AI tools for sustainability occupy an interesting position in the broader AI landscape: they are often cited as a key reason to be optimistic about AI’s net impact on the world, and the applications are genuinely impressive — energy grid optimisation, carbon footprint measurement, materials discovery for clean technology, deforestation monitoring from space. At the same time, training and running large AI models has a non-trivial energy footprint, which means the sustainability impact of AI tools is not uniformly positive without careful consideration of which applications justify the energy cost. For a broader walkthrough, our AI Tools for Every Industry is a good next read.
I want to engage with this honestly rather than treating AI for sustainability as straightforwardly beneficial: the best AI sustainability applications deliver environmental returns that clearly exceed the energy cost of the AI; the worst are sustainability-washing that uses AI terminology without meaningful environmental impact. The filter I apply throughout this guide: does this application deliver environmental returns that are clearly larger than its energy cost?
Carbon measurement and reporting — the foundation
Watershed (enterprise pricing) is the AI-powered carbon accounting platform used by a growing number of large companies to measure, track, and reduce their carbon footprint across Scope 1, 2, and 3 emissions. The AI capabilities are specifically useful for Scope 3 emissions — the supply chain emissions that are the most difficult to measure accurately because they require data from hundreds or thousands of suppliers. Watershed’s AI processes supplier data, applies appropriate emission factors, and surfaces the emissions hotspots where reduction efforts will have the most impact. For companies with sustainability commitments and reporting obligations, accurate Scope 3 measurement at scale is one of the genuinely difficult measurement problems that AI makes tractable.
Persefoni (enterprise pricing) is a Watershed competitor with particular strength in financial services — helping banks, asset managers, and insurers measure and report the financed emissions in their portfolios, which is increasingly required by financial regulators. For financial institutions with growing climate disclosure obligations under frameworks like TCFD and the SEC’s climate disclosure rules, AI-powered portfolio emissions measurement addresses a regulatory requirement that was previously impossible to fulfil accurately at portfolio scale.
Plan A ($1,500+/month, SME focus) is the AI carbon management platform most accessible for mid-size companies — providing AI-assisted carbon footprint calculation, reduction planning, and sustainability reporting without the enterprise implementation complexity of Watershed or Persefoni. For companies making their first serious sustainability measurement efforts, Plan A’s guided approach reduces the expertise barrier to meaningful carbon accounting.
Energy optimisation
Google DeepMind’s AI for data centres demonstrated a 40% reduction in cooling energy consumption in Google’s own data centres using AI-powered optimisation — one of the most cited examples of AI delivering measurable environmental return. The approach — using reinforcement learning to continuously optimise cooling system settings based on thousands of real-time variables — is now being applied beyond Google through commercial platforms.
AutoGrid (enterprise, utility and energy companies) provides AI-powered flexibility management for electricity grids — predicting demand, optimising distributed energy resources (batteries, EV charging, demand response), and reducing the need for fossil fuel peaker plants that run only during demand spikes. For utilities and energy companies managing grids with increasing proportions of variable renewable generation, AI grid flexibility management is becoming operationally necessary rather than merely beneficial. The specific problem AI addresses — optimising across thousands of distributed energy assets simultaneously — is one where AI has a genuine capability advantage over human planning approaches.
AI-powered building energy management systems apply AI to commercial and industrial building energy optimisation — learning building usage patterns and weather dependencies to minimise energy consumption while maintaining comfort conditions. For large buildings where energy is a significant operational cost, AI building management consistently delivers 10–30% energy reduction with payback periods that make the business case straightforward. This is one of the clearest cases where AI sustainability applications pass the environmental ROI test with room to spare.
Supply chain and materials
Altruistiq and similar AI supply chain sustainability tools help companies measure and reduce the environmental impact of their supply chains — processing supplier data, identifying high-impact categories, and suggesting procurement decisions that reduce environmental footprint. For companies with complex supply chains where sustainability impact is concentrated in specific categories or geographies, AI-assisted analysis identifies where intervention will have the most impact.
Citrine Informatics applies AI to materials discovery — using machine learning to identify new materials with specific properties from vast chemical spaces that would take decades to explore experimentally. Applications include discovering new battery materials for energy storage, catalysts for clean hydrogen production, and building materials with lower embodied carbon. The materials discovery timeline compression — from years to months — is one of the clearest cases where AI delivers environmental value by accelerating the development of clean technology.
Monitoring and measurement
Global Forest Watch with AI (free for researchers and policymakers) uses satellite imagery and AI to monitor deforestation in near real-time across the world’s tropical forests — detecting forest loss within days of occurrence rather than the months that manual interpretation of satellite imagery requires. For conservation organisations, governments, and companies with deforestation commitments in their supply chains, AI-powered forest monitoring provides the verification capability that makes forest protection commitments credible and enforceable.
Planet Labs with AI analysis (commercial pricing) provides AI-powered analysis of daily satellite imagery across the entire Earth’s land surface — enabling monitoring of agricultural practices, industrial emissions, urban heat islands, and ecosystem health at global scale. For environmental monitoring applications where the question is “what is happening where, at what rate,” AI analysis of satellite imagery has genuinely transformed what is knowable. The monitoring that previously required either expensive in-situ measurement networks or infrequent satellite pass analysis is now continuous and comprehensive.
Sustainability AI tools reference
| Sustainability application | Best AI tool | Environmental return |
| Corporate carbon accounting | Watershed or Plan A | Enables measurement and reduction of real emissions |
| Grid flexibility and renewables integration | AutoGrid | Reduces fossil fuel peaker plant use |
| Building energy optimisation | AI building management systems | 10–30% energy reduction in large buildings |
| Clean materials discovery | Citrine Informatics | Accelerates development of clean technology materials |
| Deforestation monitoring | Global Forest Watch AI | Near real-time detection enabling faster intervention |
| Supply chain sustainability | Altruistiq or Watershed | Identifies highest-impact reduction opportunities |
The energy cost question — being honest about it
The honest treatment of AI for sustainability requires acknowledging that AI model training and inference has energy costs that vary significantly by model size and application. Training large language models requires substantial energy — GPT-4-scale training runs consume energy comparable to small towns over weeks. Individual inference queries are much smaller, but they add up across millions of daily interactions.
The principle for evaluating AI sustainability applications: the environmental return must demonstrably exceed the energy cost of the AI. Grid optimisation that reduces fossil fuel use by megawatt-hours clearly exceeds the energy cost of running the AI. Using a large language model to generate sustainability reports describing reductions that were achieved without AI assistance does not represent a net environmental contribution regardless of how sustainability-focused the content is.
The applications in this guide — carbon measurement, energy optimisation, materials discovery, forest monitoring — all represent cases where the environmental return is orders of magnitude larger than the AI energy cost. Evaluating new AI sustainability applications against this return criterion is a useful filter for distinguishing genuine environmental impact from sustainability-washing that uses AI terminology without meaningful environmental benefit.
AI for climate research and policy
Beyond the operational sustainability tools above, AI is being applied in climate research and policy in ways that will matter for long-term climate outcomes:
Climate modelling acceleration: AI emulators that can run climate model simulations orders of magnitude faster than traditional physics-based models — enabling higher resolution projections and more rapid exploration of climate scenarios. NVIDIA’s FourCastNet and DeepMind’s GraphCast weather forecasting AI demonstrate that AI can match physics-based models on atmospheric prediction at a fraction of the computational cost, which has direct implications for climate projection capabilities.
Biodiversity and ecosystem monitoring: AI systems that identify species from imagery, track population changes from camera trap data, and model habitat change from satellite imagery — providing the monitoring data needed to understand ecosystem impacts of climate change and conservation intervention effectiveness.
Policy impact modelling: AI-assisted analysis of climate policy scenarios — modelling the emissions impacts, economic effects, and distributional consequences of different policy combinations. For policymakers trying to understand the likely outcomes of complex climate policy packages, AI modelling assistance is increasingly part of the policy development process.
The trajectory is clear: AI will play a progressively larger role in the scientific and operational infrastructure of humanity’s response to climate change. The applications where that role is most clear-cut — monitoring, measurement, optimisation, and discovery — are the ones where the tools are most mature and the returns most demonstrable. Our guide on AI tools for scientific research covers the research computing tools that underpin climate data analysis and environmental research. Our guide on best AI tools for supply chain covers the supply chain tools that intersect with sustainability — particularly supplier risk monitoring that increasingly includes environmental risk alongside commercial risk.
Sustainability reporting and the AI toolchain
One increasingly common application of AI in sustainability that doesn’t fit cleanly into any of the above categories: AI assistance in sustainability reporting itself — drafting GRI reports, TCFD disclosures, CDP questionnaire responses, and sustainability strategy documents.
This is a legitimate and useful application of general AI writing tools (Claude or ChatGPT) for organisations whose sustainability reporting burden has grown substantially as disclosure requirements expand. Writing a comprehensive TCFD report or GRI disclosure is a significant documentation task that AI writing assistance makes more efficient. The discipline required: the data, the targets, the governance structures, and the risk assessments that populate sustainability reports must be real and accurate — AI helps communicate them, it should not be used to generate them or to make them appear more comprehensive than they are.
The risk in AI-assisted sustainability reporting is the same as the risk in AI-assisted reporting in general: that the polish of the AI-generated content suggests more substance than the underlying programme has. A beautifully formatted climate risk disclosure that describes risk assessment processes that don’t actually exist is greenwashing, regardless of how it was produced. AI writing tools make it easier to produce credible-looking sustainability disclosures; sustainability programmes that back them up with real data, real targets, and real progress remain the actual measure of sustainability performance.
Choosing the right AI sustainability tool for your organisation
The right AI sustainability tool depends heavily on where your organisation is in its sustainability journey:
- Early stage (basic footprint measurement): Plan A or similar accessible carbon accounting platforms provide a guided starting point without enterprise implementation requirements. The goal at this stage is getting a credible baseline footprint measurement that can be used to set reduction targets.
- Intermediate stage (target-setting and reporting): Watershed for companies with complex supply chains needing accurate Scope 3 measurement; dedicated ESG reporting tools for companies with growing disclosure obligations to investors and regulators.
- Advanced stage (operational reduction): Energy optimisation AI for large facilities; supply chain sustainability tools for companies with significant supplier footprints; sector-specific tools for companies in industries with material-specific emissions challenges.
- Research and innovation stage: Citrine Informatics and similar AI materials discovery tools for companies whose sustainability requires developing new materials or processes rather than optimising existing ones.
The organisations that get the most from AI sustainability tools are consistently the ones that approach them as measurement and optimisation infrastructure rather than as reporting tools. The measurement enables the targets; the targets enable the reduction; the reduction is the actual sustainability outcome. AI tools that help measure more accurately and optimise more effectively are valuable; AI tools that help report more impressively without enabling better decisions are window dressing.
The AI sustainability tools that will prove most valuable over the coming decade are the ones that make it possible to act on information that previously couldn’t be collected — to see supply chain emissions that were invisible, to monitor deforestation that was unreported, to identify energy waste that was unmeasured, to discover materials that were undiscovered. That capability to expand what is knowable and optimisable is where AI genuinely advances sustainability, not in the documents that describe it.
Sustainability is ultimately about real-world outcomes — tonnes of carbon not emitted, forests not cleared, energy not wasted, materials not produced. AI tools are valuable to the extent that they enable those outcomes. Keeping that grounding in mind — asking “does this tool help us produce better actual environmental outcomes?” rather than “does this tool help us tell a better sustainability story?” — is the most useful framework for evaluating any AI sustainability investment. You might also run into AI Tools Limitations in Real‑world Decision Making.
