The climate tech sector has absorbed an enormous amount of capital over the past few years. What’s interesting now is seeing which AI applications are moving from pilots to production - and which remain stuck in research mode.

I’ve spent the last quarter talking to founders and operators in the space. The picture that emerges is more nuanced than the pitch decks suggest, but also more encouraging than the skeptics claim.

Grid Optimization: The Quiet Winner

The least glamorous application is probably the most impactful: using AI to manage electricity grids.

Renewable energy is intermittent. The sun doesn’t always shine. The wind doesn’t always blow. Managing a grid with high renewable penetration requires predicting supply and demand with high accuracy, then balancing them in real-time.

This is fundamentally a forecasting and optimization problem - exactly what modern AI is good at.

DeepMind’s collaboration with Google on data center cooling got attention years ago (they claimed 40% reduction in cooling energy). But the grid applications are more significant. Google’s work with various utilities has demonstrated meaningful improvements in renewable energy utilization.

What I find interesting is how quickly this has become table stakes. Every major grid operator I’ve talked to is either deploying AI for load forecasting and optimization or actively building the capability. The question isn’t whether to do this - it’s how sophisticated your approach is.

The edge cases matter. An AI system that’s 2% more accurate in predicting wind generation across a state-scale grid translates to millions of dollars in avoided curtailment and reduced need for backup generation.

Carbon Accounting: Boring But Essential

Here’s an unsexy truth: you can’t manage what you can’t measure. And carbon accounting remains a mess.

Most companies have terrible visibility into their actual emissions. Scope 1 (direct emissions) is relatively straightforward. Scope 2 (purchased electricity) is manageable. Scope 3 (supply chain) is a disaster - often 80-90% of a company’s footprint, and typically estimated using industry averages rather than actual data.

AI is helping in a few ways. Natural language processing can extract emissions data from supplier documents at scale. Machine learning can improve the accuracy of estimates where direct measurement isn’t possible. Pattern recognition can identify anomalies that might indicate data quality issues.

Persefoni, Watershed, and Planetly (before Persefoni acquired them) are the names to know here. They’re building the infrastructure layer that makes corporate climate commitments actually verifiable.

This might seem like accounting rather than climate tech, but the downstream effects are significant. Better measurement enables better decision-making, more meaningful carbon markets, and actual accountability for corporate commitments.

Materials Discovery: High Potential, Early Days

AI-accelerated materials discovery could be transformative for climate tech. Better batteries, more efficient solar cells, catalysts for green hydrogen - all depend on finding new materials with specific properties.

The approach mirrors what’s happening in drug discovery. Train models on existing materials data, use them to predict properties of hypothetical new materials, synthesize and test the most promising candidates.

Microsoft’s collaboration with Pacific Northwest National Laboratory generated some attention last year - they claimed to have identified promising new battery materials 80x faster than traditional methods. That’s impressive if it holds up, though we’re still waiting to see if those materials perform at scale.

The limitation is similar to drug discovery: AI can accelerate the search phase, but validation still requires physical experimentation. A promising computational prediction still needs to be synthesized and tested, which takes time and expertise.

Still, the potential here is significant. The materials constraints on many climate technologies are real. Batteries need better energy density and less reliance on problematic materials like cobalt. Solar cells have theoretical efficiency limits we’re nowhere near reaching. Electrolyzer efficiency determines green hydrogen economics. Better materials could move the needle on all of these.

Methane Detection: Satellite AI in Action

Methane is a potent greenhouse gas - far more warming per molecule than CO2 over short time horizons. And a lot of it leaks from oil and gas infrastructure in ways that operators don’t track or report accurately.

This is a problem AI can help solve. Satellite imagery combined with computer vision can identify methane plumes at scale. Companies like GHGSat and Kayrros are building this capability.

What makes this interesting is the accountability function. When you can independently verify methane emissions from specific facilities, it changes the conversation around regulation and corporate reporting. You’re not relying on self-reported data anymore.

The technology has improved dramatically. Early satellite-based methane detection could identify major leaks at large facilities. Current systems can detect smaller leaks with higher accuracy and attribute them to specific sources. The data is increasingly being used by regulators and investors.

What’s Still Struggling

Not everything is working. Let me be honest about where I’m seeing more hype than substance.

Carbon capture optimization gets a lot of attention, but the fundamental economics still don’t work for most applications. AI can make capture systems more efficient, but if the baseline economics don’t pencil out, marginal improvements don’t change the picture.

Climate modeling is improving, but the AI applications here are more about research than deployment. Better climate predictions are valuable for policy and planning, but the connection to near-term commercial applications is tenuous.

Precision agriculture has promise but scaling challenges. The AI works, but the deployment models for getting sophisticated optimization tools to farmers at scale remain unsolved.

The Investment Lens

For VCs and innovation managers evaluating climate AI opportunities, here’s my framework:

Strong conviction: Grid optimization, carbon accounting infrastructure, satellite-based monitoring. These have clear value propositions, proven technology, and paths to scale.

Cautious optimism: Materials discovery, building energy management, supply chain optimization. The technology is promising but the paths to commercial deployment are less clear.

Wait and see: Direct carbon capture, most carbon credit verification schemes, consumer-facing climate apps. Either the underlying economics don’t work, the AI doesn’t add enough value, or the market isn’t ready.

The climate transition is happening regardless of AI. But AI can accelerate specific pieces of it - particularly the optimization and measurement problems that scale poorly with traditional approaches.

The opportunity for innovators is finding those specific leverage points where AI capability meets climate necessity.