Stop Blaming AI for the Climate Crisis – Put It to Work

When people talk about AI and the climate crisis right now, I hear two stories competing for airtime.

One is apocalyptic: “AI will eat the grid, boil the oceans and exhaust the planet.”

The other is techno-utopian: “AI will magically solve climate change for us.”

Both are wrong.

What I want to argue – and what I’ll be talking about at Altercop in Singapore – is this:

AI is an energy multiplier, not just an energy consumer.

In the long run, the cost of intelligence collapses to the cost of energy.

If that energy is clean, intelligence itself becomes a climate technology.

This blog lays out the storyline, research and transition pathways behind that claim.

1. The planetary clock is ticking – and electrification is our main lever

The science is brutally clear.

The latest IPCC assessment and its 1.5 °C pathways tell us that to avoid the worst climate damage, global CO₂ emissions must fall by about 45% by 2030 (relative to 2010) and reach net-zero around mid-century. In those pathways, unabated coal is virtually gone by 2050, and oil and gas use decline steeply unless backed by carbon capture. 

Behind that sits the wider planetary boundaries and tipping points work from Johan Rockström, Will Steffen and colleagues, warning that we are close to pushing the Earth system into a “Hothouse Earth” trajectory if we continue to overshoot climate and biosphere thresholds. 

At the same time, we are “moving at speed into the age of electricity”, as IEA Director Fatih Birol puts it: electricity’s share of final energy use has already grown to around 20%, and is expected to rise sharply as we electrify vehicles, heating and industry. 

So our task is twofold:

Decarbonise electricity as fast as possible (renewables, storage, nuclear, CCS, efficiency). Electrify everything we reasonably can – transport, heat, industry – while staying within our planetary budget.

Into this already-high-stakes equation, we have now dropped a voracious new demand centre: AI.

2. AI’s energy footprint – real, but smaller than the headlines

Let’s get the hard numbers on the table.

According to the IEA’s Energy and AI analysis:

Data centres consumed about 415 TWh of electricity in 2024 – roughly 1.5% of global electricity use.  AI workloads currently account for an estimated 5–15% of data-centre power, rising potentially to 35–50% by 2030 as generative AI scales.  Even in a high-growth “lift-off” scenario, data-centre electricity demand plateaus around 700 TWh by 2035, keeping their share below 2% of global electricity use. 

Put differently: yes, AI is a serious new load – especially regionally – but it is not about to consume “all the world’s energy”.

The IEA then does something important: it asks not just “How much energy does AI use?” but also “How much can AI save?”

In its Widespread Adoption case, deploying existing AI applications across industry, transport, buildings and power could cut around 1.4 Gt of CO₂ in 2035 – roughly 5% of current energy-related emissions – even before counting any breakthroughs AI might unlock. 

Those avoided emissions are three to four times larger than the emissions from data centres themselves in the same scenarios. 

So the arithmetic is actually quite simple:

If we use AI intelligently, it is 1–2% of the energy problem – and at least 5% of the climate solution.

UNEP reaches a similar conclusion from the environmental side. Its Navigating New Horizons report and follow-up issue note acknowledge AI’s rising energy, water and materials footprint, but focus strongly on governing AI so that its enabling effects on climate and nature vastly outweigh its direct impacts. 

The World Economic Forum captures this tension nicely in its recent paper on AI’s “energy paradox” – a technology that both drives and helps manage a surge in electricity demand. 

The challenge is not that AI uses energy; the challenge is what energy, where, and for what purpose.

3. How AI helps us stay within the carbon budget

The next question I always get is: “Show me where AI actually helps.”

Let’s walk through the evidence – not as marketing slides, but through the peer-reviewed work and major institutional assessments.

3.1 AI as a climate decision engine

A growing cluster of work in Nature and other journals shows how AI can sharpen climate science and policy:

A 2025 Nature Communications paper describes AI as a frontier tool for climate policy, using machine-learning to detect patterns and causal links in policy mixes, and help governments sequence interventions more effectively.  The UNFCCC’s Artificial Intelligence for Climate Action technical paper and its ongoing AI for Climate Action initiative set out concrete use-cases – from early-warning systems to climate-risk analytics – and call on countries to integrate AI into national climate strategies, especially in developing nations. 

In simple terms: AI lets us see the climate system, the economy and our policy options in higher resolution, and adapt faster.

3.2 AI optimising today’s energy system

Power systems

AI-enhanced forecasting improves solar and wind predictions, reducing balancing costs and curtailment; reviews of smart-grid applications find AI can significantly improve unit commitment, congestion management and dynamic line rating, effectively “creating” extra grid capacity without building new wires.  The IEA estimates that using advanced digital tools (AI included) in grids could unlock up to 175 GW of additional transmission capacity on existing lines and cut power-sector operating costs by up to USD 110 billion per year by 2035. 

Buildings and industry

Buildings are responsible for close to a fifth of final energy use; pilots using AI to control HVAC systems report 8–16% reductions in electricity use and significant CO₂ savings, with one New York case saving 37 tCO₂ annually in a single building.  In heavy industry, systematic reviews show AI improving process control, predictive maintenance and waste-heat recovery in steel, cement and chemicals, lowering energy intensity and emissions. 

Every one of those gains is more “room” within the carbon budget.

3.3 AI accelerating low-carbon technologies

The IPCC 1.5 °C pathways are unambiguous: renewables, electrification, efficiency, and some carbon dioxide removal are non-negotiable. 

AI now shows up across that entire toolkit:

Renewables & storage – AI is used to optimise siting, resource assessment, inverter control, and battery dispatch. Reviews in npj Climate Action and related journals show AI improving performance and economics across the wind-solar-battery system.  Carbon capture & storage (CCS) – machine-learning aids solvent design, process modelling, leak detection and storage monitoring, improving capture efficiency and safety. 

The UNFCCC’s climate-technology bodies now explicitly encourage Parties to use AI in these ways, especially in developing countries, but stress that AI must be “the most suitable tool for the task”, not a buzzword. 

3.4 AI inventing tomorrow’s energy system

Here the story gets more speculative but also more exciting.

Fusion

In 2022, researchers from Google DeepMind and EPFL used deep reinforcement learning to autonomously control the plasma in the TCV tokamak, shaping and stabilising configurations that were previously extremely hard to manage. 

Fusion is not solved. But this is a glimpse of a future where AI co-pilots the control systems of complex, zero-carbon power plants.

Materials and batteries

On the materials side, AI is starting to compress the search for new battery chemistries, solar materials and catalysts:

A 2025 study on LLMs for battery research shows how language-model-driven workflows can guide materials discovery and design, creating “goal-driven” search instead of blind trial-and-error.  Reviews compiled by the ITU’s AI for Good initiative and others highlight AI-identified candidates for next-generation batteries and solar materials, which experts suggest could “completely change the game when it comes to electrifying our grids.” 

Demis Hassabis at Google DeepMind has been explicit that this is the real prize: AlphaFold was a step towards using AI as a general engine for scientific discovery, from biology through to climate and energy. 

4. What the leading voices are actually saying

If we zoom out, both AI researchers and climate experts are converging on a nuanced but powerful position.

4.1 AI pioneers: safety, abundance and power

Geoffrey Hinton has warned repeatedly that there may be a 10–20% chance that advanced AI could end in a loss of human control or even extinction, and he is “kind of glad” to be 77 because he may not live to see that scenario. 

Yoshua Bengio led the International Scientific Report on the Safety of Advanced AI, now a major reference for governments. That report explicitly lists environmental externalities from compute-intensive AI as a key risk cluster alongside misuse, labour impacts and loss of control. 

Fei-Fei Li, often called the “godmother of AI”, has been equally clear that AI must be human-centred: “AI is a tool, and its values are human values… the mission of AI is not to replace humans, but to understand and empower the human mind,” she says, warning that if AI strips away human dignity “something is wrong.” 

Demis Hassabis, fresh off a Nobel Prize for AlphaFold, frames intelligence as “the ultimate general-purpose tool” and predicts that, if managed wisely, AGI could help solve crises like climate change and resource scarcity – but only with serious governance and safety frameworks. 

In other words: the people building the frontier systems are telling us two things at once:

AI is dangerous if left to grow unchecked in a narrow arms race. AI is uniquely powerful as an engine for science, optimisation and design – including in climate and energy.

4.2 Climate & sustainability leaders: AI as a “green and intelligent” enabler

On the climate side, Nicholas Stern and co-authors have just published “Green and intelligent: the role of AI in the climate transition”, which identifies five core channels through which AI can support climate action: improving science and risk assessment, spurring low-carbon innovation, optimising operation of energy and land-use systems, steering finance, and enabling participation. They estimate that AI applications in power, food and mobility could reduce annual emissions by up to 5.4 Gt CO₂ by 2035 – more than the current annual emissions of the United States. 

UNEP, the UNFCCC Technology Executive Committee, and ITU’s AI for Good platform are now building a bridge between these analytics and practice: technical papers, an AI for Climate Action award focusing on Least Developed Countries and Small Island States, and even guidance on procuring energy-efficient data centres. 

On the finance side, initiatives like the Science Based Targets initiative (SBTi) and GFANZ are providing the plumbing for net-zero transition plans – sectoral pathways, portfolio-alignment tools and target-setting standards – even as they absorb political blow-back and course-corrections. 

All of this signals something important: the climate community is not naive about AI, but it is increasingly convinced that we cannot hit 1.5 °C without it.

5. AI safety meets climate safety: energy as the new governance frontier

Here’s where my worlds of AI safety and climate & energy collide.

The International AI Safety Report 2025, the Annual AI Governance Report, and analyses of frontier-lab safety policies (OpenAI, Anthropic, Google DeepMind and others) all highlight compute, energy, water and supply-chain risk as central to future AI governance. 

At the same time:

A new AI Safety Coalition under UNEP is looking at how to put AI development on a more sustainable path and steer procurement toward efficient infrastructure.  Anthropic’s recent energy report is blunt: “Energy is central to winning the AI race” and the United States will need massive investments in clean generation and grid modernisation to support frontier AI safely.  The UK and US AI Safety Institutes are partnering with leading labs to pre-test models and monitor risks, including environmental impacts, as part of responsible-scaling policies triggered by capability thresholds. 

You can feel a quiet consensus forming:

You cannot talk seriously about AI safety without talking about energy;

you cannot talk seriously about climate safety without talking about AI.

This, to me, is the intellectual hinge of the decade.

6. Three horizons: from scarcity to abundance

So how do we move from today’s messy reality to a cleaner, abundant “intelligence-as-energy” future?

I find it useful to think in three horizons.

Horizon 1 (now–2030): Efficiency and “no-regrets” deployments

Accept that AI data centres will be a fast-growing but still single-digit share of electricity demand and a tiny share of global emissions.  Regulate and incentivise hard for: Location on rapidly decarbonising grids, High energy-efficiency and waste-heat recovery, Additional clean generation (PPAs, on-site renewables). Go all-in on AI for grid optimisation, building efficiency, industrial process control, logistics and agriculture – the “boring” use-cases that collectively take a 5% bite out of global emissions. 

If we don’t use AI to squeeze every wasted kilowatt and tonne out of the system this decade, we will be fighting the climate crisis with one hand tied behind our back.

Horizon 2 (2030–2040): AI as the operating system of a clean, electrified economy

By the 2030s:

Electricity will be the dominant growth vector in energy, driven by EVs, heat pumps, green hydrogen, and clean industry.  AI will be deeply embedded as the coordination layer: Operating high-renewables grids at scale, Orchestrating storage, EV charging and flexible demand, Steering CCS and carbon removal where truly needed, Targeting capital through SBTi- and GFANZ-aligned transition plans. 

From the outside, people might stop talking about “AI use-cases”. It will just be how the energy system works.

Horizon 3 (2040+): When intelligence rides on clean energy

Beyond 2040, the futures fork.

In my optimistic “Commonwealth” scenario:

We crack a combination of ultra-cheap solar, advanced storage, deep geothermal and maybe commercial fusion. AI helps us design, control and optimise these systems – from RL-controlled tokamaks to materials-discovery engines for new catalysts and batteries.  As a result, the marginal cost of a useful unit of computation plummets, because its main input – clean energy – is abundant.

At that point, something profound happens:

The fundamental cost of running intelligence is the cost of energy beneath it.

If that energy is near-zero-carbon and abundant, intelligence becomes an inexhaustible tool for repairing the damage we’ve done – and for avoiding new damage.

In the darker “Fortress” scenario, of course, the same capabilities are hoarded behind borders and corporate walls, powered by whatever is cheapest, and used to entrench fossil-fuel interests and surveillance states.

The technology doesn’t choose the branch. We do.

7. So what should we actually do?

Let me end with a practical agenda – for governments, companies, and the AI community.

For governments & regulators

Mandate disclosure of AI energy and emissions footprints for large models and data centres; link approvals to credible clean-energy sourcing and efficiency standards (UNEP’s forthcoming guidance is a good starting point).  Integrate AI explicitly into NDCs and climate-technology roadmaps, drawing on UNFCCC’s AI for Climate Action work.  Align digital-infrastructure planning with net-zero pathways, using IPCC-consistent scenarios and SBTi/GFANZ guidance for the financial sector.  Invest in AI capacity in the global South, so developing countries are not merely data mines and hosting locations, but co-authors of AI for climate solutions.

For companies & financial institutions

Stop treating “AI transformation” and “net-zero transition” as separate strategies. Use SBTi-aligned targets and GFANZ transition-planning frameworks to ensure every major AI investment advances, rather than undermines, your climate commitments.  Prioritise AI applications with clear, measurable climate and resilience benefits – energy management, process optimisation, low-carbon product design – before vanity GenAI deployments. Support independent AI-safety and climate-science institutes with access to your models and data, so they can evaluate both systemic risk and systemic benefit. 

For AI labs and the research community

Treat energy and environment as first-class citizens in responsible-scaling policies, alongside misuse and catastrophic-risk concerns.  Build open benchmarks and tools for AI-enabled climate solutions – from grid-stability testbeds to fusion-control simulators and materials-discovery datasets.  Embed Fei-Fei Li’s principle of human-centred AI into climate applications: no solution that undermines dignity, equity or justice should qualify as “green”. 

8. Closing thought: from energy anxiety to intelligence abundance

We live in a moment where electricity demand is rising faster than expected, driven by electrification, digitalisation and AI. 

That understandably triggers anxiety. We are, after all, still burning coal and gas to keep the lights on.

But I believe we can reframe this moment:

AI is not a free pass to ignore hard choices on fossil fuels – IPCC science still demands deep phase-out.  Nor is AI a demonic force fated to wreck the grid. Its direct footprint is manageable if we govern it wisely and decarbonise fast.  Most importantly, AI is one of the sharpest tools we have to design, operate and accelerate the clean-energy future we keep talking about. 

If we get this right, then in a few decades we may look back and say:

“This was the decade when we realised that the cost of intelligence is the cost of energy – and then we chose to make that energy clean, abundant and shared.”

That is the storyline I’m taking to Altercop.

Not AI as a climate scapegoat.

AI as a disciplined, governed, electrified ally in the fight for a liveable planet.

What AI Is Actually Doing Today on Climate and Energy – and Where the Real Breakthroughs Will Come From

When people say “AI will save the climate”, I always push back: show me the deployments, not the demos.

So let’s get concrete. Below is a map of current AI innovations that are already tackling climate change and the energy crunch – and then where I expect the next wave of progress to come from.

1. Smarter power systems: AI as the grid’s “autopilot”

What’s happening now

Renewables forecasting. Grid operators are using machine-learning models to predict wind and solar output more precisely, cutting reserve margins and curtailment. Reviews of smart-grid applications show AI models improving forecast accuracy for wind by up to ~50% and for solar by double-digit percentages, which translates directly into lower balancing costs and fewer backup fossil plants.  Dynamic line rating and congestion management. AI systems ingest weather, sensor and load data to calculate the real-time capacity of transmission lines, rather than relying on static conservative limits. The IEA estimates that digital optimisation (AI included) could unlock up to 175 GW of effective extra grid capacity on existing lines and save up to USD 110 billion per year in power-sector operating costs by 2035.  Grid stability and fault detection. Deep-learning models are used for fast frequency control, fault localisation and islanding decisions, especially in high-renewables systems. This is exactly the kind of “invisible AI” that quietly keeps lights on in markets like the UK, Germany and parts of the US.

Where improvements will come from

Whole-system digital twins. National and regional TSOs are beginning to build AI-augmented digital twins of the grid – live simulations that can test scenarios (extreme weather, EV surges, interconnector failures) and optimise dispatch and investment. Expect this to become standard infrastructure, especially as IEA and Eurelectric push digital-by-design strategies.  Agentic market participants. As power markets become more granular (15-minute or 5-minute settlement; locational prices), we’ll see AI agents bidding flexible assets (batteries, EV fleets, industrial loads) into markets in real time – effectively turning flexibility into a traded commodity.

2. Buildings and cities: invisible energy efficiency at scale

What’s happening now

AI energy management systems (EMS). In commercial buildings, AI-driven EMS optimise HVAC, lighting and equipment based on occupancy, weather and price signals. Eurelectric estimates that widespread adoption of AI EMS in buildings could save about 300 TWh of electricity annually – roughly the combined generation of Australia and New Zealand.  District-level optimisation. Some cities are starting to use AI to orchestrate entire district heating and cooling networks, using predictive control to shift load away from peak periods and integrate thermal storage. Urban planning. Computer-vision and geospatial AI help map roofs for solar potential, identify heat-island hotspots, and guide where to prioritise tree planting, cool roofs and retrofits.

Where improvements will come from

Plug-and-play AI in legacy buildings. The big unlock will be cheap, retrofit-friendly AI controllers that can be installed in old buildings without massive BMS upgrades, especially in Asia and the global South where most future floorspace will be built. City-scale coordination. As cities adopt digital twins and congestion-pricing schemes, AI will be used to orchestrate transport, buildings and distributed energy together – not in silos.

3. Heavy industry and manufacturing: AI squeezing carbon out of processes

What’s happening now

Process optimisation in steel, cement, chemicals. AI models are already used to tune furnaces, kilns and reactors for optimal temperature, flow and mix, improving energy efficiency and throughput. Recent reviews show meaningful reductions in energy intensity when AI-based process control and predictive maintenance are applied across heavy industry.  Predictive maintenance. By spotting anomalies in vibration, acoustic or temperature data, AI extends equipment life and reduces unplanned outages – which means fewer backup plants and wasted energy.

Where improvements will come from

Process re-design, not just tuning. The next frontier is AI-assisted design of entirely new industrial processes – different chemistries for cement (e.g. LC3 cements, novel binders), low-temperature pathways for chemicals, and better catalysts – discovered via materials-science AI rather than incremental tuning.  End-to-end optimisation. Linking supply-chain, production scheduling and logistics AI together can cut not just direct energy, but also the embodied emissions in materials and transport.

4. Transport and logistics: routing, fleets and new mobility

What’s happening now

Route and network optimisation. AI is widely used in logistics and aviation to optimise routing, loading and scheduling – shaving fuel use across shipping, trucking and airlines. Some airlines report 1–2% fuel savings from AI-assisted flight-path optimisation, which is huge at sector scale. EV charging and fleet management. AI platforms schedule when and where electric buses, taxis and trucks charge, avoiding grid peaks and making better use of cheap renewables. Traffic management. Computer-vision and reinforcement learning are used in smart traffic lights and congestion management, cutting idling and smoothing flows in congested cities.

Where improvements will come from

Autonomous & semi-autonomous freight corridors. Once regulatory barriers fall, AI-driven platooning and autonomous trucking on defined corridors could dramatically reduce fuel use and congestion. System-level modal shifts. The real win is when AI helps optimise the whole mobility mix – public transport, bikes, micromobility, freight – rather than just making individual cars smarter.

5. Agriculture, land use and nature: AI for carbon and resilience

What’s happening now

Methane and deforestation monitoring. UNEP’s International Methane Emissions Observatory and related systems use AI on satellite and sensor data to detect methane plumes from oil, gas and agriculture, giving regulators and companies near-real-time visibility.  Similarly, AI-enhanced earth observation is now standard for tracking deforestation, peatland degradation and illegal mining. Precision agriculture. AI models guide irrigation, fertiliser use and pest control, boosting yields while lowering inputs. Reviews highlight this dual benefit: higher productivity and lower nitrous oxide and methane emissions.  Ecosystem monitoring. Biodiversity projects deploy AI to classify species from audio and camera-trap data, helping quantify nature gains and losses – increasingly important for TNFD-style disclosures.

Where improvements will come from

AI-native MRV for carbon markets. High-confidence, low-cost measurement, reporting and verification (MRV) of soil carbon, forest carbon and ecosystem services will make or break credible carbon markets. AI is central to building that evidence base, especially in the global South.  Climate-smart crop and livestock systems. AI-accelerated breeding and management for heat- and drought-resilient crops, lower-methane livestock and alternative proteins will be a major mitigation and adaptation lever.

6. Climate science, risk and policy: “intelligence on the policy stack”

What’s happening now

Better climate models and downscaling. ML-based emulators can approximate complex climate models orders of magnitude faster, enabling probabilistic forecasts and downscaling that were previously too expensive. This is feeding into early-warning systems and local risk maps.  Policy analytics. AI is being used to mine climate policies, NDCs and SDG reports to understand where ambition, implementation and finance are misaligned – particularly in developing countries.  Financial risk and transition planning. Banks and asset managers are using AI to map physical and transition risk across portfolios and to build SBTi- and GFANZ-aligned transition plans faster than traditional spreadsheet modelling. 

Where improvements will come from

Integrated climate–economy digital twins. Think of macro-scale digital twins of cities, regions and sectors where you can test combinations of carbon pricing, subsidies, resilience investments and behaviour changes – with AI learning which combinations deliver the best emissions cuts and resilience at lowest cost. AI-assisted governance. ML tools will increasingly help negotiators and policymakers track who is doing what, spot gaps in NDCs and just-transition commitments, and expose greenwashing.

7. Frontier science: fusion, materials and “AI for discovery”

What’s happening now

Fusion control. DeepMind and EPFL’s work on deep-RL control of tokamak plasmas is a proof-of-concept showing AI can shape and stabilise fusion plasmas in real time, a key ingredient for future fusion plants.  Materials discovery. AI systems are being used to search enormous chemical and structural spaces for better battery chemistries, solar absorbers and catalysts. Reviews and pilot projects show AI can cut candidate lists from thousands to a few hundred while still hitting state-of-the-art performance in perovskite solar cells and next-gen lithium-ion batteries. 

Demis Hassabis and others are explicit that this is where AGI-class systems could have their biggest positive impact: turning scientific discovery – including energy breakthroughs – from a linear, trial-and-error process into something closer to guided search at planetary scale. 

Where improvements will come from

Closed-loop labs. Autonomous labs where AI designs an experiment, robotics runs it, instruments measure it, and the model learns will massively accelerate energy-materials discovery. Fusion, deep geothermal and beyond. Any technology that depends on controlling complex physical systems in hostile environments – from fusion reactors to super-hot deep-geothermal wells – is a natural playground for advanced control AIs.

8. So where does the next big jump come from?

If I zoom out, the improvements and progress over the next decade will come from four converging trends:

Hardware and algorithmic efficiency. The IEA’s Energy and AI report shows that stronger efficiency gains in chips, model architectures and data-centre design can cut AI-related electricity use by >15% by 2035 for the same level of service.  AI deeply embedded in climate-critical sectors. Nicholas Stern and co-authors estimate that scaled AI applications in power, mobility and food systems alone could reduce annual emissions by up to 5.4 Gt CO₂ by 2035, if supported by the right policies and investment.  Locally-led AI in the global South. UNFCCC and UNEP are now explicitly backing locally led AI for climate solutions in LDCs and SIDS – from flood early-warning in the Mekong to drought analytics in the Sahel – to avoid a world where only rich countries enjoy AI-enabled resilience.  Convergence of AI safety and climate governance. Frontier AI safety reports now treat energy and environmental externalities as first-class risks, while climate-technology progress reports stress that digital and AI systems must be aligned with Paris-consistent pathways.  That convergence is exactly where we start to see “the cost of intelligence = the cost of energy” become a design principle: we push AI toward clean, transparent, auditable compute, and we push energy systems toward AI-optimised, low-carbon abundance.