscar nuclear restarts have been a favorite among Hyperscaler customers for meeting their data centers since the last time that we spoke.
The hyperscalers prefer nuclear power for two clear reasons. Number one, they bring a reliable power load that sits behind the meter and is available 24/ 7 /365. Number two, they're an emission free power source. So really the best of both worlds. That's the good news. The bad news is on a go forward basis, we have sold out of these low hanging fruits of nuclear restarts. So, the new the nuclear story going forward is really going to have to deal around investing in and building new greenfield nuclear generating plants, and it's not going to be a walk in the park here in America today.
as of today, there is not a single utility scale nuclear reactor being constructed in the United States. I do think there are some green shoots here. Some utilities have started to move nuclear projects back into its resource space, which is a very early phase before you decide to invest in a project.
But I think the harsh truth is we don't expect to see a new utility scale reactor being built in the states for the next decade. But there's many, there's a lot of reasons to be optimistic here.
The Department of Energy has set a very ambitious goal to triple America's nuclear capacity by 2050.
So, all of this makes me really optimistic that nuclear is going to have this key long-term role to play and expanding our power infrastructure to power AI in a way that is also carbon free and also fully dispatchable to ensure reliability.
Oscar Pulido: So, Will, we've covered a lot of ground here. We've talked about the demand for energy that is going to be created as AI becomes more pervasive across our day-to-day. You mentioned natural gas. We talked about nuclear, solar, and wind, but I just want to come back to you for a second. You're a fundamental analyst. You come to work and you're looking at companies and you're trying to determine where to invest, what companies, what sectors, what industries, how has ai impacted the way you go about doing your role and the way you go about doing fundamental investing.
Will Su: Oscar, I think AI for me has really become an increasing part of my daily life. But as an investor, like you said, I think one of the key questions that I've always had is, look, we accept at this point that AI is increasingly intelligent and highly efficient. The real question and the million-dollar question really is it any good at investing and trading in the real world?
we're in luck because there's data for this. So, there is a live-stock trading competition hosted on LLM Stats. In that competition, only one out of 16 AI models is actually currently making money, 15 out of 16 unprofitable. On a separate trading competition that's hosted on Alpha Arena, again, only one out of eight AI models being tested has made money on average. So not the most impressive results so far.
And my hypothesis for this really is AI as a standalone model is not nearly as powerful or profitable as an experienced investor who can leverage that AI to improve their investment and research and risk management processes. And this combination of man and machine is really, in my view, what's going to define investment talent in the years to come.
When you combine the scale efficiency and speed of AI with the deeply knowledgeable investors that bring decades of experience, deep industry knowledge through the different economic cycles and their irreplaceable human connections in the real world. That is a secret to delivering better investment outcomes and more value to our clients and investors in the future as we march into this world of AI ahead of us.
Oscar Pulido: Will, you've done it again. You've given us a lot of great things to think about. You talked about power is knowledge, which I thought was, really memorable. You mentioned BYOP, bring your own power, which is something that builders of data centers have to think about. And you also gave us a lot of terminology that we'll have to go back and look at around the sheer quantities of power demand that are being created. But we appreciate you sharing an update on where we are with the AI and energy story, and we appreciate you doing it here on The Bid.
Will Su: Thanks for having me, Oscar. Look forward to part three of our conversation in the future.
Oscar Pulido: Thanks for listening to this episode of the Bid. Next week I'm joined by Lisa Yang to talk about the K-shaped economy. What behavior investors can expect from the US consumer.
<<SPOKEN DISCLOSURES>>
This content is for informational purposes only and is not an offer or a solicitation. Reliance upon information in this material is at the sole discretion of the listener. Reference to the names of each company mentioned is merely for explaining the investment strategy and should not be construed as investment advice or recommendation. For full disclosures, visit blackrock.com/corporate/compliance/bid-disclosures.
MKTG0226-5212730-EXP0227
