CONTRIBUTORS
Stephen Dover, CFA
Chief Market Strategist,
Head of Franklin Templeton Institute
Blair Schmicker, CFA
Portfolio Manager, Research Analyst
Franklin Equity Group
Matt Cioppa, CFA
VP/Portfolio Manager and Research Analyst
Franklin Equity Group
United States
Matt Adams
Portfolio Manager, Research Analyst,
Franklin Equity Group
Chief Market Strategist and Head of Franklin Templeton Institute Stephen Dover, together with Franklin Equity Group portfolio managers Blair Schmicker, Matt Cioppa and Matt Adams, explore the implications of artificial intelligence on different areas of the economy and the challenges that may arise. They discuss the evolution of data centers, the key players driving demand growth, and the potential future impact on the energy sector. As investors located in the heart of Silicon Valley, with a keen focus on identifying and investing in innovative growth trends, Franklin Equity Group offers a compelling perspective on this changing landscape.
Key takeaways:
- The advancement of artificial intelligence (AI) is expected to require large increases in capital investments in semiconductors, servers, networking gear and data storage, as well as the construction of new and larger data centers.
- New AI data centers will require dramatically greater amounts of electric power—far beyond current power capacity levels.
- The potential demand for AI power generation could have a significant positive impact on natural-gas-focused companies in the exploration and production (E&P) industry, as demand could require 10%–15% supply growth from today’s levels.
- Procuring real estate for data centers is impacted by increases in construction costs, development margins and prices for developable land.
- Smaller, more efficient AI models and advancements in nuclear fusion may provide alternative solutions to meeting existing power challenges.
Stephen Dover: With the recent explosion of AI, the implications for data center growth are significant. How has this space been evolving?
Blair Schmicker: Data centers are a particularly unique category of commercial real estate. Consistent with the inexorable march of technology and our insatiable demand for compute, data center capacity had expanded at a consistent double-digit rate in the years leading up to this decade. In late 2022, demand growth further accelerated and eventually outstripped the capacity of its lifeblood: power. This was first felt in Northern Virginia, the largest market for data centers in North America. Case in point, local utility provider Dominion stopped expansion plans until they could devise a way to alleviate the emerging power shortage they were experiencing.1 According to Dominion, widescale adoption of remote work during the pandemic doubled new data center connections in 2022 and drove prices higher. This challenge quickly spread to other major data hubs as demand began to broadly outstrip supply. All of this happened before OpenAI released ChatGPT, which, as you alluded to in your question, landed like a lit match on a pile of gunpowder.2
Stephen Dover: What’s behind the growth in demand for electric power, and who are the key players?
Matt Cioppa: Several well-known mega-cap technology companies are leading the charge in building both the large language models that power the generative AI movement and the cloud infrastructure on which both startups and Fortune 500 companies are building and consuming AI models and applications. Four in particular—Microsoft, Meta, Alphabet and Amazon—have dramatically increased forecasted capital expenditures (capex) to capitalize on the opportunity ahead of them. The figures are staggering. Combined, expected capex from the "Big Four" is estimated at around $190 billion in 2024, up almost 50% from 2023.3 To state it another way, this is more than the other 90 technology and communication services companies in the S&P 500 Index, combined. Those capex dollars will flow in several directions, benefiting providers of advanced semiconductors, servers, networking gear, data storage and, of course, the rising cost of powering and cooling data centers.
Microsoft and OpenAI are reportedly working on plans for an AI supercomputer that could cost more than US$100 billion—100 times as much as a standard data center—with the hope of launching the first phase by 2028. Estimates suggest a supercomputer of this size could require somewhere between three and five gigawatts (GW) of power. This is a massive increase from standard data centers today, where large new data centers typically handle 100 to 150 megawatts. If Microsoft's mega-cap peers follow suit, it's clear that this will require a tremendous amount of innovative work across all supply chain participants, as well as cooperation from regulators, to enable this.
Stephen Dover: Do we currently have enough power sources to supply the anticipated demand?
Blair Schmicker: Vacancy is quickly approaching sub-1% across most markets, with Northern Virginia the lowest at 0.67%. To address this tight supply, roughly 4GW is currently under construction in the United States, though most of it—nearly 80%—is already pre-leased, according to market participants. This isn’t enough to meet long-term demand needs. McKinsey projects demand will grow by about 35GW by the end of the decade, and investor-owned utilities have already announced more than 40GW of potential service requests.4 We expect further advancements in energy innovation to help solve this supply/demand dilemma over time.
Exhibit 1: US Data Center Energy Consumption
2023–2030 (Projected)
Source: McKinsey Energy Solutions Global Energy Perspective 2023; McKinsey datacenter demand model. There is no assurance any forecast, projection or estimate will be realized.
Stephen Dover: Utilities need resources and capital to generate additional power; what does this mean for the energy sector?
Matt Adams: The potential demand for AI power generation could have a significant positive impact on natural-gas-focused companies in the E&P industry. Natural gas is viewed as one of the more accessible and cleaner-burning fossil fuels used as baseload power, while alternative forms of power generation for AI may not be ready for years (e.g., nuclear power) or as reliable (e.g., wind/solar). If the AI power demand estimates turn out to be correct, it could require an additional 10%–15% of natural gas supply on top of current demand in the United States, marking a significant acceleration in natural gas demand, which has been growing at just 1%–2% per year. The AI demand call could also come at a time when the United States is building out its liquefied natural gas (LNG) export capacity by 25%. This would create more customer competition for US domestic natural gas and higher prices.
The combination of these two factors has been a driving force behind the substantially higher-than-normal expectations for US natural gas prices reflected in today’s 2025+ financial futures markets. These prices stand in contrast to today’s lower prices, which reflect excess supply and relatively full inventory storage levels. The caveat is that there is more than enough potential supply capacity in the United States to meet this demand, in our view. Higher natural gas futures prices could incentivize E&Ps to bring back more natural-gas-focused drilling and supply to feed growing demand from AI.
Stephen Dover: None of this comes without a cost; what are some of the key challenges companies are facing?
Blair Schmicker: From a real estate perspective, all this demand is driving an increase in construction costs, an increase in development margins and, thus, an increase in the marginal price of new/existing supply. Several large owners of industrial real estate are reconsidering the highest and best use case for developable land and either selling to or partnering with data center developers. Data centers and their development are not easily accessible within private investment markets. Because of these dynamics, we perceive a very powerful global cycle that appears poised to last for several years.
Matt Adams: One of the biggest challenges we see is transportation, as most proposed trans-state natural gas pipelines have been blocked under the Biden Administration due to environmental and political pushback against fossil fuels. The two major sources of natural gas come from Appalachia (PA, WV) and the Haynesville Shale (LA); both would require new pipelines to reach potential AI demand centers in other states. This may cause a push to locate AI data centers in natural-gas-producing regions and reduce transportation risk and time. For example, in West Texas and the Permian Basin, natural gas is a byproduct of crude oil drilling and trades at very low prices. I imagine this could have additional effects on the land development initiatives Blair highlights.
Matt Cioppa: The increasing demand for power also calls into question whether mega-cap tech companies can fulfill their sustainability commitments without stifling AI progress. Nuclear power is also an option. Amazon recently purchased a Pennsylvania-based nuclear-powered data center from Talen Energy, while others are pursuing power purchase agreements (PPAs) with nuclear utilities.5 However, there are clear constraints to large-scale nuclear development that can limit this as a solution. It is no surprise that many leaders in the AI field are calling for—and investing their own money in—a faster path to nuclear fusion, a technology that's still in its infancy but holds tremendous potential.
Stephen Dover: As you said, nuclear fusion is still in its infancy, though it was initially discovered nearly a century ago. Meanwhile, multiple technology developments in recent years have felt like they move at the speed of light. Are other potential solutions in the works?
Matt Cioppa: One potential positive side effect from this power "dilemma" is that it can incentivize more rapid innovation around building smaller, or more efficient, AI models. We already see smaller models becoming much more capable. Several can successfully run on a smartphone and take far less computing capacity and power to train. New breakthroughs in model architecture may also eventually make it possible for model builders to dramatically reduce energy requirements without compromising on scale or capability. One example would be Microsoft Research's proposed 1-bit Transformer architecture (dubbed "BitNet"), which in theory can dramatically reduce a large language model's energy consumption by reducing the precision of model parameters, without materially degrading output quality. While that sounds very encouraging, it's important to emphasize that this and other similar efforts are still in their research phase.
Stephen Dover: The early optimism greeting the arrival of AI capabilities remains strong, but investors are already beginning to think carefully about the daunting challenges of fueling this technology. The jump in electricity supply needed to operate new AI data centers is well above any recent demand on US utility operators, which were already pushed to their capacity limits in key regions of the United States. Mega-cap technology companies are beginning to allocate staggering amounts of capital investment to support AI—investment that will likely stimulate demand for a range of industries, including semiconductors, servers, networking gear and data storage. It will also have secondary effects, driving demand for real estate where new data centers are established. Blair Schmicker, Matt Cioppa, and Matt Adams and their research teams at Franklin Equity Group are fully engaged and thinking ahead about what companies need to do to make AI visions a reality. We believe this deep understanding and the ongoing contact with companies who are grappling with these challenges will be the keys to understanding the potential returns and risks of AI technology. As we have tried to demonstrate, it may be equally as important to be able to formulate new questions as this historic transformation unfolds.
Endnotes
- Dominion Energy, Inc. is a Virginia-based corporation that produces and distributes electricity and natural gas to homes, businesses and wholesale customers in the US.
- Open AI is an AI research and deployment company that produces services including ChatGPT.
- Microsoft, Inc. develops, manufactures, licenses, supports and sells computer software, consumer electronics, personal computers and related services. Meta Platforms, Inc. operates as a social technology company that builds applications and technologies that help people connect, find communities and grow businesses. Alphabet. Inc. is an American multinational technology conglomerate holding company and the parent company of Google. Amazon is an American multinational company engaged in e-commerce, cloud computing, online advertising, digital streaming and artificial intelligence (AI).
- There is no assurance that any estimate, forecast or projection will be realized.
- Talen Energy, Inc. is a leading producer and energy infrastructure company with a focus on supplying power to technology companies looking for carbon-free energy sources to develop their artificial intelligence capabilities.
WHAT ARE THE RISKS?
All investments involve risks, including possible loss of principal.
Equity securities are subject to price fluctuation and possible loss of principal.
Investment strategies which incorporate the identification of thematic investment opportunities, and their performance, may be negatively impacted if the investment manager does not correctly identify such opportunities or if the theme develops in an unexpected manner. Focusing investments in information technology (IT) and technology-related industries, carries much greater risks of adverse developments and price movements in such industries than a strategy that invests in a wider variety of industries.
Any companies and/or case studies referenced herein are used solely for illustrative purposes; any investment may or may not be currently held by any portfolio advised by Franklin Templeton. The information provided is not a recommendation or individual investment advice for any particular security, strategy, or investment product and is not an indication of the trading intent of any Franklin Templeton managed portfolio.