Oct 16, 2026 · 1h 6m

AI abundance pushes venture capital toward deep tech

E443: Why the Next Trillion-Dollar Startup Won’t Be a Software Company

As software becomes easier to build and harder to differentiate, investors face a higher-stakes shift toward hardware, infrastructure, and long development cycles.

3 key takeaways

1Software commoditization may redirect venture opportunity toward energy, manufacturing, defense, mining, and other physical industries.
2Deep-tech investors must assess technical progress, regulatory timelines, financing needs, and founder depth before revenue appears.
3The thesis remains fragile: several prominent failures in robotics, chips, or diagnostics could chill funding across the sector.

Don't miss

Leo explains why a next-generation sawmill became a favorite investment despite initially looking more like private equity than venture capital.

The brief

Leo Polovets argues that AI is making software abundant and commoditized, shifting scarce value toward physical infrastructure, hardware, and deep technology.

The investment challenge changes when companies need years of technical and regulatory work before revenue: investors must judge teams, milestones, financing, and failure risk.

A next-generation sawmill illustrates the opportunity in overlooked industrial markets, where limited startup competition can matter more than fashionable consensus.

Leo favors more shots on goal for binary hardware outcomes, while arguing that sufficiently large exits can justify early dilution and heavy capital needs.

The thesis has a dangerous weak point: failures in robotics, chips, or blood diagnostics could trigger reputational spillovers and freeze deep-tech funding for years.

What was said on this episode

13 statements · 9 positive · 2 negative · 2 neutral

Leo Polovetson AI-driven software commoditizationNegative0:00
AI is commoditizing software development.
“Building software is becoming more and more commoditized.”
Listen at 0:00
Leo Polovetson HardwarePositive0:10
Software commoditization will shift value toward hardware.
“I think it just means more value will move to other areas, and hardware is, like, to me, the next natural one.”
Listen at 0:10
Leo Polovetson AntaresPositive14:04
Antares nuclear reactors will probably begin selling within 12–24 months.
“They'll probably be selling in the next, you know, 12, 18, 24 months.”
Listen at 14:04
Leo Polovetson Ambitious hardware companiesNeutral14:10
Ambitious hardware companies typically need about five years to deploy units.
“you know, about five years from zero to like having their units deployed in the field.”
Listen at 14:10
Leo Polovetson Hardware businessesNeutral15:11
Many hardware businesses take seven to eight years to become public companies.
“the overall time period to go from, you know, like start to public company is still seven, eight years for a lot of these businesses.”
Listen at 15:11
Leo Polovetson Deep-tech companiesPositive39:59
Bias for action and execution are especially important in deep-tech companies.
“Bias for action execution are really important on the physical side”
Listen at 39:59
Leo Polovetson Data-center power technologyPositive41:50
Data centers will buy less-efficient power technology if delivered much faster.
“if you make something that's 80% as efficient, but you can get it out in six months, like data centers will line up to buy it from you”
Listen at 41:50
Leo Polovetson AI-assisted data-center designPositive43:32
AI and software could reduce data-center design time from a year to weeks or months.
“what if you can use AI and software to bring that year down to a few weeks or maybe a couple months?”
Listen at 43:32
Leo Polovetson AI infrastructure energy availabilityNegative44:03
Energy availability is currently the biggest physical bottleneck for AI infrastructure.
“I think energy is probably the biggest one”
Listen at 44:03
Leo Polovetson Specialized LLM-inference chipsPositive46:33
Specialized LLM-inference chips could be 10–100 times faster on supported tasks.
“the things it can do, it's going to be 10 or 100 times faster.”
Listen at 46:33
Leo Polovetson HumanoidsPositive48:29
Humanoids may become a trillion-dollar category over 10–20 years.
“over 10 or 20 years, there's a good chance that like humanoids might be like one of the, you know, new like trillion dollar categories”
Listen at 48:29
Leo Polovetson Task-specific robot armsPositive48:50
Task-specific robot arms will be more efficient than humanoids for narrow tasks.
“robot arms that are specialized and trained for just that one task are going to be way more efficient”
Listen at 48:50
Leo Polovetson Humanoid companiesPositive49:57
One or two humanoid companies will become huge winners within 10–20 years.
“there will probably be like one or two huge winners in that space in the next 10, 20 years.”
Listen at 49:57

Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.