Artificial intelligence is poised to become the dominant force shaping the future of global infrastructure, but its exponential rise is already straining the physical systems needed to support it. A new report from Deloitte warns that power demand from AI data centres could increase more than 30 times by 2035, potentially overwhelming grid capacity and exposing deep systemic vulnerabilities unless cross-industry coordination accelerates.
The report, Can U.S. Infrastructure Keep Up With the U.S. Economy, draws on new survey data from 120 data centre and power company executives and paints a sobering picture. In 2024, AI accounted for just 12 per cent of total data centre electricity use in the US. By 2035, that share is expected to rise to 70 per cent, requiring a total of 123 GW, equivalent to powering over 90 million homes.
Such figures mark a turning point for AI infrastructure. What was once a software challenge is now a hardware and energy crisis. As Deloitte principal Martin Stansbury puts it, “There is an opportunity in infrastructure development to support the national strategic priorities of AI and energy dominance. However, it is a complex undertaking… Collaboration will be critical.”
From chipsets to grid stress
While most of the public conversation around AI centres on algorithms and models, the physical cost of deploying those systems at scale is becoming increasingly difficult to ignore. Hyperscale AI data centres now consume hundreds of megawatts per site. Some in early development are aiming for gigawatt-scale footprints, comparable to entire power stations.
As a result, mentions of data centre in manufacturing and energy sector investor calls have increased fivefold year-on-year, Deloitte notes, with total capital expenditure on infrastructure expected to exceed $1 trillion for both utilities and hyperscalers within the next five years.
But building the AI future is proving more difficult than financing it. More than 70 per cent of data centre and power executives surveyed by Deloitte cited grid capacity constraints as a major obstacle, alongside supply chain disruption, security, and misaligned construction timelines. The survey also revealed sharp differences in expectations: 92 per cent of data centre operators believe access to power is now a competitive risk, compared with 71 per cent of power companies.
Beyond the electricity itself, other infrastructure elements, from transformers and transmission to chip manufacturing and cooling, are also at risk of becoming chokepoints. Even as AI promises new efficiencies, the urgency of its rollout is stressing the systems that underpin modern economies.
Collaboration or fragmentation
Despite broad agreement that partnership is essential, only 15 per cent of data centre leaders and eight per cent of utility executives describe cross-sector collaboration on infrastructure as “highly effective.” This mismatch suggests that while intent is strong, execution remains fragmented.
Deloitte identifies three levers for bridging the gap: technology, regulation and funding. Data centre executives prioritised increasing energy efficiency, deploying intelligent infrastructure, and integrating renewables. Power companies, meanwhile, face the additional challenge of shielding consumers from rate increases while supporting faster capacity expansion.
A proposed framework in the report calls for greater transparency in infrastructure planning, removal of speculative development proposals, and fast-tracking of priority projects. But it also acknowledges that regulation must evolve in parallel with technological innovation to prevent AI adoption from outpacing the systems that sustain it.
The report concludes that unless infrastructure development becomes as strategic as AI development itself, the industry risks hitting a physical ceiling. As AI moves from experiment to essential service, the race is no longer just about building better models, but powering them reliably, sustainably and at scale.
