The AI energy crisis: When innovation outpaces infrastructure

Artificial intelligence is reshaping economies at remarkable speed—and straining the energy systems that power it. As AI moves from niche to ubiquitous, embedded in everything from virtual assistants to industrial automation, demand for high-performance computing and the data centres behind it is surging. Global data centre electricity use is expected to rise sharply this decade, raising questions about grid capacity, environmental sustainability, and whether current policy settings are fit for purpose. This article examines the scale of AI’s energy appetite, how Australia’s National Electricity Market (NEM) can adapt, and what other countries are doing to manage the wave.

Why AI is pushing power systems to the brink

Generative AI and other compute-heavy workloads require densely packed servers, accelerators, and round-the-clock availability. That translates into concentrated, continuous electricity demand and significant cooling needs. As adoption spreads, more training, fine-tuning, and inference at scale will drive up consumption in both new and existing facilities.

The capital requirements to meet this growth are vast, with global investment in data infrastructure, generation, and transmission expected to climb into the trillions over the decade. Without coordinated planning, the result could be higher energy prices, grid congestion, and delayed decarbonisation—costs that would ultimately spill over to consumers and businesses.

Australia: Can the NEM keep up?

Data centre development in Australia is accelerating, particularly in finance, telecommunications, and cloud services. These facilities require exceptionally reliable power and low-latency connectivity, concentrating demand in urban load pockets. Estimates suggest data centres already account for roughly 4–5% of national electricity consumption and could reach around 8% by 2030.

The immediate challenges are threefold: grid connection timelines, local network constraints, and the need for firmed, low-emissions supply. While Australia is rapidly adding renewable capacity, transmission expansion and storage integration must keep pace—especially where large compute campuses cluster on metropolitan feeders. Fragmented state–federal responsibilities and voluntary guidance on AI use add further uncertainty for long-term investment planning.

What would help:

• Mandatory energy performance standards for data centres, including power usage effectiveness (PUE) targets and transparent emissions reporting.
• Clear, enforceable pathways for grid connections, with streamlined approvals tied to demonstrated contributions to system security (e.g., on-site storage, demand flexibility).
• Better integration of projected AI loads into national transmission planning and Renewable Energy Zone development.
• Incentives for on-site or contracted renewable generation, waste-heat recovery, water-efficient cooling, and load shifting.
• Cost-reflective tariffs and locational signals to steer new capacity toward grid-strength hubs.

The United States: Investment, standards, and uncertainty

The United States is combining investment, regulation, and research to manage AI-driven demand. Federal programs have directed funding toward grid modernisation, clean energy build-out, and semiconductor manufacturing central to AI hardware. Agencies are also assessing data centre efficiency and considering performance standards as workloads grow.

However, policy continuity remains uncertain, and changes in administrative priorities could reshape timelines for clean energy incentives or efficiency requirements. As a global AI leader, shifts in the U.S. policy landscape will reverberate through supply chains, project pipelines, and technology deployment worldwide.

China: Aligning compute with renewables

China has repositioned its digital infrastructure strategy as AI becomes a national priority. Under the “Eastern Data, Western Computing” approach, energy-intensive computing is encouraged in western provinces with stronger renewable resources and available land, easing pressure on eastern urban grids.

New facilities face strict efficiency and carbon-intensity thresholds, with approvals contingent on meeting those benchmarks. The model aims to balance regional development, improve grid stability, and advance long-term climate targets—peaking emissions before 2030 and achieving carbon neutrality by 2060—while still supporting rapid growth in digital services.

Ireland: A stress test for grid and policy

Ireland offers a real-time glimpse of what happens when compute demand outruns infrastructure. In 2024, data centres accounted for a substantial share of national electricity consumption, prompting concerns over prices, grid reliability, and progress toward climate goals.

Regulators have proposed requiring data centres to match their load with equivalent new power supply—spanning generation, transmission, and storage—effectively making approvals contingent on bringing new capacity to the system. This shifts the cost and delivery risk of expansion from ratepayers to developers and aligns growth with system capability.

The emerging playbook—and what’s next

Across jurisdictions, a common pattern is emerging:

• Shift infrastructure costs to developers through connection conditions and capacity-contribution requirements.
• Set minimum efficiency and emissions standards for new facilities, verified through continuous reporting.
• Link approvals to renewable procurement, storage, and demand flexibility that support grid stability.
• Locate large compute clusters where transmission is strong and expandable, and where cooling water and land can be used responsibly.

For Australia, the priority is to move from fragmented, voluntary guidance to a coherent, enforceable framework. That means mandatory disclosures on energy use and emissions, accelerated transmission and storage planning that explicitly models AI load growth, and approval pathways that reward projects contributing firming, flexibility, and decarbonisation.

AI’s energy demands are no longer a distant concern; they’re a defining constraint on how fast—and how sustainably—the technology can scale. Keeping the energy transition on track in an AI-driven economy will require policy to evolve as quickly as innovation, aligning compute growth with grid security, affordability, and climate commitments.