The AI Power Problem: Can Energy Infrastructure Keep Up With the Data Centre Boom?

Global data centre capacity is set to nearly triple by 2030. The real question isn’t whether AI is coming – it’s whether the power to run it will be there. Speaking at a technology and data centre conference in Frankfurt taught me something I hadn’t fully appreciated as an AI advisor: the next phase of the AI revolution will be determined by something much more physical than algorithms. It will be determined by power. We tend to talk about artificial intelligence as a software story – better models, faster chips, autonomous agents. But behind every AI interaction sits a growing physical infrastructure: data centres, GPUs, cooling systems, networks and, above all, electricity. The AI economy may be digital, but its foundations are not. And those foundations are under enormous pressure.

AI is creating an unprecedented demand for compute
McKinsey estimates, reflected in the data presented at the conference, suggest that global demand for data centre capacity could almost triple by 2030. AI is the engine: AI workloads are expected to grow more than three-fold, while all other workloads grow by less than double. In other words, around four in five of the new gigawatts will be built for AI. This is not simply about more people using generative AI tools. The next wave is AI embedded into businesses, software, customer service, coding, research, decision-making and increasingly autonomous agentic workflows. Every agent, image, piece of generated code or automated decision creates demand for compute. And compute requires electricity. The International Energy Agency estimates that global data centre electricity consumption could more than double by 2030, or around 3% of global electricity consumption.

The infrastructure we need does not yet exist
Much of today’s data centre estate was designed for traditional cloud computing, enterprise applications and storage. AI demands much higher power density and more sophisticated cooling. That creates a paradox. AI demand is accelerating rapidly, but the infrastructure to support it has to be built, upgraded or completely redesigned. The question is not simply how quickly companies can adopt AI. It is how quickly the physical world can adapt to support it.

Power is becoming the defining constraint
For years, building a data centre meant thinking first about land, buildings, connectivity and capital. Today, power is increasingly the first question. In Europe’s established markets – Frankfurt, London, Amsterdam, Paris and Dublin, known collectively as FLAP-D – the problem is acute. JLL research shows that hyperscale greenfield sites planned for 2026-2028 are, on average, 175 kilometres from major hub cities, compared with just 46 kilometres for earlier projects. Greenfield developments have risen from 8% to 39% of the pipeline. Data centres are increasingly being built where power and land are available, rather than where customers are located. The grid was not designed for this. In established European hubs, grid connections can take seven to ten years, while building the data centre itself takes roughly 18 to 24 months. The European Commission has identified grid access as a major bottleneck, with connection times of three to ten years and some projects delayed by up to 13 years. Meanwhile, JLL data shows the price of powered land in FLAP-D has risen by around 82% since 2021, from approximately €1.24m to €2.26m per MW. Power is no longer simply an operating cost. Power availability is becoming an asset.

The map of Europe’s AI infrastructure is changing
This is why some future AI infrastructure is moving away from traditional hubs. The Nordics are particularly interesting, combining relatively abundant renewable electricity, cooler climates and growing investment. The emerging model is land + power + connectivity + energy strategy, and the location of the next generation of data centres will be determined by questions such as:

• How much electricity can the site access, and how quickly can it be connected?
• Is that electricity reliable, and what is its carbon intensity?
• Can additional generation be built alongside the data centre?
• Can excess heat be reused, and can batteries provide flexibility?
• Is there room for future expansion and high-density AI infrastructure?

We need to innovate on energy as fast as we innovate on AI
The question is not whether we need more electricity. We clearly do. It is where that electricity will come from and how quickly we can make it available. The IEA expects renewables to provide around half of the additional electricity data centres will require, with nuclear, natural gas, storage and other sources all playing a role. Technology companies are already major participants in the energy market. The IEA reports that the sector accounted for around 40% of corporate renewable power purchase agreements signed in 2025, and it is increasingly supporting nuclear and geothermal projects, including conditional offtake agreements with small modular reactor developers. The data centre of the future may not simply be a large building connected to the grid. It could become an integrated energy and compute system. Battery storage deserves particular attention. AI workloads can create very high and rapidly changing power demands, while renewable generation is variable. The IEA has highlighted the potential for on-site batteries to support AI data centres, particularly where grid connections are constrained. With the right incentives, data centre storage and backup assets could even provide flexibility to the wider grid, turning data centres from large consumers into participants in a more flexible energy system.

Efficiency matters, but it will not solve the problem alone
Efficiency will absolutely help. The IEA notes that power consumption per AI task is falling significantly as hardware and models improve. But when technology becomes cheaper and more efficient, people use more of it. Goldman Sachs estimates that global token consumption could reach approximately 120 quadrillion tokens per month by 2030, around 24 times today’s level, driven by AI agents and enterprise applications. Efficiency can reduce the energy needed for each task while the number of tasks grows exponentially. It may even enable more adoption.

The real AI race may be an energy race
We tend to frame global AI competition around models, chips and software. But underneath it is another race, to secure power, land, grid capacity, cooling, connectivity and capital. It is not an incremental technology upgrade. It is a new industrial build-out. And the implications reach well beyond data centres. Financial services, healthcare, manufacturing, logistics, retail, defence and thousands of emerging businesses will depend on AI-enabled infrastructure. If access to compute is constrained by access to power, energy infrastructure becomes part of the technology strategy of every AI-intensive business. That creates opportunities for entrepreneurs working on generation, storage, grid optimisation, cooling, power management and new infrastructure models, and for investors willing to finance the physical foundations of the digital economy. It also creates an urgent challenge for governments and grid operators.

The infrastructure challenge of the AI decade
AI does not run in the cloud. It runs in physical data centres. Data centres do not run on algorithms. They run on electricity. And electricity does not simply appear when demand arrives. It requires generation, grids, storage, investment and planning, and increasingly sophisticated coordination between all of them. The technology industry has shown it can innovate at extraordinary speed. Now the energy and infrastructure industries need to match that pace. The AI revolution may have started in software, but its next chapter will be written in steel, silicon, batteries, turbines, transmission lines and data centres. The question is no longer whether AI will transform the economy. It is whether we can build the physical infrastructure fast enough to power that transformation.

Data points in this article draw on research from the International Energy Agency, JLL, McKinsey and the European Commission’s 2026 Cloud & AI infrastructure study.