BMO Equity Research on the AI + Data Center Build Out: Sustainability Impacts, Second Order Beneficiaries
BMO Equity Research on the AI + Data Center Build Out: Sustainability Impacts, Second Order Beneficiaries
Directeur, Stratégie ESG
M. Morrow s’est joint à l’équipe de recherche sur les actions de BMO Marchés des capitaux en septembre 2020 en quali…
Première directrice générale et chef, Recherche sur les actions, Canada et Royaume-Uni
Camilla s’est jointe à BMO Marchés des capitaux en 2020 à titre de première directrice générale, Recherche sur les a…
M. Morrow s’est joint à l’équipe de recherche sur les actions de BMO Marchés des capitaux en septembre 2020 en quali…
VOIR LE PROFIL COMPLETCamilla s’est jointe à BMO Marchés des capitaux en 2020 à titre de première directrice générale, Recherche sur les a…
VOIR LE PROFIL COMPLET-
Temps de lecture
-
Écouter
Arrêter
-
Agrandir | Réduire le texte
Disponible en anglais seulement
In this special episode from BMO’s IN Tune Podcast, Camilla Sutton and Doug Morrow sat down to discuss how the AI/data center build out is contributing to a new era of electricity demand.
IN Tune features Equity Research analysts from BMO Capital Markets and explores key emerging themes, trends, and important issues to our institutional clients globally.
Subscribe to this free podcast and never miss an episode or visit the BMO Equity Research website for more great episodes of IN Tune.
Camilla Sutton: One of the most important investment themes right now is the AI and data center build out. I'm thrilled to have our ESG analyst Doug Morrow, join us today to walk us through how to think about this from an ESG investing lens. Doug, let's get this kicked off. In late June, your team published a note titled, The AI + Data Center Build Out: Sustainability Impacts, Second Order Beneficiaries. Walk us through it. What are the key takeaways for investors?
Doug Morrow: Thanks, Camilla. I think the first key takeaway for investors is that AI is really going to push us into a new era of electricity demand. So in the US right now, electricity demand over the last 10 years or so, maybe 12 years has actually been relatively flat. So even though GDP went up by over 60% over the last 12 years, the amount of electricity used in the economy was relatively stable. This is because of efficiency. Basically, as in other economies, the US economy has become much better at squeezing more value out of each electron due to more efficient manufacturing processes, industrial processes, consumer products, etc. But what's going on now is that AI is coalescing with other trends, including the electrification of transport, the electrification of heating, reshoring, and even climate change, to just fundamentally change how much electricity we use. So generative AI models such as chat GPT, are really fun to use, you know, my kids like to use them to ask questions, but they are 10 times more energy intensive than a Google search. Second, there's going to be a supply response to this trend, and it's not going to be green, or at least not fully green. And what I mean by that is, if you look again, at the US, for example, about 77% of all new generation capacity brought online last year was renewable. So basically a mixture of wind, solar and storage. And that's because the US like most countries is trying to green its grid. So yeah, to kind of nail at home, eight out of every 10 megawatts added to the grid last year in the US was zero carbon. But I don't think we're going to see that ratio again, for a while. The demand numbers coming from these data centers are just so big that they're going to require more natural gas plants to come online. And we're also going to see existing coal and nuclear assets stretched out as much as possible. So the demand numbers coming from the data centers are just so big that they're going to require more natural gas to come in line, and existing coal and nuclear assets to be stretched out as much as possible. I mean, we're definitely going to continue to see more renewables come online too, but it's not going to be 77% of the total. Our best guess is something in the neighborhood of 50% gas and 50% renewables. And then the final takeaway for investors here is just to point out that data centers power these generative AI models use large amounts of water to cool their IT equipment. So the typical air-cooled hyperscale data center uses 2.1 million liters of water per day. And most of that has to be potable, i.e., freshwater. And that means that data centers are competing with farmers, local communities, and other industrial users for a resource that is showing more and more signs of strain, especially in places like California, the Colorado River Basin in parts of the Midwest. So Phoenix is actually the world's second largest data market by commission power. And Phoenix is a city that is under growing water strain. So approximately 20% of data centers in the US right now, are based on are based in watersheds that face moderate to high stress. And we do not believe that the water and grid congestion impact to data centers are widely appreciated by local communities despite some of the protests that we've seen. So yeah, I think this means there is a risk of community opposition to data centers that is probably being underestimated. And we think investors should be engaging with the management teams, and hyper scalars such as Google, Microsoft, Meta Platforms, and Amazon, as well as the colocation companies such as Equinix and Digital Realty, to see how they're thinking about mitigating this risk.
Camilla Sutton: Some of these stats are mind boggling, Doug, and they really do support your view that the era of flat electricity demand is completely over. Let's shift gears a little bit, kind of where you left off in the last answer. You know, what are the large tech companies like the Microsoft, Amazons, Metas, Alphabet thinking and doing regarding sustainable approach to AI and data centers?
Doug Morrow: Yeah, I mean, generally, they're doing a really good job, they are highly incentivized to green these data centers as much as possible. Not only does this drive efficiency and help them lower costs, but it also helps counteract that community opposition angle, at least to a certain extent that we talked about. So yet, like I said, generally speaking, we think these companies do an excellent job on sustainability. But what we're finding is that they face a trade off right now between growing their AI business and hitting their sustainability targets, at least over the short run. So where it really hits home for them is on scope three emissions. So scope three, this is another one of the takeaways from the note is really going to be the next frontier in greening data centers. So let's look at Microsoft, for example. Microsoft, in our opinion, is a great climate company. They've slashed scope one and two, they have a strong commercial upside to our warming planet, they are part of the BMO climate opportunities screen, but their scope three emissions are 76% off target. And this is mostly due to the embodied carbon in their data centers and IT equipment. It's a similar story at Meta. We would say Google is probably doing the best on scope three, but they do take a bit of a different approach compared to Microsoft and Meta. Amazon trails the other hyperscalers generally on sustainability. One of the things we highlight in our note is they're the only ones amongst that peer group that do not publicly report their water consumption. And they actually stopped reporting their electricity use back in FY22. So those are some things that we flagged in our note.
Camilla Sutton: That's amazing Doug, appreciate it. You argue also that several sectors are going to benefit as the green data center theme takes hold. What are the sectors and specifically which companies did you identify with that revenue upside?
Doug Morrow: The themes we like here are grid stabilization, clean power production, alternative cooling, low carbon building materials and data center engineering. We see these themes as second derivative plays on the broader AI and data center thematic. On grid stabilization, we like Fluence Energy. This is an energy storage provider that is exploring behind-the-meter solutions for data centers. On clean power, we highlight companies like Constellation Energy and Vistra, as they are well positioned to benefit as US electricity consumption increases. And they also have nuclear in their generation mix which opens up the possibility of doing nuclear PPAs with some of the hyperscalers, which is a new trend, and we think that could play out in the next little bit. Companies offering alternative cooling would include names like Chemours and Ecolab, and for engineering, we highlight Stantec and WSP. Low-carbon building materials, this is a bit more challenging, it's still an emerging field. But for low carbon aluminum, IT equipment and green steel, we highlight names like Alcoa, Nucor, and Palo Alto Networks as solution providers. And all these things are going to be essential for the large hyperscalers to tackle the scope three emissions in their data centers.
Camilla Sutton: By the way, Doug, your table on page three of the report, it's an incredible summary of just what you've said all of these names with a lot of details, they were really useful as a reader. Why don't we dig in a bit more to the alternative cooling theme? Can you tell us what Chemours is doing versus companies like Stantec and WSP?
Doug Morrow: Yeah, so right now, most data centers use air cooling, to dissipate the heat from their IT equipment, which as you can imagine running 24/7 very powerful machines, they throw off a lot of heat. So this system basically involves traditional air conditioning, they also sometimes use a heat exchanger that creates chilled water that runs through pipes that's used to dissipate the heat. Now, as mentioned already one challenge with this type of technology is it uses a lot of water. But another perhaps even more important drawback over the short term, is it’s starting to bump up against technological limits, it’s starting to prove incapable of simply dissipating the amount of heat that's being produced by these higher performance chips that are used in AI applications. So as a result, over time, we think we're going to see more data centers shift to these alternative liquid cooling technologies, especially hyperscale size AI dedicated data centers. Now some of these new technologies offer significant sustainability benefits. For example, if you think about two-stage immersion cooling so this is where it equipment is actually fully immersed in a tank of dielectric fluid. In, this can yield 100% reduction in water consumption and a 95% reduction in energy consumption that's tied to cooling. This is all compared to air cooling. So really some significant sustainability advantages as well as more powerful and dense cooling potential. The company from our coverage with the most torque to this theme, the two, the two phase emergent is Chemours. Now, you mentioned Stantec, and WSP, you know, these are really strong data center engineering names, and they may be involved in data center layout and design, but it's really Chemours that has the major upside exposure to this two-phase immersion cooling.
Camilla Sutton: It's amazing how all that technology is changing so quickly. Well, why don't we close out here with a big blue sky question? With everything that you've learned in researching for this note, do you think AI will have a net positive or a net negative for the environment overall?
Doug Morrow: Well, that's really the top of mind question for many people in the industry. It's simply difficult to say at this point. But as a natural optimist, I'm inclined to say net positive. I think AI is showing promise in improving sustainability outcomes in many fields. If we think about things like traffic and route optimization, precision agriculture, resource forecasting, etc. There's a certain irony, though, in that AI is also being used to optimize power grids, when the massive power requirements of AI models in the first place are probably contributing to grid instability. So I think there's also a certain irony in how we're using AI to solve problems that are in part being driven by AI. Look, I think, I think over the short run, it's really crucial that investors understand the electricity and water appetites of generative AI models, not only from an environmental standpoint, but also because we think mitigating these impacts over time, is going to provide really compelling investment opportunities in and of themselves.
Camilla Sutton: Doug, I love where you've taken us on this podcast and within your note, it's been a really interesting exploration deep dive into where we are with data centers in the AI build out from a sustainability lens. That was Doug Morrow, Director of ESG at BMO Capital Markets.
