After SaaS: The Physical Decade

A photonics founder expects AI to make their simulations perhaps 30 to 40% faster. The chips still have to be sent out for manufacturing, a design that works first time is likelier but not certain, and a team of about ten still needs the specialists who run the tools. The capital does not shrink. It was the clearest version of what founders reported at the Ripple on hardware after SaaS. AI takes time out of designing climate hardware but not yet the money, and that sent the discussion looking for moats elsewhere.

Mark Windeknecht of World Fund had warned at the start that the gains may go mainly to incumbents, the companies that already own the relationships and the access to hardware. Software's moat, one of the hosts said, had all but gone from their deal flow. What counts now is integration with the customer and their data. An investor added that listed SaaS stocks have recovered from this year's sell-off, because distribution and customer relationships are hard to displace. One participant doubted software was ever the moat. It had always been about unique data and distribution, they said, and another added network effects.

where even we don't know how things work, so why would an AI know?

Where the tools help

Pina Fritz of Deep Science Ventures, who works on AI for the earliest stage of hardware development, walked through tools along the chain. Flow Engineering tracks requirements and early engineering decisions. Physics-based AI can interpolate between simulations instead of running every model in full, and Siemens is pushing digital twins. A materials founder said their company builds system-level targets and techno-economics into material design, so the materials stop hitting bottlenecks once in use. One of the hosts described a portfolio company pairing an automated lab with machine learning to speed up catalyst work that used to take hundreds of repeated tests.

An organic-battery founder was the most precise about limits. Be clear about what AI can and cannot do, they said, and stop what fails. It does not yet help find new battery molecules. It does track what the company promised investors and grant bodies, flag milestones that are slipping and take over junior, non-core roles. The company also built its own battery-cycler software and electronic lab notebook, and its lab engineers now spend 90% less time keeping records. Pina agreed that AI is powerful for optimising well-understood technology, but not at the edge of science, "where even we don't know how things work, so why would an AI know?"

What does not shrink

One investor hoped faster cycles would ease the old clash between a fund's life and the time hardware takes to develop. Another participant pointed out that what AI removes is mainly labour, so the saving depends on how much of a company's costs were people and how much were materials. An IoT founder was blunter. Some things cannot be hurried, they said, and sometimes it is "like hiring two pregnant women to give birth in six months." At their stage the moat is the customer relationship, and AI helps by speeding up each round of changes made on customer feedback.

A moat with a small m

The defences on offer were familiar. One participant named patents and early data, from building the simplest product that yields the most data soonest. Another argued for owning assets and more of the supply chain. An investor said patents protect only where they are respected, so some companies they look at keep production processes as trade secrets. Pina held that creativity, asking the right question at the right moment, is something AI cannot replicate. The investor who had asked about moats disagreed. The limits of today's models erode by the day, they said, and it is hard to think anything is that precious a decade out.

A founder building across space, AI and IoT asked whether the line between hardware and software companies still means anything. Someone pointed to Toyota's Woven City in Japan, where companies can pilot products for cities, as an incumbent reaching beyond its sector. Others said start-ups should stay narrow and move up or down the chain later. A participant then summed up. Solving one problem really well for a specific niche seems to be the moat, even if it is not one in the old sense. The reply was that the low-hanging fruit has been picked, so staying on top of a problem is its own kind of moat, one with a small m. An investor drew the lesson for their trade. If the moat is how fast a team stays on top of the problem, underwriting the founder matters more than ever.

The last challenge came at the very end. One participant said the discussion had dwelt on design tools and administration while climate tech has to rebuild the world's factories to run on electricity instead of coal or gas. Robots that can build factories are perhaps ten years away, they said, and those who build the factories in the meantime will own them. On that view the moats are being dug by construction workers, manufacturing staff and R&D technicians. A founder replied that their company already builds its hardware with robotics.

This Ripple was hosted by Mark Windeknecht (World Fund) and Pina Fritz (Deep Science Ventures) at The Drop 2026 on 16 September.