Interview: AI’s energy appetite vs. the innovation..

As artificial intelligence accelerates into the core of the global economy, a new constraint is coming sharply into focus: energy.

At a time when electricity demand in advanced economies had largely plateaued, the rapid expansion of data centers-driven by AI-marks a structural shift. For investors, operators, and policymakers alike, the question is no longer whether demand will grow, but how systems will adapt.

In this conversation, Christoph Frei, Partner and Head of Energy at Emerald, speaks with Simon Bennett, Energy Analyst at the International Energy Agency (IEA), about the agency's latest work on AI and energy. The discussion explores demand projections, infrastructure bottlenecks, efficiency gains, and where the next wave of energy innovation may emerge.

(Original Interview date: September 2025)

The scale of AI-driven demand

Christoph Frei:
Simon, data center energy demand has quickly become one of the hottest topics in energy. You've been closely involved in the IEA's recent report on AI and energy. What are the key demand projections?

Simon Bennett:
You're right this has become a central issue not just in energy, but also in broader economic and political discussions.

We at the IEA felt there was a need for a more common understanding of how much energy data centers actually use today, how that might evolve with artificial intelligence, and how we can meet that demand. That's what led to the report we published on Energy and AI.

Now, projections in this space are inherently uncertain, so we developed a central case alongside a range of alternative scenarios. But in our base case, we estimate that global electricity demand from data centers could increase by around 500 terawatt hours by 2030, largely driven by AI.

In global terms, that's not enormous. But what really matters is where and how that demand shows up.

Why it matters: concentration, scale, and permanence

Christoph Frei:
Because it's not evenly distributed.

Simon Bennett:
Exactly. Data centers are highly concentrated geographically. In some regions, they already account for more than 20% of electricity demand.

At the same time, the scale of individual facilities is increasing dramatically. Some of the largest data centers under construction today will consume as much electricity as two million households.

And unlike many other loads, they operate continuously. They are always on. What this means is that, in advanced economies where electricity demand had been flat or even declining, we are now seeing a reversal. Data centers are becoming a major driver of demand growth.

A structural shift in the US energy system

Christoph Frei:
Your report also highlights the US specifically. The numbers there are striking.

Simon Bennett:
They are. In our base case, around half of electricity demand growth in the United States through 2030 could come from data centers. One statistic that really captures the scale is that data centers in the United States are on track to consume more electricity than the combined US demand of heavy industries like chemicals, steel, and aluminum.

That's a profound shift. For decades, we've thought of industry as the dominant electricity user. Now, digital infrastructure is taking that role.

Bigger, faster, more concentrated by design

Christoph Frei:
What's striking is not just the scale, but also the concentration. From an energy system perspective, we've typically tried to avoid clustering critical infrastructure. Yet data centers are becoming larger and more centralized. Isn't that a concern?

Simon Bennett:
There are two sides to that: technical and economic. From a technical standpoint, larger data centers benefit from economies of scale. You can deploy capital more efficiently, achieve higher operational efficiencies, and optimize system performance. There's also the issue of latency - being close to demand improves performance, which creates an incentive to cluster capacity near population and economic centers.

But you're right to point out the potential risks. From a system perspective, there are questions around resilience, grid constraints, water use, and even geopolitics. Whether we see a shift toward more distributed data center architectures will depend largely on policy and incentives. The underlying technologies favor scale-but that doesn't mean the system has to evolve that way.

Efficiency: the underestimated lever

Christoph Frei:
Historically, in highly innovative sectors, efficiency gains tend to surprise on the upside. Are you seeing that in data centers as well?

Simon Bennett:
Yes, and it's a crucial part of the story. Efficiency improvements in data centers have been ongoing for years, particularly through advances in chip design and server architecture. But we may now be entering a new phase. In the past, the priority was speed - building capacity as quickly as possible to meet demand. Now, constraints around electricity availability are becoming more binding, and that is shifting incentives.

We expect to see innovation not just in computing efficiency, but also in cooling technologies, system design, and operational flexibility. Some of these technologies-particularly in cooling-are already starting to scale. Others are still emerging, but the direction of travel is clear.

Scenarios: how wide is the range?

Christoph Frei:
Your report includes multiple scenarios. How much variation are we really talking about?

Simon Bennett:
Quite a lot. In our "liftoff" scenario, global data center electricity demand could be an additional 500 terawatt hours higher than the base case by 2035. That's a very significant increase. On the other hand, there are scenarios where demand is lower, driven by greater efficiency gains or different trajectories in AI adoption.

It's important to note that efficiency is only one part of the equation. The evolution of AI models, their usage, and the scale of deployment will also play a major role. So while we've tried to be balanced in our central case, there is substantial uncertainty in both directions.

The regulatory dilemma: growth vs. affordability

Christoph Frei:
Let's turn to policy. In some regions, we're already seeing electricity prices rise. Should regulators accept this as the cost of digital growth?

Simon Bennett:
It's a complex issue. Price increases are not solely driven by data centers, but they are certainly a contributing factor in some regions. From our perspective, the goal should be to avoid situations where data center expansion leads to significant price increases for consumers. That requires better coordination between data center development and power system planning-particularly in ensuring that new demand is matched with new generation capacity. There are also opportunities to make data centers more grid-friendly, for example by increasing flexibility or aligning demand with renewable generation.

The regulatory frameworks we've traditionally used were not designed for this kind of demand growth. So there is a need to adapt.

AI as part of the solution

Christoph Frei:
At the same time, AI itself could help solve energy system challenges.

Simon Bennett:
Absolutely, and this is often overlooked. AI has the potential to optimize grid operations, improve industrial efficiency, accelerate energy innovation, and enhance system reliability. So the question is not simply whether AI increases energy demand, but whether it can also enable a more efficient and sustainable energy system overall. The answer is likely yes, but it depends on how we manage its deployment.

Where innovation is happening now

Christoph Frei:
Beyond data centers, where do you see the most exciting developments in energy innovation?

Simon Bennett:
One of the key findings from our State of Energy Innovation report is just how broad the landscape is right now. There's significant progress in areas like next-generation geothermal energy, where we've seen projections of substantial cost reductions, though those still need to be proven at scale.

But some of the most interesting developments are happening at the intersection of sectors. For example:

• Electrification of heating at both residential and industrial scale
• Thermal energy storage
• Advances in cooling technologies, including solid-state cooling

Cooling, in particular, is underappreciated, but it's becoming increasingly important as both data center demand and global temperatures rise.

Fusion, geothermal, and the long-term horizon

Christoph Frei:
And what about frontier technologies like fusion?

Simon Bennett:
Fusion has long been seen as perpetually "20 years away." What's different now is that we may actually be entering that window. There has been a significant increase in private investment, and some companies are targeting pilot plants and even grid-connected systems in the 2030s. Not all of those ambitions will be realized. But the level of confidence and capital behind fusion today is very different from the past. If the engineering challenges can be solved, fusion could offer a clean, widely available energy source with minimal waste and no geographic constraints.

Geopolitics reshaping innovation

Christoph Frei:
How is geopolitics influencing technology development?

Simon Bennett:
We're seeing a strong impact, particularly around critical minerals for batteries.

This is driving innovation in three main areas:

1. Alternative battery chemistries that reduce reliance on scarce materials
2. New extraction technologies that diversify supply
3. Recycling, which is receiving significant attention and funding

For example, extracting lithium from geothermal brines is gaining traction-something that might not have been commercially viable in a less constrained geopolitical environment. At the same time, international cooperation remains essential. When countries collaborate on technology development, they accelerate innovation and improve outcomes for everyone.

The "valley of death" for startups

Christoph Frei:
One concern often raised is whether innovation can scale. Where do you see the biggest risks?

Simon Bennett:
One of the biggest risks is the gap between early-stage innovation and commercialization-the so-called "valley of death." We've seen a surge in venture capital investment in clean energy startups in recent years, thousands of companies working on promising solutions. But that funding has declined recently, and it hasn't yet been fully replaced by public or alternative sources. Over 1 800 energy start-ups raised equity in the peak years of 2021 and 2022, and another 1 400 since then. If those companies cannot scale, we risk losing a significant portion of the innovation pipeline.

What success looks like

Christoph Frei:
If we look ahead five to ten years, what would success look like?

Simon Bennett:
I would highlight two areas. First, large-scale decarbonization challenges, particularly in heavy industry and long-distance transport. These are areas where we still lack clear commercial pathways, and where public-private partnerships will be essential. Second, ensuring that smaller, modular technologies, many of which are highly promising, have a clear route to market. If we can unlock both ends of that spectrum, we will have made real progress.

A shifting perspective on electrification

Christoph Frei:
Final question: where has your own view changed most in recent years?

Simon Bennett:
I've been surprised, positively, by how quickly electrification solutions have advanced. In areas like high-temperature heat and building heating, we now see more viable pathways than we did even five years ago. That has led to a more balanced view of how we can decarbonize sectors that were previously considered very difficult.

Closing thought

The rise of AI is often framed as an energy problem while cheap electricity in turn is a critical enabler for potential AI leadership. But as Simon Bennett makes clear, AI is equally a catalyst, forcing systems to evolve, accelerating innovation, and exposing both constraints and opportunities. For those operating at the intersection of energy, capital, and technology, the message is clear and the stakes are high: the next decade will not just be about meeting demand. It will be about redesigning the system that supplies it.

More on energy at Emerald:

Fusion at an Inflection Point: what the latest Emerald Energy Sprint revealed

Tech to Solve the Data Center Energy and Resource Dilemma

Reimagining Data Centers: Insights from Emerald's Latest Energy Sprint