At San Francisco Climate Week, everyone wanted to talk about artificial intelligence.
“I was looking through all of the events on SF Climate Week, and it seemed like every single one of them had AI somewhere in the name,” joked (sort of) Rohan Nuttell of OpenAI last week, while moderating a panel called AI for Climate.
Sure, with over 300 events, there were opportunities for climate nerds to learn about carbon dioxide removal or sustainable fashion or grid infrastructure. But AI was inescapable. I heard from companies using AI to monitor flood risk, model forest carbon sequestration, and help utilities identify vulnerabilities from climate threats. I even learned about a company using AI to decarbonize pet food.
Yet one notable section of the climate world wasn’t buying the hype: Investors. In my one-on-one conversations with venture capitalists and other financiers throughout the week, the prevailing approach was wait and see. It was a striking departure from the rest of Silicon Valley, where 6-month-old AI startups are getting multi-billion-dollar valuations.
“I think there are very few large business opportunities that have single-handedly been unlocked,” Sophie Purdom, managing Partner at climate tech VC Planeteer Capital, told me, with regards to AI. “Maybe they make it better or faster or whatnot. But I don’t think we’ve seen a whole lot of new large markets that have suddenly been uniquely unlocked in climate.”
One problem is that AI can mean anything from “we have a machine learning algorithm” to “we use a large language model to help write your climate grant applications,” as this company does. But that distinction is important. Generative AI, which takes in reams of data and spits out brand-new content (think ChatGPT or DALL-E), is what’s been driving the AI hype machine since OpenAI released ChatGPT in November 2022. Eventually, generative AI could have powerful climate implications — think the development of novel EV battery chemistries or synthesis of new, more climate-friendly proteins.
But not quite yet, Shawn Xu, a partner at climate tech VC Lowercarbon Capital, told me.
Xu said he was left disappointed after a Climate AI hackathon that Lowercarbon hosted with OpenAI last year. “To be honest there was a lag between the number of interesting AI engineers and founders who wanted to go build real climate applications coming out of that hackathon.”
In the last couple of months though, Xu has been excited to see AI companies proposing “foundational models” for sectors like materials science and biology. These are generative models trained on large datasets that can perform a wide variety of tasks, like a ChatGPT for meteorology or architecture that could build weather models or design green buildings. “But I don’t think that there has been a slam dunk case on a company that we’re excited about yet,” Xu said.
This doesn’t mean that Lowercarbon and other climate tech investors are avoiding AI investments. There are plenty of well-funded climate tech companies using increasingly powerful machine learning models and algorithms to analyze patterns in large datasets and predict outcomes. It’s just that this isn’t exactly new. Companies across many industries have been using this type of predictive AI for much of the last decade. Now incorporating generative AI in the form of large language models is becoming relatively common too.
“Anything that’s solving workflow inefficiencies, anything that’s helping you get context from somewhere else, anything that’s helping you understand more data,” are well understood applications of AI that Juan Muldoon, a partner at climate software VC Energize Capital, told me he’s excited about.
“I think you’re going to see it materially impact long-running operational costs for [energy] projects,” Scott Jacobs, co-founder and CEO of the sustainable infrastructure investment firm Generate Capital, told me. “It’s just another use of technology replacing humans.”
That doesn’t always make for a particularly flashy business. Muldoon cited one of Energize’s portfolio companies, Jupiter Intelligence, which “takes very, very large amounts of climate, weather, and terrain data to be able to more accurately predict asset level risks associated with particular climate events,” he explained. “So that’s a data AI company. But it’s not really marketed that way.”
Maybe that’s because in this era, the term is almost self-evident. As an old editor once told me, writing that a tech company uses “machine learning” or “AI” to perform data analysis can be as mundane and obvious as advertising that a company uses “the internet.” But as generative AI moves beyond advanced chatbots and towards the type of broader foundational models that Xu is most excited about, investment could heat up.
Xu told me that Lowercarbon has made a yet-unannounced investment in a company that gathers vast amounts of earth observation data, which could hopefully one day be used to create a “foundational model for earth science.” This model could potentially do things such as generate custom maps to track natural disasters or the climate risks to crops and built infrastructure. Xu says a company like this would be “a holy grail.”
Yet the main holdup to some of these “holy grail” companies is that we often lack not only enough data but a comprehensive understanding of how to characterize that data, said Clea Kolster, partner and head of science at Lowercarbon.
“We’ve seen a lot of pitches on AI for chemistry,” she told me. And while AI could spit out new atomic and molecular combinations for use in novel battery cells, “the amount of those new things that are actually going to be good is probably very small until you actually start to have a better understanding of how many of these materials work in different structures and environments.”
Even if scientists and researchers get a better handle on the datasets they’re working with, Purdom told me she’s generally skeptical of investing in companies that use AI to do basic R&D, citing the buzzy example of AI being used in critical minerals exploration and extraction “The competency of the prospecting and the R&D approach seems distinct to me from the actual value extraction, physical resource extraction part of the business,” she told me. The same could be said of using AI for battery design or protein development. “I have seen few examples where the platform approach of just the research and identification part is where there’s been a big standalone business.”
Not to say everyone takes that point of view. Bay Area-based KoBold Metals, an AI-enabled minerals exploration company, has raised over a billion dollars, with Bill Gates’ climate tech VC, Breakthrough Energy Ventures as a leading investor.
But overall, the potential for novel applications of AI in the climate space is still largely being figured out. And in these early stages, many climate investors are treading carefully.
“I have talked to a number of these AI companies,” Jacobs told me. “They’re talking about climate impacts and they have real value propositions that they’re going after. Great! But they don’t have real success stories yet.”
