The Art of a Good Unicorn: The presence of AI doesn’t guarantee defensibility

Investors face AI hype FOMO-are they upskilling to spot real tech or just chasing trends? Can financial pros evolve into true tech-savvy backers amid the noise?

At a time when no conversation is complete without mentioning the word “AI”, temptations to present themselves as companies leading in the space have never been severe, especially from the side of startups. But what about those who are putting money behind these ventures; are they equipped enough to cut through the AI (artificial intelligence) washing and identify those upstarts which will stand the test of time? Have they done any up-skilling themselves especially when the rate of success is hugely dependent on technical understanding? How are the generalist fund managers who are experiencing FOMO (fear of missing out) in the midst of the AI tsunami taking critical investment decisions ? Are these investors who are well equipped at financial engineering willing to reinvent themselves as tech investing mavens?

Rahul Bhasin underwent a course at the MIT to gain a deeper understanding of new-age technologies. (HT sourced photo)

We speak with private equity veteran Rahul Bhasin, managing partner of Baring Partners Private Equity, managing upwards of a billion dollars in assets under management. Bhasin underwent a course at the Massachusetts Institute of Technology (MIT) to gain a deeper understanding of new-age technologies like generative AI and to build a fundamental understanding of the same. Edited excerpts from the interview below:

How has your thesis around AI evolved in the past 18 months, especially as open-source models become increasingly competitive with proprietary ones?

Over the past 18 months, our thesis has shifted from focusing solely on model supremacy to understanding deployment context, data moats, and vertical differentiation. The rise of open-source models has commoditized core capabilities, so we now look for startups that can create enduring value through proprietary data, strong distribution, or deep integration into industry workflows, rather than just building yet another foundational model.

What have been the biggest lessons from your previous investments in AI, particularly in understanding where real value lies versus inflated hype?

The biggest lesson: the presence of AI doesn’t guarantee defensibility. We’ve learned to look beyond the tech demo and assess whether AI meaningfully transforms unit economics, customer stickiness, or productivity. Hype can mask fragile business models — we now focus much more on execution, distribution, and repeatable use cases that show real traction.

Is the private equity and VC ecosystem, as it stands, structurally equipped to evaluate the technical depth of modern AI models? Or are investors still playing catch-up?

The ecosystem is catching up — but unevenly. Some firms are building real in-house technical depth, while others still rely heavily on external validation. We’re in a transitional phase where strong technical due diligence is becoming a competitive advantage, especially when investing in companies claiming frontier capabilities. A continuous learning mindset and culture is critical for any investing firm which succeeds in the long term.

Are you seeing investors actively upskilling in AI literacy, from LLM (large language models) architectures to benchmark evaluation, or are they largely outsourcing technical validation to consultants?

There’s definitely a push toward upskilling, especially around LLM architectures, inference costs, and fine-tuning techniques. However, in high-stakes or deep-tech deals, most still complement that with specialist consultants or internal AI advisors. The best firms are doing both — learning the language while still leaning on experts for depth. I have an unfair advantage that some of my closest friends and my son are experts in ai and all I have to watch is whether they roll their eyes when I bounce off an idea with them.

How do you differentiate between a true AI-first startup and a wrapper product with API (application programming interface) dependency? Has that distinction become harder or easier in 2025?

It’s actually gotten easier in some ways. We look at four signals: (1) how dependent the company is on third-party APIs; (2) whether they’ve built any of their own training or fine-tuning infrastructure; (3) if the AI materially alters the product’s value proposition; and (4) if there’s a clear feedback loop or proprietary data layer. A “wrapper” product rarely scores high across all four.

As AI starts to reshape traditional industries, from insurance to manufacturing, how do you approach investing in ‘AI-for-X’ versus foundational AI? Which layer do you believe will yield the most sustainable returns?

We’re more focused on the ‘AI-for domain specific application layer particularly vertical applications where AI becomes embedded into legacy workflows. These tend to have stronger defensibility via industry-specific data and stickier B2B (business-to-business) relationships. Foundational AI is capital-intensive and winner-take-most; unless you have a truly unique angle, it’s hard to compete with hyperscalers. Also in line with human adaptation of any transformative new technology there is always an investing bubble which creates huge foundational capacity and lowers the costs for people to build out application use cases.

AI is moving faster than most industries’ ability to adapt, including yours. What’s the biggest blind spot PE (private equity) investors have today when it comes to evaluating AI-native businesses?

The biggest blind spot is overvaluing technical potential without understanding commercialization risk. You can have a brilliant model, but if go-to-market is weak or customer onboarding is friction-heavy, the business won’t scale. Investors sometimes underestimate how long it takes for traditional industries to absorb change — and how critical domain expertise is in that process.

(Shrija Agrawal is a business journalist. The views expressed are personal.)