Energy Experts Tackle AI's Voracious Appetite for Electricity

Adrian Anderson, Senior Vice President of Global Energy at Equinix, has procured more than 55 gigawatts of carbon-free electricity—enough to power tens of millions of homes. Miranda Ballentine, a Distinguished Industry Fellow at Dartmouth's Irving Institute for Energy and Society, has reshaped how corporations and governments think about energy. Together, the two made for a compelling pair at an Irving Institute for Energy and Society fireside chat, held as part of Tuck's Revers Center for Energy, Sustainability, and Innovation's 10th Anniversary celebration: The Future of the Energy Industry Conference.

Sam Brigham, a Tuck '26 student and Revers Center Fellow, opened the session by thanking the Class of 1972 for their generosity in making the speaker series possible. The conversation drew an audience of students, faculty, Tuck alumni, and energy professionals eager to unpack one of the most consequential questions facing the modern economy: What does the artificial intelligence revolution mean for the future of energy—and what does energy mean for the future of AI?

Ballentine, whose career spans the founding of the Clean Energy Buyers Association, a tenure as Assistant Secretary of the Air Force for Installations, Environment, and Energy, and a role as Director of Sustainability at Walmart, got straight to the point. She noted that corporate energy buyers—companies like Walmart, Microsoft, Google, Amazon, and Apple—have been reshaping the energy landscape for over a decade, driving the adoption of carbon-free electricity long before AI entered the popular lexicon.

Anderson, who leads the Global Power team at Equinix and previously held energy innovation roles at Apple, Amazon, Google, and Microsoft, brought a front-row perspective to the discussion. His resume reads like a who's who of the hyperscaler energy revolution, including involvement in landmark deals such as the Constellation/Microsoft Crane Clean Energy transaction, the Brookfield 10-gigawatt Clean Energy Collaboration, and a ~12-gigawatt supply chain initiative with Hanwha Qcells.

AI and Energy: How Big Is the Demand, Really?

The conversation quickly turned to the question on every utility executive's mind: just how much energy does AI actually require?

"According to Rhodium, AI accounts for somewhere between 40 and 80 percent of current demand growth," Ballentine noted, adding that electrification and industrial reshoring are also significant contributors.

Anderson offered important nuance, pushing back gently on the projections circulating in the industry. "McKinsey projected something like 300 gigawatts of AI load globally," Anderson said. "Today we're at 40 to 60 gigawatts, and that includes cloud deployment." He acknowledged that projections for electricity demand have been climbing relentlessly—from 12 gigawatts to 18 gigawatts to 90 gigawatts by 2030—but cautioned that the timeline assumptions embedded in many of those figures are unrealistic. "There is not enough concrete, not enough labor, not enough grid interconnection," he said.

He also offered a taxonomy of data centers that he argued is critical to understanding the real shape of the challenge ahead:

1. Massive training centers (500 MW+): Located anywhere, they house large-scale GPUs for AI model training and don't require the same reliability as other facilities. Microsoft's large-scale GPU training center was among the first of its kind.
2. Inferencing data centers: Regionally located, metro-area facilities requiring high reliability and low latency. Smaller, but critical to real-time AI applications.
3. Retail data centers: The workhorses of the internet, average around 35 MW, and carry a current majority of internet traffic and keep everything from Netflix to 911 emergency services running.

"Most of the hype is about what's being built by 2030 or 2035," Anderson said. "It's all happening now, but 1.2 gigawatts took years to build. AI growth will happen. We're just looking at unrealistic timelines."

Bring Your Own Everything: New Models for a Strained Grid

The conversation also explored emerging frameworks that hyperscalers and policymakers are testing to work around the constraints of an overtaxed grid, including Bring Your Own Capacity (BYOC), Bring Your Own Generation (BYOG), and Bring Your Own Transmission (BYOT).

Anderson was direct in his assessments. BYOC, where data center operators bring their own power generation capacity, is a bad idea in his view. "Fixed costs stay the same. There's a whole host of issues. You have to go slow to go fast," he said. 

The concerns go beyond economics. An interconnected grid offers resilience benefits that isolated generation simply cannot replicate, including the ability to balance supply and demand across regions, absorb shocks from extreme weather events, and maintain reliability when individual generators fail. Fragmenting that system through large-scale behind-the-meter generation risks undermining the shared resilience that all customers depend upon.

BYOT, however, is a different story entirely. "Bring your own transmission is a great idea," Anderson said. He pointed to a pioneering partnership between Google and CTC Global that uses advanced conductor technologies to increase transmission capacity in as little as 18 months—compared to the seven or more years typically required for new transmission builds. In markets like the UK, Ireland, and Spain, BYOT is already a requirement.

"We don't have a generation issue," Anderson argued. "We have a time-of-use and transmission issue. We need more interconnection, more transparency, and more real-time data."

He also addressed the growing chaos in interconnection queues—a pain point familiar to anyone working in grid development. In Ohio, he noted, a queue that stood at 6 gigawatts ballooned to 30 gigawatts in months. "Of those folks, 90 percent are not real," he said, advocating for higher capital thresholds and a shift to cluster or batch studies to bring order to the process.

Crucially, Anderson argued that hyperscalers are not adversaries of the grid or ratepayers–they want to be partners. "We understand our load ramp. We will help pay for infrastructure. We can bring capital that doesn't impact ratepayers. Companies can build out a substation, get interconnection, and transfer it to the utility."

AI for Energy: The Other Side of the Equation

Ballentine steered the conversation toward a dimension often overlooked in the public debate: not just energy for AI, but AI for energy.

Anderson highlighted several compelling examples from his time at major hyperscalers:

• At Google, machine learning applied to energy trading delivered a 27 percent improvement in returns.
• At Microsoft, an AI-powered hackathon reduced the time required to complete a Nuclear Regulatory Commission application from six to nine months down to six hours.
• Companies like Kevala are using granular, real-time data, down to the household level, to enable sophisticated grid modeling and scenario planning that was previously impossible.

"AI can help us make load processes and grid modeling dramatically more efficient," Anderson said. "There is so much efficiency to be gained. Real-time data allows real-time reactions."

The group also took up the challenge of forecasting, which is a perennial frustration for utilities and hyperscalers alike. Anderson was candid: "We will never fix forecasting. Customers drive load growth. We don't control customer behavior." But he argued that AI-enhanced real-time data could dramatically improve the ability of grid operators and utilities to respond dynamically, even if perfect prediction remains out of reach.

Carbon-Free Electricity: Green Hushing and the Race Against Time

Ballentine raised the tension between the industry's sustainability commitments and the relentless "speed to power" demands of the AI buildout.

"Carbon-free electricity by 2030 was always a moonshot," Anderson acknowledged. "I've procured more than 55 gigawatts of carbon-free electricity, including the first nuclear and fusion deals. But now we're in a new cycle. The shift has gone from sustainability first, to time to power first."

Natural gas, he said, will be a central fuel in the near term. But he pushed back against the narrative that sustainability has been abandoned. "Companies are still doing incredible work. It's just that you aren't supposed to say it out loud right now."

Ballentine gave that phenomenon a name: green hushing—the active choice by companies to avoid publicizing their environmental initiatives out of fear of reputational risk in the current political climate. She distinguished it from the now-outdated concern about greenwashing. "Green hesitancy is about worrying about negative reputational risk with advancing environmental initiatives," she said. "It's always been a reputational risk to address a global challenge."

A Decade of Progress, and the Road Ahead

As the session drew to a close, the throughline of the conversation was clear: the energy transition and the AI revolution are not separate stories. They are the same story, playing out simultaneously on the same constrained infrastructure, with the same shortage of capital, labor, and transmission capacity, and with enormous consequences for the climate, the economy, and society.

For the future energy leaders, the message from two of the industry's most experienced voices was both sobering and energizing: the problems are real, the timelines are brutal, but the tools, the capital, and the ingenuity exist to solve them.

"We are digitizing no matter what," Anderson said, summing up the session.

The Arthur L. Irving Institute for Energy and Society's Class of 1972 Speaker Series fireside chat was held as part of Tuck's Revers Center for Energy, Sustainability, and Innovation's 10th Anniversary Conference, "The Future of the Energy Industry," at Dartmouth College.

Watch a recording of the Fireside Chat on AI & Energy