The AI debate has been dominated by model capability, adoption, capital expenditure and the prospect of new revenue. But AI also depends on physical infrastructure.
The International Energy Agency estimates that global data-centre electricity consumption will almost double between 2025 and 2030, while electricity use in AI-focused data centres is expected to triple.
That makes power, water, grid capacity and cost increasingly important to the investment case. These factors can affect where AI infrastructure is built, how quickly it can scale and whether it remains competitive over time. Sustainability therefore belongs in investment decisions, not just retrospective reporting.
Start with the investment problem
The strongest starting point is not whether an asset or company can carry a sustainability label, but whether resource constraints, regulation or technology choices could change its economics.
For AI and digital infrastructure, investors need to look beyond headline demand growth to power availability and price, grid capacity, water, compute utilisation, cooling and hardware refresh cycles.
These choices can lock in opportunity and liability for decades. A data-centre campus, grid connection or cooling system commissioned today may still be operating long after the current generation of AI hardware has been replaced, creating cost, regulatory and transition risks investors need to understand now.
The common pattern is simple: sustainability becomes most financially relevant when it can change the technology choice, investment decision, operating model or commercial proposition itself.
AI is making sustainability an infrastructure issue
AI makes this shift impossible to ignore. Its business case may be productivity, automation, insight or new products, but delivery depends on compute, data centres, power, cooling, water, networks, semiconductors and capital investment.
The resource profile also varies dramatically by use case. The IEA notes that energy use per AI task has been falling rapidly, but newer applications such as video generation, reasoning and agentic tasks can consume hundreds or thousands of times more energy per query than simple text generation.
Efficiency gains and higher total consumption can therefore happen at the same time.
For investors, ‘AI exposure’ is not one infrastructure profile: efficient inference on well-utilised assets has very different economics from poorly governed agentic workloads on underused GPU capacity.
The opportunity is not to slow AI adoption. It is to distinguish scalable growth from growth that depends on increasingly expensive or constrained resources.
Power is becoming a transition-risk variable
Power is already a material constraint. The UK’s Compute Roadmap says at least 6 GW of AI-capable data-centre capacity may be needed by 2030, around three times today’s capacity.
Other markets are responding differently: Ireland, where data centres used 23% of metered electricity in 2025, has tightened connection requirements for data centres, while US policy is prioritising faster permitting for AI infrastructure.
For investors, location increasingly determines connection times, capital requirements operating flexibility, energy and carbon exposure.
Reporting is not enough
Disclosure remains important, but the regulatory picture is becoming more fragmented at the same time as resource use becomes more material.
The EU Energy Efficiency Directive requires energy-performance reporting for significant data centres, and the European Commission is developing a data-centre rating scheme and considering minimum performance standards.
The UK published final Sustainability Reporting Standards S1 and S2 in February 2026 for voluntary use, while regulators consider future requirements.
Regulatory divergence does not make sustainability information less useful. It increases the importance of understanding the underlying operational evidence rather than relying only on a jurisdiction’s disclosure template.
The better model is a common evidence layer linking technology operations with energy, asset, procurement, supplier and sustainability data.
Its value is decision-grade evidence: where infrastructure is underutilised, where power or water constraints sit, where supplier assumptions are weak and where lifecycle choices affect cost and carbon.
Build sustainability into the investment thesis
For portfolio managers, advisers and asset owners, this points to a more practical analysis. Rather than asking only whether an AI or infrastructure company has a net-zero target, diligence should test whether management understands the resource economics behind growth.
Due diligence should increasingly test five things: the availability, price and resilience of power; workload efficiency and compute utilisation; water and location constraints; hardware lifecycle assumptions; and whether the underlying evidence can be trusted.
The emerging Software Carbon Intensity for AI specification is one sign of his maturation. It provides a standardised approach to measuring AI carbon intensity across the lifecycle and uses a functional unit so that efficiency can be assessed against useful output, rather than relying only on aggregate emissions.
Putting guardrails around AI growth
At company level, CIOs increasingly need practical energy and carbon guardrails around AI. For investors, the equivalent question is whether those guardrails exist — and whether they are strong enough to influence capital allocation and operating decisions.
Oversized models, repeated calls, poor caching and underutilised infrastructure can inflate cost, energy and carbon simultaneously, particularly as agentic AI increases orchestration and compute demand. This is where FinOps, GreenOps and AI Governance converge.
A credible investment case should show how management governs those consequences, from capacity planning and workload design to site selection and hardware procurement. These are questions of capital discipline and durability of returns, not specialist ESG.
Sustainability is becoming an eligibility layer for technology investment
Many of AI’s most consequential sustainability exposures will not look like traditional sustainability investments; they sit inside power, cooling, semiconductors, networks and water systems. For investors, sustainability is increasingly an eligibility and control layer: helping determine what infrastructure can be built, where it can operate, which resources can support it and whether its economics remain durable.
The question is not simply which companies can capture AI growth. It is which AI strategies can scale without power, infrastructure, resource or regulatory constraints undermining the investment case. Increasingly, those factors will help determine which assets can turn AI demand into durable returns.
Read more: Data drought: AI’s water problem
