AI and Climate Change: Can Technology Save What Politics Cannot?

The Central Question: Can AI Solve What Politics Has Failed to Address?

We’re living through a genuinely peculiar historical moment. Governments globally have committed to net zero targets. International agreements like the Paris Agreement represent a rare moment of planetary consensus. Yet our actual emissions continue rising. Renewable energy deployment is accelerating, yes, but global energy demand is outpacing renewable capacity additions. The gap between our stated climate goals and our actual trajectory grows wider every year.

Into this gap steps artificial intelligence. There’s a genuinely compelling case that AI can help us decarbonise faster and more efficiently than we otherwise would. The question I find myself returning to is whether AI can function as a substitute for the political will we’ve failed to demonstrate. The answer, I think, is more complicated than the technology enthusiasts would have us believe.

I’ve invested in climate tech ventures for the better part of the last five years. The entrepreneurs working in this space aren’t idealists—they’re pragmatists with deep domain expertise. They understand that climate action happens when you align economic incentives with environmental outcomes. AI is interesting to them not because it’s ideologically pure, but because it can create those alignments at scale.

How AI Can Actually Reduce Emissions

Let’s start with climate modelling. Understanding what’s happening to our climate requires processing enormous datasets—temperature records, atmospheric measurements, satellite imagery, ocean data, ice core analysis. Human researchers have been doing this work for decades, and their models have become progressively more sophisticated. AI systems can ingest these datasets, identify patterns, and generate predictions that help us understand climate dynamics with greater precision.

More importantly for near-term action, AI can help optimise energy systems in real time. A modern electricity grid is an enormously complex system. Demand fluctuates. Generation sources vary—solar and wind are intermittent, nuclear and coal provide baseload, batteries store and discharge. Managing this system to minimise costs whilst maintaining reliability is a problem that requires processing millions of data points simultaneously. AI systems excel at this kind of optimisation. Google’s DeepMind division has demonstrated that machine learning approaches can reduce the energy required to cool data centres by 40 per cent. That’s not marginal improvement—that’s transformative.

Smart grid applications represent another significant opportunity. When you integrate AI with distributed energy resources—rooftop solar, battery storage, electric vehicles—you can optimise how electricity flows across entire networks. Instead of centralised control, you get dynamic, real-time balancing. This enables higher penetration of renewable energy without requiring massive battery storage investments or grid reinforcement. Early deployments show that AI-optimised grids can integrate renewable energy far more efficiently than conventional approaches.

Carbon capture technology benefits from AI optimisation as well. Direct air capture facilities pull carbon dioxide directly from the atmosphere, which is energetically expensive. AI can optimise which materials to use, what operating conditions maximise capture, and how to integrate capture systems with industrial processes to minimise overall energy requirements. Companies like Carbon Engineering have been applying machine learning to improve capture efficiency for years.

Industrial efficiency is another domain where AI creates genuine opportunity. Manufacturing facilities consume enormous quantities of energy. Process optimisation—adjusting temperatures, pressures, flow rates, timing—is something human operators do based on experience and intuition. AI systems trained on historical operational data can identify energy-saving adjustments that reduce consumption without compromising output. When you’re operating a cement plant, a steel mill, or a chemical facility, small efficiency gains compound across continuous operations into massive annual energy reductions.

The Uncomfortable Paradox: AI’s Massive Energy Appetite

This is where the story becomes more complex. Training and deploying large language models requires staggering amounts of electricity. The actual numbers are disputed—companies don’t always disclose energy consumption publicly—but credible estimates suggest training a large AI model like GPT-4 consumes somewhere between 500 and 2,000 megawatt-hours of electricity. That’s equivalent to the annual electricity consumption of 50 to 200 American homes, consumed in the course of a single training run.

Data centres running AI inference—answering queries, making predictions—consume substantial electricity continuously. If you scale these operations globally, you’re talking about electricity consumption that rivals entire countries. Most of this electricity still comes from fossil fuels, though this is gradually changing as renewable energy penetration increases. The bottom line is uncomfortable: building and deploying the AI systems that will help us fight climate change requires enormous energy inputs that themselves contribute to climate change.

This paradox is genuine and unresolved. Some of the enthusiasts for AI as a climate solution underplay or ignore this aspect. It’s worth taking seriously. If AI energy consumption grows exponentially as the technology scales, and if that energy remains sourced primarily from fossil fuels, then AI could actually accelerate emissions rather than reduce them.

There’s an important caveat here: the energy efficiency of AI systems is improving rapidly. Training a model today uses substantially less energy per unit of useful computation than training a model of comparable capability three years ago. Data centres are becoming more efficient. Model architectures are being designed with energy efficiency as a primary concern. The trajectory isn’t predetermined—it depends on how the industry prioritises efficiency versus raw capability.

DeepMind’s Energy Optimisation Work: What’s Actually Possible

Google’s DeepMind division has published work on using reinforcement learning to optimise data centre cooling. The problem is straightforward: keep servers cool without wasting energy on overcooling. Traditional approaches use fixed rules—cool the facility to X degrees, adjust based on load. DeepMind trained AI systems to learn the thermal dynamics of actual data centres and find cooling strategies that maintain reliability whilst reducing energy consumption.

The results were striking. One deployment achieved 40 per cent energy reduction for cooling systems. Scale that across Google’s global data centre fleet, and you’re talking about enough electricity savings to power hundreds of thousands of homes. This isn’t theoretical—it’s deployed, measured, and documented. It demonstrates that AI can create tangible climate benefits by optimising complex systems.

The UK Net Zero Commitment and the Role of Technology

The UK has committed to reaching net zero emissions by 2050, with interim targets of reducing emissions by 81 per cent by 2035 compared to 1990 levels. These are legally binding targets. Achieving them requires massive deployment of renewables, energy efficiency improvements, electrification of transport and heating, and industrial transformation.

AI will play a role in this transition—that seems clear. The question is how large a role. Some of the decarbonisation is straightforward technology substitution: replace coal plants with wind farms, replace petrol cars with electric vehicles. These things happen because the economics align, often with some policy support. AI can help optimise these systems once they’re deployed, but it’s not the fundamental driver of change.

Other aspects of decarbonisation are harder. Heavy industry—steel, cement, chemicals—cannot be easily electrified. These sectors require either massive technology innovation or fundamental production changes. AI might help with industrial efficiency, but it can’t solve the deep engineering challenges. Similarly, aviation and maritime shipping lack zero-carbon alternatives today. AI might contribute to efficiency improvements, but won’t eliminate the emissions.

What this means in practice is that AI is a valuable complement to decarbonisation policy and technology deployment, not a substitute for either. You still need clear policy signals and investment incentives. You still need engineers building renewable energy systems, grid infrastructure, and industrial technologies. AI can make these systems work better, but it can’t replace the hard work of transformation.

Nexatech’s Approach to Sustainability Technology Investment

When we evaluate climate tech ventures at Nexatech, we look for companies applying AI to solve specific decarbonisation challenges where the technology genuinely creates advantage. A company optimising industrial processes to reduce energy consumption—that’s interesting if the optimisation is significant and the market opportunity is large. A company using AI to improve renewable energy forecasting—that’s valuable because better forecasting reduces the need for peaking plants and battery storage. A company deploying AI to optimise building HVAC systems—that solves a real problem affecting millions of structures.

What we’re sceptical of is climate tech that requires AI to be economically viable. If your only competitive advantage is having better algorithms, you’re in trouble. Better algorithms are copied quickly. What matters is whether the underlying problem is real, the solution is economically sound, and the market opportunity is substantial. AI is the tool that helps solve the problem, not the problem itself.

We also pay close attention to the energy footprint of the ventures we invest in. If a climate tech company is deployed on massive compute infrastructure that requires coal-powered electricity, the net climate benefit becomes questionable. We ask founders hard questions about their energy sources, their efficiency, and their roadmap to reduced computational requirements.

The Honest Assessment: What AI Can and Cannot Do

Artificial intelligence can help us decarbonise faster and more efficiently than we otherwise would. It can optimise energy systems, reduce industrial emissions, improve renewable energy integration, and enable better climate modelling. These are non-trivial contributions to a genuinely difficult problem.

But AI cannot substitute for political will. It cannot replace the hard work of building renewable energy infrastructure, electrifying transport, retrofitting buildings, and transforming industrial processes. It cannot overcome the economic interests aligned against decarbonisation. It cannot solve the deep equity questions about how climate costs are distributed across society.

More uncomfortable still: if AI deployment itself becomes a major source of emissions, and if we’re deploying massive AI systems to solve climate problems created by other consumption, we’ve merely shifted the problem rather than solved it. The honest truth is that our climate challenge is ultimately political and economic. Technology—including AI—plays a supporting role.

What This Means for Investors and Entrepreneurs

For entrepreneurs working in climate tech, the lesson is straightforward: build solutions that solve real problems in ways that make economic sense. Don’t rely on AI to be the magic solution—use it where it genuinely creates advantage. Understand your energy footprint. Think carefully about whether your approach actually reduces global emissions or merely shifts them.

For investors, the opportunity in climate tech remains enormous, particularly in industrialisation of clean technology and deployment at scale. AI will play a supporting role. The real value creation happens in companies that solve hard technical problems, operate efficiently, and build solutions that work in real markets. That’s harder than waiting for AI to solve everything, but it’s also more honest about where real progress comes from.

Related reading: How Emotional AI Claims to Read Your Feelings — and Why It Probably Can’t, Developing Ethical Frameworks for AI Implementation and What is information communication technology ict: A concise guide to ICT basics.