Why we are focusing on Atomic Layer Deposition
Paris, May 21st 2026
Building a deep tech startup on AI for materials discovery sounds like a clear enough thesis. The hard part is deciding where to focus.
Over the past year, we evaluated various markets across electrolysis, mining, batteries, catalysis. This process led us to a clear conclusion: Atomic Layer Deposition (ALD) is where our technology can have the most impact today.
The filter that eventually mattered wasn’t market size or growth rate. It was simpler: at our current stage, is this industry pulled by value or pulled by cost?
Most materials markets are cost-driven, and we worked through them one by one:
- Catalysis. Margins aren’t what you’d expect. IP lives more in process integration than molecule patents, and a new catalyst only matters if a large plant trusts you enough to run it at scale. Few startups clear that bar quickly.
- Batteries and energy storage. The LFP wave out of China compressed much of the near-term innovation space down to optimization, doping, and coating. There is still meaningful room for new chemistries, especially in interfaces and next-generation batteries, but the path to value capture today is harder and more cost-driven than what we are looking for.
- Electrolysis. This domain is in retreat after a wave of over-investment. We identified complex problems to optimize on electrode design and coating, but the market itself was pulling back.
- Mining and industrial chemistry. These markets are characterised by very thin margins, very high volumes, almost no room for differentiated chemistry to capture value. The big players in mining don’t do chemistry discovery, and industrial chemicals are dominated by cost logic in a way that leaves little oxygen for a startup proposing a new molecule.
The pattern was consistent: volume logic dominates. The path from a discovered molecule to captured value is long, uncertain, and crowded.
What we were actually looking for, and took a while to name, was a market where a single molecule could be genuinely proprietary, where customers had margin to spend on performance, and where the chemistry was complex enough that computational tools would give a real edge over trial and error. Defense is one example. Specialty surface treatments another. Semiconductors is the biggest.
That’s where ALD lives. Atomic layer deposition is the process used to deposit ultra-thin films, sometimes a single atomic layer at a time, onto surfaces. It’s how the industry builds the insulators, barriers, and gate oxides inside modern chips. Without ALD, you can’t make a 3nm transistor. Without better ALD precursors, the molecules that drive the deposition chemistry, you can’t push further. Every new node, every new memory architecture, every advanced packaging design that the next decade of compute depends on runs through this chemistry.
There’s also an ecological case here that took us a while to fully internalize. Every gain in transistor density and gate quality translates directly into lower energy consumption per computation. Data centers already account for a meaningful and growing share of global electricity demand. Better precursors mean thinner, cleaner, more uniform films, which means more efficient chips, resulting in less energy burned per token, per query, per training run. At the scale the industry is operating at, it’s not a marginal effect.
And importantly: ALD itself isn’t confined to semiconductors. The same atomic-scale surface engineering is starting to show up in battery material and electrode coatings to extend cycle life, and in catalyst supports where a few atomic layers of the right oxide can change selectivity dramatically. Some of the markets we explored might come back to reach later, through a different technological entry point, as ALD matures for those applications.
The molecules are complex enough that computational tools give a real edge over traditional trial and error. Patent value is real. A novel precursor with the right thermal window and film quality is genuinely protectable. And the bottleneck, it turns out, isn’t synthesis: it’s predicting synthesizability and process compatibility before you ever run a reaction. That’s exactly what our DFT pipelines and ML models are built to attack.
Which brings us to the decision behind this post:
We’re going all in on ALD.
Not as a short-term experiment, or a temporary beachhead, but as the thing we do. As the market where our technology currently has the strongest fit and where computational chemistry can create a real competitive edge.
The AI for materials space has seen rounds raised at valuations that, frankly, don’t match the maturity of the science yet. Some of those companies are trying to be horizontal across every materials market at once. At our size, that’s not a strategy, it’s a way to be mediocre at five things. We’d rather be the best in the world at one. ALD is that one, and we intend to stay the best in the world at it.
We wouldn’t have found any of this without spending a year looking at markets we eventually walked away from. The wrong bets clarified the right one.
