How AI’s Energy Hunger Could Reshape Climate Goods
Climate-Positive Shopping
🌱
Earn carbon credits on every sustainable purchase
Same prices as direct · 25,000+ partnered stores.
AI is hungry for electricity. But that hunger could accelerate renewable energy deployment—and reshape how we measure and optimize climate action.
Dear IMPT Family,
Artificial intelligence is one of the most computationally intensive technologies ever created. Training a large language model consumes hundreds of thousands of kilowatt-hours of electricity—sometimes more than a small country uses annually. Running AI inference at scale requires sustained electricity consumption that rivals industrial processes.
As AI deployment accelerates globally, this energy demand raises an urgent question: will AI’s environmental hunger derail climate progress? Or will it accelerate solutions? The answer is probably both—and understanding it matters for how you shop and make climate choices.
🔥 Key Highlights 🔥
1️⃣ AI model training (especially large language models) consumes 100,000–1,000,000+ kWh per model
2️⃣ Running AI at inference scale requires sustained electricity—growing by 15–30% annually as adoption increases
3️⃣ Data centres powered by coal are carbon-intensive; those powered by renewables are near-zero-carbon
4️⃣ AI is increasingly used for climate solutions: optimising renewables, predicting weather, designing green materials
5️⃣ The real question isn’t whether AI has carbon cost—it does. It’s whether AI’s value as a climate tool justifies it
1️⃣ The Numbers
A single training run of a large language model (like GPT-3 scale) consumes roughly 1.3 million kWh of electricity, generating 550 tonnes of CO₂. Smaller models consume far less: 5,000–50,000 kWh for typical commercial models. Inference—running a trained model to answer a query—uses vastly less energy per operation than training, but scales to trillions of operations annually across millions of devices.
Estimates suggest AI-powered data centres now consume 1–4% of global electricity, growing to 10–15% by 2030 if trends continue unchecked.
2️⃣ The Renewable Energy Wild Card
Here’s the crucial nuance: AI data centre carbon depends entirely on electricity source. A data centre powered by 100% renewable energy has zero per-kWh carbon. One powered by coal has roughly 1 kg CO₂ per kWh.
Major AI companies (Google, Microsoft, Meta) have committed to renewable energy. Google achieved 100% renewable-matched electricity for its operations in 2023. Microsoft is building nuclear power plants to power AI data centres. The trend is clear: AI operators are investing in renewables to reduce operating carbon and improve efficiency.
This creates a virtuous cycle: AI companies deploy renewables to power AI, which increases renewable energy capacity, which decarbonises the entire grid.
3️⃣ The Climate Solution Potential
AI is increasingly used to solve climate problems:
✔ Renewable optimization: AI predicts wind and solar output, optimising grid dispatch and reducing fossil fuel backup.
✔ Weather prediction: Better modelling reduces energy use in heating and cooling.
✔ Materials science: AI accelerates discovery of better batteries, carbon capture, synthetic fuels.
✔ Supply chain optimization: AI reduces waste and transport emissions in manufacturing and logistics.
✔ Monitoring: AI detects deforestation, illegal fishing, emissions leakage in near-real time.
The carbon cost of training and running AI might be justified if it accelerates these solutions by years.
4️⃣ The Embodied Carbon Trap
There’s a secondary concern: data centre infrastructure itself. Building a data centre—steel, concrete, cooling systems—generates embodied carbon. If data centre growth outpaces useful output, you’re accumulating carbon-intensive infrastructure for marginal value.
The counterargument: data centres are already being built (cloud computing isn’t stopping). The question is whether they’re powered by renewables and operated efficiently. Pushing data centre operators toward renewable energy and efficiency standards matters more than trying to reduce compute overall.
5️⃣ The Personal Shopping Angle
Why does this matter to you? Because AI is increasingly used in supply chain optimisation, product recommendations, and demand forecasting. E-commerce platforms use AI to predict what you’ll want, optimise inventory, and reduce overproduction and waste. Climate-conscious shopping might be powered by AI—and the electricity cost of that recommendation is justified if it prevents unnecessary purchases.
Additionally, IMPT uses data and optimisation (increasingly AI-assisted) to match your purchases with carbon-offset projects. That optimisation has an energy cost, but it’s offset by the climate project it enables.
6️⃣ The Efficiency Question
AI models are becoming more efficient. Early large language models required enormous training runs. Modern techniques (distillation, pruning, few-shot learning) achieve similar capability with less compute. As efficiency improves, the carbon per useful output decreases—even with increased deployment.
7️⃣ The Policy Front
The real lever here is policy: carbon pricing for data centres, renewable energy mandates, efficiency standards. If governments price carbon properly and require renewable energy, AI operators will optimise for decarbonisation automatically.
Looking Ahead — AI as Climate Tool, If We Get It Right
AI’s energy hunger is real and accelerating. But it’s not inherently a climate problem—it depends on electricity source and whether AI’s value justifies its cost. The resolution is straightforward: ensure data centres run on renewable energy, impose carbon pricing on high-emission compute, and accelerate AI deployment for climate solutions.
As an individual, you don’t need to feel guilty about using AI-powered tools or shopping on AI-optimised platforms. You should care that the underlying compute is powered by renewables (which major operators are already doing) and that AI’s output justifies its input. Earn carbon credits on your purchases through IMPT, and you’re automatically offsetting the energy cost of the systems that recommend and serve you products.
Let’s keep building — together. 🌍💚
