From Seattle to Shanghai, Chicago to Cape Town, and seemingly everywhere in between, the construction of data centres is proving to be one of the great energy infrastructure challenges of the 21st century.
But while hyperbolic forecasts of AI use have data centres consuming as much power as entire countries, in reality, total electricity use for the sector is expected to reach just 3% of global consumption by 2030. To put that into context, that’s just a fraction of what will go into air conditioning.
But while this dose of realism helps to contextualise this major societal evolution, immediate challenges remain which the sector must be prepared to confront. Data centres are, indeed, being built at record pace, with a forecast of 6,111 public data centers constructed by the end of 2025 worldwide. In North America, much of this incoming new capacity is clustering in already congested grids and power-hungry AI workloads are stretching local systems to their limits.
AI’s energy demand curve is already bending towards efficiency, with advances in chip design, cooling and computing architecture mitigating initial exponential growth and driving a more linear uptake. But even a slower rise in demand can strain grids already struggling to keep up with the electrification of transport, heating and industry.
The challenges in the grid
There are two clear bottlenecks: semiconductors and electricity. But there are levers that can be pulled.
Built near solar and wind resources, modular, distributed data centres can be scaled incrementally and operated flexibly, with activity ramped up during periods of abundant renewable generation and dialled back during peak grid stress. The data centre would, in effect, become an adaptive load, integrated with the rhythm of local energy production rather than fixed to the limitations of centralized grids.
The approach would also diversify risk. In today’s model, hyperscale data centres often depend on large, long-distance transmission lines that take time to permit and build. Modular siting around renewables could circumvent grid bottlenecks while accelerating deployment. It could also make room for new hybrid power models, combining natural gas, carbon capture, battery storage, and renewables, to ensure the almost perfect level of reliability which is demanded by digital infrastructure to function.
While renewables should anchor the buildout of data centres, hybrid energy systems, where local solar or wind generation is complemented by dispatchable gas turbines or small modular reactors, can offer both reliability and a path toward decarbonization. The private sector is already experimenting with this power couple model, pairing thermal generation with renewables behind the meter to secure energy independence.
Of course, there remains scepticism as to whether this hybrid approach is sustainable in the long-term or just a short-term stopgap. Moreover, it remains to be seen if these solutions will remain niche or evolve into a mainstream strategy. Utilities, regulators and investors all have a role in determining that outcome.
The AI era will not collapse under the weight of its own electricity bill, but the way in which data centres are located and powered will shape how equitably and sustainably that era unfolds.
Dr Ali Ghorashi is a senior vice president, energy infrastructure finance and M&A at DNV
Read more: Why Europe’s data privacy framework needs a common blueprint
