The views expressed are those of the authors at the time of writing. Other teams may hold different views and make different investment decisions. The value of your investment may become worth more or less than at the time of original investment. While any third-party data used is considered reliable, its accuracy is not guaranteed. For professional, institutional, or accredited investors only.
Accelerating investment in and adoption of artificial intelligence (AI) may boost economic productivity and inclusive wealth creation for decades to come. At the same time, AI’s intensive computing operations have raised questions about potential strains on energy and water resources, risk of delays to data center development, and the tech sector’s ability to meet its ambitious decarbonization goals. As the size and number of data centers run by the largest cloud service providers (known as hyperscalers) expands, their ability to scale compute capacity, manage costs, and protect their social contracts — including their right to operate — has come into sharper focus. The differences in hyperscalers’ ability to manage these issues could create fault lines that determine long-term outperformers and laggards.
Here, we outline our research aimed at identifying hyperscalers with responsible resource-stewardship practices. We also share examples of our engagements, which aim to help companies increase their financial and social value and create lasting competitive advantages.
In the grand scheme of things, “big tech” is generally viewed as cleaner than many other sectors. Some of the largest hyperscalers and enterprise software players are known for their ambitious climate goals, efficient operations, and significant clean-energy purchasing. Despite the sector’s robust growth, data centers, their value chains, and downstream computing are responsible for just 1.8% to 3.9% of greenhouse gas emissions today.1 As compute capacity continues to consolidate among a few hyperscalers, however, these companies face more intense scrutiny on both the social impacts of AI and their environmental footprint — including their energy and water consumption, particularly within their local geographies.
Computing itself is a major component of the industry’s energy and water demand. Consumption has increased over the past two years amid the introduction of new generative AI models, including significant compute requirements for model training. The industry and these models themselves are in flux, as efforts to boost computational efficiency and shift to smaller models for more specific applications are underway. Both efforts will likely reduce energy and water use, along with costs. As model developers start to differentiate their approach, computing suppliers are also competing to develop more efficient equipment, including semiconductors, servers, networking tools, and fiber (Figure 1).
Figure 1
For illustrative purposes only. Source: Wellington Management
Suppliers that have proactively integrated resource efficiency into their research and development before this latest AI wave continue to distinguish themselves. For example, some semiconductor companies with efficiency-oriented designs have announced new energy-saving and water-cooling techniques for chips and packaging that may dispel historical assumptions about what the industry can accomplish.
Energy is not only required to power evolving computing capacities, but it is also essential for facility operations and equipment cooling. Currently, 95% of data centers are air-cooled, a relatively inefficient and difficult-to-scale solution.2 Liquid cooling technologies that chill chips or servers directly and siphon off excess heat are more energy efficient than air cooling, in many cases. The challenge is the requirement for more water. Hyperscalers’ robust water demand may not be an issue in water-rich regions, but it can be a tremendous physical, economic, and social challenge in water-scarce regions — which, according to our research with Woodwell Climate Research Center, are becoming more numerous. Notably, our analysis suggests that some of the largest data-center operators appear to be shifting to less water-scarce regions while ramping up water recycling and other efficiencies.
Hyperscalers’ robust water demand may not be an issue in water-rich regions, but it can be a material physical, economic, and social challenge in water-scarce regions [...]
A core question we seek to answer is: Which operators are best prepared for necessary data center development and long-term cost performance as they compete to lead this new wave of AI? We also seek to assess the risk of delays or pushback on data center development, owing to resource requirements and each company’s ability to manage these risks. We have found that a few factors are critical to both questions and help us determine whether a data center operator is poised to become a leader or a laggard:
Siting decisions that minimize physical limitations
Hyperscalers have a fair amount of flexibility when it comes to facility siting. They can expand into new regions to access favorable electric utility relationships and plentiful water resources. Our research finds that certain operators are more adept at siting, establishing footprints in regions with ample water supplies and electricity development potential. These first-mover advantages may ensure that companies can secure building permits and other approvals faster than their competitors and may more easily avoid future operational disruptions.
Strengthening local relationships
As AI is increasingly viewed as a “strategic sector,” with both national security importance and local employment implications, hyperscalers can make a stronger case for their possible contributions to local economic and social development. Here again, we have found certain companies to be more capable than others in developing these relationships and prioritizing the benefits to immediate jurisdictions, some extending to aid development of wider tech hubs and support job training in AI. These efforts may help accelerate speed to market and minimize operational scrutiny from local or national regulators.
Strategically managing power purchasing and clean-energy investments
In the same vein, arrangements with local utilities, most notably clean-energy power purchase agreements (PPAs), can be designed to create “win-win” economic outcomes for local constituents. Some operators have navigated this process well, combining utility concessions to support grid reliability and ensure they pay their fair share of rates. A few are taking an even longer-range view, securing clean-power supply through large-scale development deals in which data center operators assure the utilities of their long-term energy demand. Finally, some operators have sought to support the development of nascent clean-energy technologies that also help utilities with the long-term challenges of grid reliability, including long-duration energy storage, microgrids and other transmission solutions, and clean forms of baseload power. All of these efforts are critical to reaching clean-energy goals in partnership with utilities as power consumption from AI use increases.
Increasing energy and water efficiency
Efficiency in all forms is, and will continue to be, a cost imperative for data centers. Companies that build energy and water efficiencies into facility design and computing capacity may maintain cost-savings advantages, minimize the risk of interruption, and protect their right to operate. Water rights in particular are under increasing scrutiny at locations with high water stress or with data center operators that are dominant users. We believe hyperscalers and other operators with strong water programs are better positioned to secure and maintain their right to operate. Likewise, data center operators that prioritize supplier relationships to maximize energy efficiency — including developing custom chips to help drive longer-term efficiency gains and help keep costs under control — may be better able to manage costs, adhere to net-zero and other climate goals, and manage energy requirements. All of these efforts accrue options for siting, local relationship building, and the purchase of clean power.
Data center operators that prioritize supplier relationships to maximize energy efficiency [...] may be better able to manage costs, adhere to net-zero and other climate goals, and manage energy requirements.
As part of our research process, we engage regularly with hyperscalers to ensure that their approach to these issues is, in our view, sound and value accretive for investors. We cross-check the intelligence we gain during engagements with fundamental data on companies’ capital and operational expenditures. We also share insights from our own research and observations of evolving best practices. As awareness of AI’s energy and water demand has come to the fore over the past year, our engagements have also been an opportunity to assess how data center operators respond to pressure. In several cases, we have encouraged companies to be more transparent about their strategies to manage water and energy usage, which we believe can relieve some of the pressure and scrutiny from regulators and investors.
Example: Promoting transparency and ensuring adherence to commitments
Our engagements with two large-scale data center operators have reinforced our view that the companies are serious about their environmental commitments and take a long-term view on strengthening local relationships, power purchasing strategies, and investments in future technologies. These two firms have also faced challenges with data center buildouts this past year, due in part to energy and water supply issues.
Over the course of our interactions, the companies have gradually increased transparency about these challenges and how they plan to manage them while keeping their climate commitments. These engagements have enabled us to convey strong investor interest in their ability to manage these challenges and have led to commitments from both companies for additional communications and disclosures. We now better understand their strategies for water investments, power arrangements with specific utilities, and rationales for using or avoiding alternative mechanisms such as carbon offsets, which can lower costs but are controversial among some stakeholders.
Example: Monitoring improvement and shifting to potential leadership with utility partners
Our engagements with a company revealed how far it has come and where it might focus its longer-term growth strategy. The company’s right to operate came under pressure a few years ago because of its less-well-developed energy and water management strategy. During the course of our interactions, we have seen this company establish more robust practices, confirm its commitments to adhere to these practices at the site level, and catch up to competitors on clean-power purchasing and general efficiency building. Recently, the company significantly increased its focus on energy as a critical resource and started to emphasize specific clean-energy technology priorities that could potentially be an advantage as it partners with utilities.
Example: Understanding an approach to supplier partnerships
We recently began engaging with a company on environmental management as it navigates increasing AI demand, shifting relationships with data center partners, and emerging European disclosure regulations. The company previously depended on cloud and colocation partners and had minimal involvement in climate/environmental management. It recently convened an internal team to manage corporate energy and water use amid plans for direct data center buildouts. Our engagement helped us gauge how proactively the team is in building best practices, engaging with regulators, shifting from indirect to direct energy purchasing, and pushing for colocation supplier transparency on energy and water. We saw early signs that the company may aim to develop efficiencies such as liquid cooling and compute optimization and pursue siting options for its data centers. We believe the company may be able to realize a mix of cost-efficiency gains and rapid expansion as its energy and water management matures.
Amid rapidly evolving power and infrastructure needs for AI systems development, this research is invaluable in helping us identify potential outperformers in this space.
Active management and a deep understanding of sustainability are crucial in navigating this rapidly expanding investment area and the growing resource demands of AI-driven data centers. Hyperscalers’ ability to proactively manage energy and water usages, make strategic siting decisions, and build strong local relationships will be among the keys to establishing and securing competitive advantages and sustainable growth. We believe that companies that excel in these areas can lead in the AI future, ensuring cost efficiency and adherence to climate goals. In our experience, engaging with hyperscalers to promote transparency and best practices can drive financial and social value, and help further position them as long-term outperformers.
1Sums determined via a combination of Wellington’s proprietary modeling and company meetings, along with academic models of long-term compute energy demand, International Energy Agency (IEA) forecasts, and financial-industry models. | 2Yang Jie, “Novel ideas to cool data centers: Liquid in pipes or a dunking bath,” Wall Street Journal, 11 August 2024.
Stay up to date with the latest market insights and our point of view.
