What Deep Tech Climate Investing Taught Me About Backing Software and AI
A few years ago I was helping build the thesis for what became a $500M climate tech fund at BDC Capital, backing Canadian companies working on carbon removal, materials science, and industrial decarbonization. Today, as a Partner at Pender Ventures, I spend my time with software and AI-native founders. On paper those look like two very different jobs. In practice, the move taught me more about what actually makes a great investment than either world could have on its own.
A few of the biggest lessons:
1. Moats look completely different.
In deep tech, the moat is often the technology itself: hard IP, patents, years of R&D that a competitor genuinely cannot replicate quickly. It’s binary in a way, either you have the breakthrough or you don’t. In software and AI, the moat is rarely a single piece of IP. It’s a more nuanced mix of distribution, data advantages, workflow lock-in, speed of iteration, and how well a team compounds small advantages over time. Underwriting that takes a different muscle, you’re evaluating a system, not a patent.
2. Opex discipline matters the same amount. Capex does not.
Every good operator, regardless of sector, needs to run a tight ship on opex, that discipline doesn’t change. What’s completely different is capex. Deep tech companies often need significant capital before they know if the physics or chemistry even works at scale: pilot plants, hardware, long validation cycles. Software and AI companies can get to signal with a fraction of that capital. The efficiency bar is the same; the amount of capital required to reach it is not even close.
3. Bootstrapped hustle vs. big spend, grant-fueled R&D.
Some of the best software and AI-native founders I meet today have gotten remarkably far on very little capital: lean teams, fast shipping, real revenue before a big raise. Deep tech is a different game almost by design: heavy upfront R&D spend, longer time horizons, and, in Canada especially, a real reliance on non-dilutive capital and grant dollars to bridge the gap before commercial revenue shows up. Neither approach is better. They’re just solving for different constraints.
4. Time horizons and risk profiles run on different clocks.
Climate and deep tech bets are often measured in a decade or more, with technical risk resolving in stages. Software and AI move at an entirely different speed: the “state of the art” can shift meaningfully in a matter of weeks, not years. That compresses both the upside and the downside, and it changes how you think about timing a check.
5. Physical AI is where these two worlds are starting to collide.
This is the one I am still chewing on. Physical AI, robotics, autonomy, anything where software meets atoms, sits right at the intersection of everything above. It needs deep technical R&D and real capex like climate and deep tech did, but it moves and iterates with the speed and software-like economics of an AI native company. So where does the moat actually sit? Is it the hardware, the data the system generates in the real world, the software layer sitting on top, or the integration of all three at once. And importantly, who is the customer for that software layer, is it the OEM’s, the end user, or the fleet operator running the machines day to day, since each one buys on a different basis and pulls the moat in a different direction. I do not think there is a clean answer yet, and that is exactly why it is interesting. My instinct is that the moat will end up being some combination of proprietary real world data plus the speed of the iteration loop, but I am watching this space closely rather than pretending I already have it figured out.
What hasn’t changed:
The fundamentals I care about most as an investor are the same regardless of sector. Team quality, specifically, can they pivot and figure it out when the market shifts underneath them, because it always does. Whether the technology is genuinely differentiated or just interesting. And a real, honest answer to how big the market actually is.
Closing Thoughts
Looking back, I feel genuinely fortunate about the timing of this move. Right now, in AI, everyone is learning in real time: the operators, the investors, the incumbents. Things are changing week by week, sometimes faster. There’s something freeing about that. Nobody has a ten-year head start on this wave, so the advantage goes to whoever learns fastest and adapts with the most conviction.
Our Thesis at Pender
That is exactly what we look for at Pender. We back companies at the commercialization and scale inflection point, teams that have proven the early signal around product market fit, generally $2 to 3 million or more in annual recurring revenue, and are ready to scale. Past that point, what we spend the most time on is rarely the product alone. It’s the market size and whether the opportunity is actually as big as it looks on paper. It’s capital efficiency, whether the team gets real output for every dollar they raise. And it’s the team itself, whether they can read a market that is shifting under them in real time and adapt faster than anyone else in the room. We believe the best companies are built by bold ideas and diverse, multi-faceted leadership teams that reflect the world they are building for, and that conviction has only gotten stronger since I made this move. The tools change. The sectors change. What we look for in a team and the road map does not.
Back to top ↖