What It Takes to Build a Unicorn in the AI Era

“The strongest founders don’t just understand AI; they understand the industry they’re selling into. They know the workflows, regulatory constraints, purchasing dynamics, and bottlenecks because they’ve lived them.”

Angela Lee, Professor of Professional Practice, Finance Division, Columbia Business School

The average AI unicorn founder is 29 years old. Three years ago, that number was 40. Antler’s recent report analyzing more than 1,600 unicorns shows that AI companies are reaching billion-dollar valuations two full years faster than every other sector. Category-defining startups are now emerging from more than 300 cities across 45 countries. Great companies are no longer the exclusive product of seasoned executives and a handful of zip codes.

Solo founders now represent roughly one-third of all VC-backed companies. ARR per employee is rising measurably across early-stage AI startups. Founders are arriving at the starting line with less corporate experience and building further, faster than prior generations. 

These shifts are reframing what it means to build a great company. Rather than centering experience, pedigree, or geography, today’s breakout founders are optimizing for speed, capital efficiency, and distribution from day one. The data suggests that building a unicorn in the AI era requires more than a good model. It requires a fundamental understanding of the structural forces that make scale possible in the first place. 

To better understand what those forces are, how the founder profile is evolving, and what separates enduring companies from well-funded ones, we spoke with Angela Lee, professor of venture capital at Columbia Business School and founder of 37 Angels, which has evaluated more than 20,000 startups and invested in more than 120.

Angela Lee is a faculty member at Columbia Business School, where she focuses on leadership, strategy, and venture capital. With two decades of professional experience in entrepreneurship, innovation, and strategy—including roles in product management and strategy consulting at McKinsey—she brings a wealth of expertise to her teaching.

Beyond academia, Lee is an active entrepreneur who has founded several educational ventures. She also founded 37 Angels, an investment firm that trains new investors via boot camps and has assessed over 20,000 startups, investing in more than 120. Her contributions have been recognized by various outlets, including Inc., Entrepreneur Magazine, and Crain’s, and she is frequently sought out for her insights by major media networks like CNBC and Bloomberg TV. She holds degrees from UC Berkeley and Columbia Business School.

The speed at which AI startups are reaching scale is not accidental. According to Lee, three structural forces have fundamentally changed what it costs, and how long it takes, to build a company worth building.

The first is team size. Tracking ARR per employee as a core metric across her portfolio, Lee says the shift is measurable. “We’re seeing a measurable rise in solo founders in venture-backed companies. Roughly one-third of VC-backed founders today are solo, compared to much lower levels historically.” Smaller teams mean lower burn, faster iteration, and more capital-efficient early traction.

Second, we can consider distribution. Earlier technology cycles required heavy enterprise sales teams, physical infrastructure, or both. Lee argues that calculus no longer applies. Many AI products are API-based, embedded directly into existing workflows, or distributed through developer ecosystems. Startups reach global customers on day one without the overhead that once made early growth so expensive.

Third, she explains, is infrastructure. Cloud providers, foundation models, and open-source tooling have created a pre-built foundation. In Lee’s view, founders today are starting further down the road than any prior generation. Time-to-product has shortened. Time-to-revenue has followed.

For Lee, these three forces compound rather than operate independently. “That combination means lower burn, faster iteration, and more capital-efficient early traction,” she says. “The result is that startups can demonstrate meaningful traction with smaller teams, less capital, and shorter timelines — which supports faster markups and, in some cases, billion-dollar valuations earlier in the lifecycle.” These economics have opened the door to a new generation of founders building across more geographies than ever before. Who those founders are, and where they are coming from, tells its own story.

The geographic spread of unicorn creation is still in its early stages, and the structural conditions driving it are only getting stronger. Founders anywhere in the world can now build on top of existing models and cloud infrastructure, reaching customers without the proximity to capital and talent clusters that earlier generations depended on. 

“I expect more globally distributed AI startups,” Lee says, “particularly in markets with strong technical talent but historically less venture concentration.” 

The Antler data already reflects this. Unicorns are emerging from more than 300 cities across 45 countries, and Silicon Valley, while still significant, no longer holds a monopoly on where breakout companies are born.

The foundation layer tells a different story. Lee is direct about where capital is actually flowing. “Seventy-five percent of all VC went to 30 VC firms and 50 percent went to nine VC firms in 2024,” she notes. Training frontier models requires enormous compute and capital, and that reality concentrates early value among a small number of players and the mega funds backing them. Geographic democratization at the application layer exists alongside significant capital concentration at the top.

Who is doing the building is changing too. More founders are arriving directly from research labs and universities, earlier in their careers and without traditional corporate pedigrees. Lee sees this as a natural consequence of what AI products actually demand. Deep technical credibility matters more when the product itself is model-driven, and researchers are increasingly willing to start companies before accumulating the resume that prior generations considered a prerequisite.

Yet the data surfaces a gap that the broader optimism around AI’s democratizing potential does not explain away. Women represent just 6 percent of unicorn founders. Meanwhile, immigrant founders and repeat entrepreneurs continue to play outsized roles in company creation, suggesting that certain pathways into venture-backed entrepreneurship may remain more accessible than others. The opening of geography has not necessarily produced a corresponding opening of access. 

New cities and new countries are entering the map of unicorn creation, but the demographic profile of who gets funded, who gets to the starting line with the right networks and capital relationships, has moved far more slowly. For Lee, the distinction matters because the structural forces making it cheaper and faster to build a company do not automatically dismantle the structural barriers that determine who gets to try. What founders do share, wherever they come from and whatever barriers they navigate to get there, is a specific set of traits that determine whether their companies endure.

The compression of timelines and the democratization of infrastructure have made it easier than ever to build a company. They have not made it easier to build one that lasts. Lee is precise about this distinction. The surface area for failure has expanded alongside the surface area for opportunity, and the startups that endure share a specific set of traits that have less to do with their models and more to do with the judgment of the people building them.

Domain expertise sits at the foundation. Lee is emphatic that technical fluency alone is not enough. “The strongest founders don’t just understand AI, they understand the industry they’re selling into,” she says. “They know the workflows, regulatory constraints, purchasing dynamics, and bottlenecks because they’ve lived them.” In her view, AI is a tool. Domain knowledge determines whether a tool creates real value or an impressive demo that never converts into a durable business.

Problem definition is where many AI startups fall short, according to Lee. “Many AI startups are building impressive demos. Fewer are solving a painful, budgeted problem.” Enduring businesses, in her experience, are anchored in clear ROI and a genuine understanding of stakeholder needs across the full market, not just the early adopter willing to take a chance on a new product.

From there, the question shifts to distribution. Lee frames it as what ultimately separates companies that win from those that plateau. The model working is a baseline, not a differentiator. “How does this company reliably acquire and retain customers?” is the question she wants answered. Advantages in community, partnerships, embedded workflows, and network effects are what create defensibility. Without them, growth is expensive and fragile.

Underlying all three traits is something Lee argues has not changed despite everything else that has. The core venture criteria remain consistent. Exceptional ability, speed of execution, strong communication, and the capacity to attract both talent and customers are still what she looks for first. 

“In many ways, it hasn’t shifted as dramatically as the headlines suggest,” she says. The fault line she draws is between companies that optimize for valuation velocity and those that optimize for customer value and repeatability. In her experience, only one of those orientations produces something built to last.

The distinction between valuation velocity and enduring value raises a larger question about how the AI landscape is being framed. There is a framing problem at the center of how most people talk about AI investment, and Lee does not shy away from naming it. 

“Saying you ‘invest in AI’ in 2026 is a bit like saying you invested in ‘technology’ in 2010,” she says. 

The category has become so broad it has stopped being analytically useful. AI is not a sector with defined boundaries and a discrete set of players. It is an enabling layer that is quietly embedding itself across healthcare, finance, logistics, education, and manufacturing. Nearly every industry will be touched by it. Most already are.

That reframe has direct consequences for how founders should think about what they are building and how investors should think about what they are funding. The question Lee returns to is not whether a company uses AI. That bar is already too low to mean anything. The real question is what problem the company solves and for whom. Customer value, unit economics, defensibility, and execution are the measures that matter. They always have been.

Where durable value accrues over time is something Lee has a clear view on. The foundation layer, the company’s training frontier models, will continue to attract enormous capital. But the customer relationship is where she expects lasting value to concentrate. “We’ll likely see value shift from ‘who built the model’ to ‘who controls the customer relationship,'” she says. Companies embedded in enterprise workflows, consumer platforms, and industry-specific systems are better positioned to build the kind of switching costs and network effects that make a business genuinely hard to displace.

The “AI startup” label, Lee suggests, will eventually become a relic. Companies will stop being evaluated as participants in an AI wave and start being evaluated the way any business has always been evaluated. The founders who understand that earliest, who are building for customer value and repeatability rather than narrative and valuation momentum, are the ones most likely to still be standing when the label fades. Lee has spent years watching founders chase the wrong signals, optimizing for what impresses investors in a given moment rather than what creates lasting value for customers. 

Ultimately, new trends in unicorn timelines and founder profiles tell a story about speed and access. But it does not promise whether the companies being built today will matter in ten years. That question, Lee argues, has always come down to the same thing. Not the model, not the valuation, not the moment. The problem being solved, and how well.