How can Startups choose Climate Solutions that Leverage AI Effectively?

The climate crisis is staring us down, and startups are stepping up with bold ideas to tackle it. We’ve seen firsthand how artificial intelligence (AI) can be a game-changer for sustainability. But here’s the catch: AI often comes with a hefty computational price tag, guzzling energy and resources that can undermine the very environmental goals that startups aim to achieve. Sam Altman humorously confirmed on X that even polite interactions with large language models—like saying 'please' or 'thank you' to ChatGPT—can translate to significant computational costs, highlighting just how resource-intensive AI has become. So, how can early-stage companies leverage AI effectively for climate solutions without needing supercomputers or draining their budgets? Let’s break it down.

The AI-Climate Conundrum

AI’s potential to address climate challenges is undeniable. From optimizing energy grids to tracking specific environmental impacts— such as plastic waste reduction in supply chains or promoting circular economy practices— , AI is helping startups make a dent in global emissions. For instance, companies like BrainBox AI, which uses AI to decarbonize commercial buildings, have slashed HVAC energy costs by up to 25% by predicting temperature needs with historical and external data. Similarly, AgroScout, an Israeli startup, uses AI to monitor crops in real-time, cutting agrochemical use by 85% and boosting sustainable farming.

But there’s a flip side. Training large AI models, like those powering ChatGPT, can consume coomillions of liters of water for data center cooling and emit carbon equivalent to thousands of households. A Times of India feature revealed that a single interaction with a large AI model can cost fractions of a cent in computational power, but those fractions add up at scale. By 2027, AI systems could use up to 6.6 billion cubic meters of water annually—nearly half the UK’s yearly water usage. For climate-focused startups, this resource intensity is a paradox. How do you deploy AI to save the planet without contributing to its degradation?

The International Energy Agency (IEA) notes that data centers are among the fastest-growing sources of emissions globally, but their share will remain below 1.5% of the total energy sector emissions through 2035. More importantly, the IEA emphasizes that "the widespread adoption of existing AI applications could lead to emissions reductions that are far larger than emissions from data centres"—and that the power consumed by AI is expected to be dwarfed by the energy savings it can enable.

Lightweight AI: The Startup’s Secret Weapon

The good news? You don’t need a massive data center to make AI work for climate solutions. Startups can adopt lean, resource-efficient AI strategies that align with their mission and constraints. Here’s how:

1. Start Small with Pre-Trained Models

Instead of building AI models from scratch, startups can use pre-trained, open-source models and fine-tune them for specific climate applications. For example, Pendulum uses a human-in-the-loop approach to tune large language models (LLMs) for precision agriculture. By processing unstructured data into machine-readable formats, Pendulum helps farmers optimize pesticide and water use, reducing waste by up to 92%. This approach cuts down on computational needs, as the heavy lifting is already done by the pre-trained model.

2. Embrace Edge Computing

Why send all your data to a cloud server when you can process it locally? Edge computing—running AI models on devices like sensors or IoT gadgets—reduces latency, bandwidth, and energy use. Bug Mars, a Canadian startup in our network, uses NVIDIA Jetson Orin Nano modules to monitor insect farms at the edge. Their AI tracks temperature and pest levels in real-time, boosting yields by 30% without relying on power-hungry cloud infrastructure. For startups, edge computing is a budget-friendly way to deploy AI in remote or resource-constrained settings, like rural farms or off-grid energy systems. It’s also greener, as it minimizes data transfer and server load. It helps sidestep the cumulative energy draw of repeated server calls, which can add up quickly even for small tasks, as suggested by the recent reporting around ChatGPT's operating costs.

3. Prioritise Data Efficiency

AI thrives on data, but more isn’t always better. Startups can use techniques like data augmentation or synthetic data generation to train models with less real-world data. For instance, generative AI can create synthetic datasets for subsurface modeling in  geothermal energy or carbon sequestration, where real data is scarce. This approach, highlighted by AWS’s climate tech initiatives, reduces the need for extensive data collection, saving both time and compute resources. At Climate Collective, we guide startups to focus on high-quality, relevant datasets. A startup tracking urban heat, like FortyGuard, doesn’t need to process every satellite image—just the ones relevant to city landscapes. Curating data smartly keeps computational costs low and models focused.

4. Optimise for Energy Efficiency

AI's energy footprint is a real concern, but its potential benefits in the power sector far outweigh its costs. Startups can optimize their workflows to minimize energy use while leveraging AI to drive sustainability. Techniques like model pruning (removing redundant parts of a neural network) or quantization (reducing model precision) can make AI run faster on less power. VIA, a startup we’ve worked with, uses compact open-source LLMs running on CPUs instead of GPUs, slashing energy use while delivering sustainability data for building managers. Moreover, AI is already improving energy forecasting, grid optimization, and renewables integration, reducing system inefficiencies and enhancing resilience. For instance, DeepMind's AI improved wind power output predictability in the UK by 20%, leading to smarter dispatch decisions and reduced backup fossil generation. Startups should also consider cloud providers committed to renewable energy, like AWS, which aims to power its data centers with 100% renewable energy by 2025. By choosing such providers and adopting energy-efficient practices, startups can align their AI operations with climate goals and contribute to net-positive environmental outcomes.

5. Collaborate and Share Resources

No startup is an island, especially in climate tech. Collaborative platforms like Climate Change AI or Google DeepMind’s dataset initiatives provide shared tools and knowledge bases for climate-focused AI. At Climate Collective, we promote a community where startups can pool resources, share AI models, and learn from each other’s successes. Our S2C2 initiative connects entrepreneurs across Asia, Africa, and Latin America, helping them access shared AI tools without reinventing the wheel. The Climate Collective Approach.

AI for Climate Use Cases is Different

Startups can adopt lean, resource-efficient AI strategies to minimize their energy footprint—using techniques like model pruning (removing redundant components of a neural network) and quantization (reducing model precision) to run models faster and with less power. These methods are especially important as AI’s own energy use grows.

However, when AI is applied to energy-saving use cases, particularly in the power sector, its environmental benefits can significantly outweigh the energy it consumes. AI is already enhancing energy forecasting, grid optimization, and renewable integration, helping utilities reduce inefficiencies and increase system resilience. For example, DeepMind’s AI improved wind power output predictability in the UK by 20%, enabling smarter dispatch decisions and reducing reliance on backup fossil generation.

This aligns with the IEA's finding that AI's energy savings in the power sector can exceed its own energy consumption, driving net-positive environmental outcomes.

A Call to Action for Startups

As a startup founder or innovator, your mission is to solve climate challenges without creating new ones. AI can be your ally, but it’s not about having the biggest model or the most data. It’s about being smart—choosing lean models, leveraging edge computing, curating data wisely, optimizing for energy, and collaborating with others. And in doing so, your AI solution might just save far more energy than it consumes—something both the IEA and industry leaders suggest is well within reach.

Written By -- Pratap Raju, Founding Partner at Climate Collective Foundation

Read More:

How AHEAD Delivers Adaptive Digital Transformation with India-First Talent & AI

Freshworks Partner Program: Insights into the Evolving Channel Ecosystem

Partner Managed Cloud Model Supports Our GTM Strategy

Cloud Centric Cybersecurity Solutions Designed and Made in India