Using AI for good: what the evidence says, and what it requires

The conversation about AI and climate tends to run in two separate tracks. One examines AI's growing energy footprint and asks whether the technology is compatible with decarbonisation commitments. The other examines AI's analytical capabilities and asks whether it could accelerate the transition. Neither question can be properly answered without taking the other seriously.

The footprint is real and growing

Global electricity consumption from data centres is projected to double by 2030, according to the IEA, and AI is the primary driver. Electricity consumption in accelerated servers, the hardware running AI workloads, is projected to grow by 30% annually. A single generative AI query consumes approximately ten times the electricity of a conventional web search. Large technology companies, including Google and Microsoft, have reported increases in greenhouse gas emissions in recent years as AI adoption scaled, creating visible tension with their public net-zero commitments.

The investment trajectory sharpens the picture. The largest technology companies committed capital expenditure above $400 billion in 2025, with a further 75% increase projected for 2026. At least half of the new power associated with AI data centre growth is likely to come from fossil fuel sources, given current constraints on renewable capacity. For organisations that have made climate commitments, the procurement of AI services carries a material emissions implication that sits within Scope 2 and, increasingly, within Scope 3 as the disclosure frameworks governing these purchases widen.

The opportunity is proportionate

The IEA projects that existing AI applications, adopted at scale in end-use sectors, could reduce global CO2 emissions by around 1,400 Mt by 2035. That figure is roughly three times larger than total projected data centre emissions over the same period. The asymmetry is significant: if AI's climate applications are developed with intent, the technology's contribution to decarbonisation can substantially exceed its own footprint.

The applications are concrete and already operating. AI-driven forecasting is improving the accuracy of renewable energy prediction, reducing curtailment and helping grid operators balance variable supply with real-time demand. In industrial settings, AI process optimisation is demonstrating energy reductions of 30 to 50% in some contexts. For Scope 3 emissions, which represent more than 80% of total corporate emissions across many industries, AI is beginning to make supply chain data tractable, converting fragmented, estimate-based reporting into something that can be tracked and acted on at scale.

The regulatory environment is creating demand for exactly these capabilities. The EU's Corporate Sustainability Reporting Directive, IFRS S2, California's SB 253, and the Science Based Targets initiative now require large organisations to disclose emissions across all three scopes. The first wave of large listed entities began reporting in 2025. The manual processes that generated Scope 3 estimates in the past are not adequate for what regulators and investors now expect.

What the evidence actually requires

The case for AI as a net-zero tool is not a claim that the technology is inherently climate-positive. It is a claim that it can be, under specific conditions.

Those conditions are governance conditions. They concern which AI tools an organisation uses in sustainability and ESG functions, the quality of the underlying data those tools depend on, and the oversight structure applied when AI outputs inform regulatory disclosures. Without that governance layer, AI adoption in sustainability functions can introduce accountability gaps that are difficult to audit, particularly as outputs move into public reporting contexts.

The evidence from early adopters reinforces this. Where AI is improving Scope 3 accuracy, it is doing so in combination with supplier engagement and primary data collection, not by replacing them. If the supplier data does not exist, AI cannot generate it. What AI can do is make the analysis more tractable once the data is there, and identify where gaps in coverage carry the greatest reporting risk.

The organisations best placed to extract genuine climate value from AI are those that approach the technology with structured oversight. That means maintaining judgement about where AI outputs can stand on their own, building the internal capability to govern these tools under scrutiny, and understanding the infrastructure implications of procurement decisions before those decisions are made.

AI has a real and demonstrable role in the net-zero transition. The question is not whether to use it. It is whether the conditions are in place to use it well.

TCC's work in this area

TCC supports organisations to use AI responsibly in sustainability and ESG functions, and contributes applied research on AI's role in the climate economy. If you are working through what AI adoption means for your climate commitments or your reporting obligations, we would welcome a conversation.