AI is finally making DeepTech venturable
In short
Can AI fit deep tech into a ten-year venture fund?
Michelle Robson and Lisa Schneider asked four groups whether AI can close the gap between deep tech's timelines and venture's, and the answers came with a physical catch.
Michelle Robson of Odyssey Ventures has invested in deep tech for about a decade, and some of Michelle's earliest investments are still waiting to take off. They are not bad companies, Michelle said. A lot of deep tech simply takes far longer than a typical ten-year fund. The hosts put the path from research to commercial scale at 15 to 20 years. Michelle and Lisa Schneider of Aramco Ventures came to test a hopeful idea, that AI can close the gap. The groups they sent off to work on it came back convinced that AI speeds things up. The group that looked hardest at the tools was just as clear that it cannot do the physical work.
Lisa described the same gap from an industrial investor's side. When Aramco Ventures looks at a new technology, it often expects another 10 to 15 years before it can be deployed at a scale that matters. Capital is the second problem. First-of-a-kind plants can need 500 million or more before an infrastructure investor will take over, and venture funds do not have that money. The hosts split the table into four groups, each with a question, and asked them to report back.
otherwise it'll give you very overconfident garbage.
Fewer people, faster loops
The group on AI and the capital stack was unanimous that AI helps. Its spokesperson said teams can now do as much as before with far fewer people, and people are the biggest cost in building a company. In engineering that means simulation, digital twins and faster design iteration. Most in the group had connected Claude or another model to the tools they already use, such as Aspen or Excel, to speed up a day's work. On the business side they used AI to find and apply for grants, test business models, run techno-economic analysis and draft patents.
The caveats were firm. Bad data still gives bad answers, and the models tend to be overconfident, so you have to be careful what you feed them, "otherwise it'll give you very overconfident garbage." Nor can AI and simulation take a company the whole way. The physical validation step remains. What AI can do is make sure that when a company reaches it, the tests yield more useful information.
Another participant found AI better at critique than at creation. Founders can record themselves pitching, feed in their business model and ask a model to respond as an early-stage, late-stage or debt investor. One of the hosts pointed to a portfolio company, Flow Engineering, which moves from a central requirements tool towards CAD driven by prompts and has cut design time from months to days or weeks.
Old lessons, new wave
The other groups spent less time on tools than on mistakes. The group on software's lessons wanted hardware founders to talk to customers earlier and improve the product while selling it. Where adoption is the bottleneck, they suggested as-a-service models, as companies selling energy and heat as a service already do by owning and operating the assets. A participant said that a few years ago everyone wanted to build a giga-factory of something, and now founders are looking for new routes to market.
The cleantech group said investors in the last wave, in hydrogen for example, treated policy as the pull for demand when it can only push. The market has to rest on the unit economics of the product. After the ESG backlash, and with American policy proving choppy, investors trust green premiums less and favour companies with near-term revenue. The group was unsure about AI's energy demand, and about whether climate companies pivoting to serve data centres are making a sound long-term bet.
Asked about exits, with IPOs difficult and many European industrial acquirers under strain, an investor said the last wave put too much capital into slow, sticky markets at valuations built for software-like growth. Lean rounds keep more exits open. Strategic buyers tend to acquire at one of two points. The first is when a start-up has something unique at the right moment. The second comes later, when it has customers and is starting to own a market. "It's a company sell, not a tech sell," the investor said of the second.
Where AI and cleantech differ
The fourth group asked whether AI will repeat cleantech's mistakes, and mostly thought not. AI sells on economics, while cleantech sold on emissions first. AI also iterates at a speed cleantech never approached. It will be overinvested and many will lose money, the spokesperson said, but its market is large enough to absorb that. On defensibility the group sided with cleantech. With superintelligence, anything built in software can be taken away, while something built from atoms endures. The one shared weakness is image. Cleantech is caught in America's partisan divide, and a public backlash against AI is building fast.
One of the hosts put the bet plainly. If AI can compress development, from discovery through design and production to quality control, into about ten years, the next cleantech wave looks venturable. The report-backs suggested AI can supply the speed. The atoms that give deep tech its durability still have to be tested in the physical world.
This Ripple was hosted by Michelle Robson (Odyssey Ventures) and Lisa Schneider (Aramco Ventures) at The Drop 2026 on 16 September.