By Simone Ragazzi, portfolio manager, Algebris Investments
For all the noise surrounding artificial intelligence, including its breakthroughs, its risks, and its impact on productivity, the conversation still tends to revolve around the digital layer: chips, software models, data. Yet increasingly, the real story is not just about what AI can do but what it demands. And what it demands above all, is power.
Power in the literal sense. Electricity, land, cooling systems, transmission networks. These are now the constraints that define how fast AI can scale. AI is no longer just a digital or software issue, it’s becoming a physical infrastructure challenge, touching everything from the power grid to urban planning.
Data centres are already responsible for about 1.4% of global electricity consumption. That figure could triple by 2035. Forecasts from Bloomberg NEF suggest we need over 360 gigawatts of new power generation to meet AI-driven demand alone, roughly the equivalent of building 350 large power stations.
US tech giants are already on track to spend more than $1.8trn in capex over this decade, not just on computing capacity but on the energy systems that underpin it. In Asia, cloud expansion is accelerating just as fast. But the challenge isn’t only about scaling up. It’s also about reaching the limits of the physical systems needed to support that growth.
Power grids, cooling systems, and land availability are all under pressure. Infrastructure, that was built for a different era of computing, is now being stretched to accommodate entirely new demands, and in many places, it’s struggling to keep up.
See also: ClearBridge’s Langley: Infrastructure in a shifting world order
Efficiency not capacity
So, the industry is adapting. In many cases, it’s no longer just about building more capacity, it’s about using what’s available more efficiently. That’s why efficiency is now being treated as a kind of capacity in its own right. Every watt saved through better cooling or smarter operations becomes a watt that can power more computing.
Liquid cooling, for instance, is starting to move into the mainstream. It’s not a silver bullet, but it works. Compared to traditional air-cooled systems, liquid-based designs reduce power usage and make heat reuse more viable. In some cities, data centre waste heat is already being fed into local district heating systems. At Meta’s Odense facility in Denmark, the servers help warm thousands of homes.
Software also has a role to play. Tools such as digital twins (virtual models of data centre operations) are being used to simulate energy usage and identify inefficiencies in real time. That might sound like a marginal gain, but across large data centre fleets, these optimisations can significantly reduce operating costs and unlock additional computing throughput without touching the physical footprint.
Shift in value chain
What this all means for investors is that the value chain is shifting. The market has understandably focused on semiconductors and cloud platforms, but those aren’t the only stories worth following. There’s a cohort of companies enabling the physical expansion of AI infrastructure — cable manufacturers, power conversion specialists, cooling system designers, and grid connectivity providers — that are increasingly central to the AI economy.
Firms such as Vertiv, Hitachi, Prysmian, Schneider Electric, and Trane Technologies sit in this space. Their role isn’t as visible, but it’s vital. They’re building systems that allow hyperscalers to keep expanding. And they’re starting to be recognised for it. Order books are growing. Procurement pipelines are expanding. And the spending isn’t slowing down.
This is not a distant forecast, it’s already happening. US tech companies have collectively procured over 50 gigawatts of renewable energy, with more on the way. In Southeast Asia, new hyperscale projects are driving grid expansion and clean energy adoption across Malaysia and beyond. The investment case is building around those who can deliver resilience, efficiency, and scale; especially as regulatory pressures mount and sustainability becomes non-negotiable.
See also: GQG: AI boom could be worse than dot-com bubble
The next chapter
AI is not just a matter of software and computing. It is now firmly embedded in questions of resource use, energy strategy, and infrastructure development. The next chapter of the digital economy will be shaped as much by who powers it as by who programs it. And that shift should reframe how we think about opportunity in global markets.
The infrastructure story behind AI isn’t about glamour. But it is about leverage; from a political, financial, and operational perspective. For investors, it’s where some of the most durable upside may lie. Not at the bleeding edge of invention, but in the systems that make it all possible.
