AI’s Two-Way Relationship with Energy and Climate
Our fund thesis was hot at SF Climate Week According to Climate Tech VC “It became a running joke that two-thirds of the events covered at least one - if not all three - of these themes: ‘Data centers, AI, and Power.’”
CLAI Ventures organized four oversubscribed panels in conjunction with DLA Piper and IndieBio. Here were our takeaways:
AI for Energy
For this panel we brought founders from two CLAI portfolio companies Julian Green of Brightband, Matineh Eybpoosh of Powerline, as well as Titiaan Palazzi the co-founder of myst.ai.
AI is good for controls
Our panelists discussed how AI, especially recent advancements in reinforcement learning, are good for solving control problems that are traditionally controlled by humans - aka how should the system pull on available levers in order to maximize an objective function. Powerline tries to do this in determining how utility-scale battery owners control charging and discharging of their assets. However, the tricky bit is they also have to account for the fact of how changes in their behavior affects the behavior of other participants. To account for this, many of our companies have had to build digital twins to simulate optimal behavior of other participants and arrive at a game theoretic solution.
New data for AI
As we know, AI quality is primarily about data. Brightband made the point that ChatGPT was trained on public Internet data which was 570GB. There is far more weather data than that - on the order of 100-200PB, or about 200,000 times more data. This data hasn’t been fully trained on before, simply given the magnitude, and presents a huge opportunity for better AI-based weather models. Conventional weather data is based on physical simulations. There are interesting techniques for reconciling AI predictions with the physics models and speeding up weather forecasting.
Energy for AI
We were joined by an exceptional panel: Dr. Liang Min of Stanford Bits & Watts Initiative, Dr. Phitchaya Mangpo Phothilimthana of Hardware-Software Codesign at OpenAI, and Wendi Zhang, AI Practice Co-Leader at Egon Zehnder.
We focused on how the AI boom is driving greater energy usage and what optimization techniques exist to abate this. Given the long delays in interconnection queues, many hyperscalers are already colocating energy generation with large data centers, arriving at solutions that don’t involve plugging into the shared grid.
Hyperscalers are already optimizing their energy usage - both for net-zero reasons and cost savings reasons. Data center optimization includes different packing algorithms such that low priority jobs can be run at times where energy is more abundant.
Hardware-software co-design
Panelists shared that teams working on hardware design always have the workload in mind. But since chip development takes years, when the chip is ready for use, the workload it was designed for may no longer be popular or relevant. Teams are working on shortening the iterative loop between designing hardware and running software. This is called hardware-software “co-design” which significantly reduces compute and energy costs.
Model architectures
There was also discussion of model architecture choices:
Small Language Models are now in vogue. Chips are getting cheaper, on the order of less than $20. Large language models are being distilled to become much smaller and fit on chips.
Mixture of experts variations - it has been rumored that GPT-4 is a large mixture of experts model. One performance optimization is having a small model serve the query and call upon a larger expert selectively in order to save on model size.
Talent is still everything
Beyond the technical breakthroughs in energy efficiency of AI, there is a need to identify talented AI engineers who are mission aligned on sustainability. Luckily, for now the economic and impact incentives are aligned on AI’s compute costs, which makes us bullish on this domain for innovation.
Breakthroughs in Energy and AI
For this panel we brought together Anna Patterson of Ceramic.ai, Page Crahan of Tapestry at Google X, and Rahul Dubey of Rhizome Data (our portfolio company) for a conversation on how AI is reshaping the energy landscape.
There’s more renewable energy waiting than the grid can handle
Today, there are 2,600 GW of renewable energy projects stuck in U.S. interconnection queues — that’s more than twice the total capacity of the U.S. grid. Moving these projects through the queue faster is one of the biggest climate unlocks out there. But to do that, we need new tools for planning, permitting, reliability testing, and grid management.
The grid is underutilized
Panelists pointed out that the U.S. grid runs at under 50% of its capacity today. AI-powered grid modeling can help surface where we’re underusing infrastructure and squeeze more throughput out of the system we already have. Think advanced simulations, demand predictions, and dynamic optimization — all using existing wires and substations.
AI efficiency gains matter
Ceramic.ai is working on improving model training efficiency. This isn’t just about cutting compute costs or energy use; it’s about accelerating the pace of AI innovation itself. Faster, leaner training pipelines let teams experiment and iterate more quickly, which leads to better models in less time — a compounding effect on progress.
Selling AI into energy requires more than tech
Our panelists agreed that selling into the utility and energy space isn’t just about having great AI. It’s about solving very specific, high-stakes pain points — for Rhizome that is solving for grid resilience and fire-risk mitigation with concrete suggestions utilities can act on. It’s about building trust with partners who operate under strict reliability and safety mandates.
Investing in Planetary Health
We had Ajay Gupta, founder of CLAI Ventures and Alex Kopelyan, Senior Director & Partner at IndieBio discuss investing in climate + health.
Redefining Planetary Health
We explored planetary health as the intersection of human well-being and environmental resilience. From heatwaves and wildfires to flooding and food insecurity, climate impacts are increasingly public health crises. Panelists emphasized that adaptation—such as cooling solutions, flood prediction, and resilient agriculture—is at its core about protecting lives. As such, the traditional boundaries between health and climate tech are dissolving, especially as investors and founders turn their focus toward scalable adaptation solutions.
The Role of AI and Capital in Climate Innovation
The discussion then turned to how AI is accelerating innovation and attracting capital to underfunded sectors like water, insurance, and biodiversity. AI is enabling everything from grid modeling and weather forecasting to plant disease detection and sustainable mining. Startups like Rhizome and Brightband show how AI can unlock new business models in climate adaptation. Panelists noted that success isn’t just about the model—it’s about the end solution, and whether it delivers impact and returns within venture timeframes.
Ajay Gupta
Managing Partner
CLAI Ventures
We invest in AI for climate mitigation and adaptation.
ajay@clai.vc