Why AI infrastructure constraints could make the technology more energy efficient over time

A global survey by Digital Reality cited in a recent ESG News opinion piece found that AI is advancing at a pace that raises fundamental questions about whether that growth is sustainable from an energy and infrastructure standpoint. The argument put forward is counterintuitive: the very constraints slowing AI expansion, including power availability, data centre capacity, and cooling requirements, may act as market signals that accelerate efficiency improvements. Scarcity, in this reading, is not an obstacle to sustainable AI but a mechanism that shapes it.

The energy intensity of AI workloads is already a material concern for corporate sustainability teams. Data centres running large language models and inference workloads consume significant amounts of electricity, and for companies with Scope 2 emissions targets, the growth of AI in their operations creates a direct reporting liability. The pressure on power grids is particularly acute in markets where renewable energy capacity cannot keep pace with data centre demand, creating a situation where AI growth and emissions reduction targets are in direct tension.

For ESG managers and technology procurement leads, the practical implication is that AI infrastructure decisions are now sustainability decisions. Choosing where to host AI workloads, which providers to use, and how to account for the associated energy consumption within a GHG Protocol Scope 2 or Scope 3 inventory is a question that compliance teams are beginning to encounter. The opinion argues that hardware and energy constraints will force AI developers to design more efficient systems, but that efficiency dividend is not automatic and will depend on regulatory and market incentives to materialise.

The broader sustainability case rests on a bet that innovation under constraint produces better outcomes than unconstrained growth. Historically, energy efficiency improvements in computing have followed this pattern, with limitations on processing power and heat dissipation driving architectural innovations that reduced consumption per unit of output. Whether AI follows the same trajectory at the speed required to align with corporate and national climate targets is an open question that sustainability professionals will need to monitor closely.

For CSRD reporters, AI energy consumption is becoming a line item that cannot be deferred. Companies disclosing under ESRS E1 will face questions about how their technology infrastructure contributes to their overall energy consumption and greenhouse gas emissions profile. As AI adoption accelerates across finance, logistics, and professional services, the materiality of data centre emissions for non technology companies is rising, making the efficiency trajectory of AI infrastructure a directly relevant variable for sustainability reporting teams across sectors.