At Web Summit Vancouver‘s Center Stage panel on where smart money is going in 2026, Techstars Founder and CEO David Cohen sat alongside Salil Deshpande of Uncorrelated Ventures to dig into the AI cycle, the flight to quality, and what is actually worth backing for the long arc.
The takeaways are useful whether you are raising your first round or deciding what to build next.
We Are Still Very Early
Both David and Salil put the AI cycle in the early innings. Today, roughly 1 percent of the world is actively using these tools. The next chapter is the shift from chat-based AI to intelligence embedded in real workflows, products, and services. David framed it as the second major shift of his career, with a 20- to 30-year arc ahead.
Translation for founders: there is plenty of time to build, but the bar for what counts as a real company is rising fast.
Flight to Quality Means Going Lower, or Going Deeper
Salil made the sharp opening point: quality looks like gross margin. The middle layer, the AI-producing companies, is full of negative or thin margins, including the model providers themselves. That is pushing capital lower in the stack, into infrastructure and hardware, where AI is now a big-budget, price-insensitive customer. As a former software investor now writing hardware checks, Salil has been following his own advice.
David’s version of the flight to quality is about going deeper, not just lower. At Techstars, the focus is on verticals and categories that a 15-year-old with a laptop cannot copy this weekend:
-
Hardware and physical connection to the world
-
Space and other industries with access and regulation moats
-
Healthcare, with proprietary data loops
-
Categories with humans in the loop who genuinely shape the outcome
Defensibility in Three Questions
Salil offered a useful diagnostic for application-layer startups: Is your product formed on top of a database? Who does the data belong to? How hard is the schema to migrate out of?
If you are still pitching beautiful software with a thin AI wrapper, that conversation will be short. Investors want to see proprietary data, compounding loops, and the actual structure of the asset, not claims about it.
Sell Outcomes, Not Seats
David’s strongest line was about pricing models. The future is pay for outcomes, performance, and utility, not pay per seat. The new mode is software you do not see: agents that consume intelligence and deliver work on your behalf. If your business still asks customers to log in and click around, you are betting against the direction of the market.
What to Avoid, and Where to Look
The do-not-invest list from David and Salil was consistent: thin wrappers, just-software pitches, easily-extracted data, forms over a database.
The bets that excite them: infrastructure beneath AI, hardware reborn, and verticals where the asset compounds. Asked for one category to back for five years, David named healthcare. Salil picked power and energy.
The Opportunity Inside All of This
Builders who understand this technology have a multi-year window to monetize narrow AI services for real-world businesses, the kind of opportunity venture capital is not built to chase.
Not every great use of AI needs to be venture scale. Some of the best ones will be lean, profitable, and yours alone.
