Powering AI’s Growth: The Critical Infrastructure That’s Needed

Powering AI’s Growth: The Critical Infrastructure That’s Needed

By Pooja Goyal , Chief Investment Officer, Carlyle Infrastructure Group


Welcome to Up Close with Carlyle, where we look at the investment landscape from a different point of view in each edition by sharing insights from executives across our firm. Subscribe here to be notified of future editions.


Excitement, enthusiasm, and high expectations surround the future of artificial intelligence. Getting to that future, however, will require significant investments in physical infrastructure to power and deliver AI’s potential. At Carlyle, we believe this may present opportunities for infrastructure investments.

At the most basic level, the growth of AI will require two infrastructure components: data centers, which consist of huge computer installations and the structures to house them, and enormous quantities of electricity and water to power and cool the computers. Rough estimates suggest the investment to build the data center structures that are expected to be required over the next ten years could total $800 billion to $900 billion[1]. Access to power is probably the biggest bottleneck to building out AI infrastructure since demand could reach ten or twenty gigawatts, or more, over the next decade[2].

The electrical generation capacity that will be needed for AI needs could well be on the level of the World War II Manhattan Project, whose Oak Ridge, Tennessee, facility that enriched uranium for the atomic bomb consumed one-seventh of the electricity produced in the United States from 1943 to 1945[3]. Existing data centers, which were built near major metropolitan areas to serve internet users, do not require anywhere near the volume of electricity that AI data centers consume.

Utilities serving current data centers do not have the capacity currently to serve new AI centers without power being generated elsewhere and delivered via a forest of new, high-capacity transmission lines—which local communities would no doubt rail against. What’s more, the current utility regulatory environment, which was designed to ensure reliable delivery for all users, hampers AI data centers from relying on local utilities, which find it difficult to give data centers the assurance of delivery and certainty of long-term prices that are necessary for data center viability. In addition, data centers would like to be able to sell power they may generate back to utilities, but regulation currently makes those interconnections difficult as well.

To meet demand in the short run, we are likely to see arrangements worked out with data center developers, utilities, and power infrastructure owners to provide the necessary power. The ongoing participation of government at all levels also will be necessary to create the appropriate regulatory backdrop for these changes. But because the need for large and preferably green volumes of reliable power will grow dramatically, most new AI data centers are likely to be built in large areas away from current facilities and near sources of energy including solar, hydroelectric, and nuclear power. Many will be constructed in combination with their own co-located power-generation facilities.

Carlyle is aiming to help address the demand that AI will place on the nation’s power supply by drawing on its more than 15 years of investing experience in infrastructure and energy spanning all major sectors—renewable energy, thermal power, upstream, midstream and downstream oil & gas, digital infrastructure, and transportation. In 2021, Carlyle created Copia Power , a wholly owned portfolio company focused on developing utility-scale sustainable infrastructure in the United States. With an investment of about $2 billion, Copia currently is developing a 20,000-acre site west of Phoenix, Arizona, intended for use by data centers[4]. The facility has a 500-kilovolt interconnection with the grid as well as 1.5 gigawatts of electrical generation capability from solar panels, and batteries for storage[5].

With AI growth dependent on the availability of power for the massive data centers that will be required, Carlyle intends to continue partnering with established industry leaders with highly specialized skillsets to develop, own, and operate sustainable infrastructure assets. We believe the needs around AI and the need for data centers presents opportunities worth monitoring.


Carlyle Global Infrastructure is an?integrated platform which brings together the firm’s substantial resources, scale and capabilities to further invest behind the energy transition and capture significant opportunities around the world.?Our team invests in long-term projects spanning the transportation, energy, digital, water, renewables and power sectors globally. We believe our investments fund essential assets that help meet the needs of everyday life and support economic growth and productivity.

If you enjoyed Up Close with Carlyle, follow us for more insights and thought leadership. We also welcome your comments and thoughts below.


[1] Laughlin, Morgan and Footh, Jim. February 22, 2024. “Data Center Growth Abounds in the Digital Age.” PGIM Real Estate.

[2] Patel, Dylan; Nishball, Daniel; and Ontiveros, Jeremie Eliahou. March 13, 2024. “AI Datacenter Energy Dilemma – Race for AI Datacenter Space.” Semianalysis.com

[3] National Park Service. “Oak Ridge Wayside: Powering the Manhattan ProjectNPS.gov.

[4] Gonzales, Angela. April 12, 2024. “Utah solar developer taps area near Phoenix for $1.2B project “ ABC15, Arizona.

[5] HarquahalaSun.com.

David Geddes

CXO @CogSoft? || Board @ISTA @NSMA @USAe || Advisor @GraphDragon

2 周

You are making a decision to ignore other competing techniques in a large discipline or field. In ten years, businsses will not be using vast energy to crunch numbers. Did expert systems have this problem? Do you know what happened to expert systems? The technique did not die, it evolved. #ai

回复
Endeley Ndemba

Chairman | C-Level Executive | Value Creator @ BRESHPAC-SAS | Driving Growth & Social Impact.

3 周

Pooja's message sets a compelling and forward-looking tone that captures both the necessity and the appeal of infrastructure funding in the context of AI’s continued evolution.

回复
Alejandra Pedernera

IT Recruiter & Talent Acquisition Specialist | HR Professional with 13+ Years of Strategic Recruitment (IT and Generalist) | Expert in Process Optimization and High-Performance Team Development

1 个月

The rapid growth of AI hinges on robust infrastructure investments. As Carlyle highlights, powering data centers with sustainable energy sources is critical. Collaboration among developers, utilities, and regulators will be essential to meet the soaring energy demands and build a resilient framework for the future of AI.

回复

No thanks. This is simply another gravy train Corporation pulling the wool over the weak minded. All I hear is the sound of bleating sheep. Keep on going along with this PR and see where it will take you. The one thing that is certain here. Those going along with this are thinking about nothing more than protecting their so-called professional careers. How sad.

回复
Shalin Govender

Strategic Leader | Expert in High-Tech Business Development & Regulatory Navigation | iGaming Investor

1 个月

Very interesting... Gets me thinking about the African perspective... with cheaper arid land, and huge solar potential. Cyprus is also worth a look given its strategic positioning, sunlight and proximity to the sea for cooling. I see a potential investment opportunity coming.

回复

要查看或添加评论,请登录

社区洞察

其他会员也浏览了