Introduction

"It is always difficult to be born. You know the chick does not find it easy to break his way out of the shell." - Herman Hesse

With the pace of model development continuing unabated, AI models are seeing unprecedented capability improvement. What was state-of-the-art in September becomes a rounding error by March.

As we argued in Software is Eating the Work, we are now seeing a phase shift where AI systems do the work of humans, rather than just augmenting them; the line between "tool" and "worker" is blurring.

Over the last six months, public markets have started to realise the implications of these shifts, catalysed by the launch of Claude Cowork on January 12, 2026. The ensuing selloff constituted a flight to safety, with investors seeking shelter from the uncertain impact of AI. Volatility continues to be high as individual stocks are repriced in light of the changes erupting from San Francisco.

While the public markets roil, the reaction in the private markets has been more muted. Trillions of dollars in dry powder are sitting idle, as investors attempt to understand the implications of these shifts. Some have unknowingly already lost billions, with AI invalidating the assumptions they made prior to ChatGPT. Many firms are offsides, with their portfolios underwater and no plan to get back in the black.

But while the flight to safety is sensible in the short and medium term, management and ownership at even the most effected companies will not go down without a fight. Over the next few years, some of these companies will truly transform their businesses, as their peers fail to adapt to new realities.

Private equity investors, much more so than their colleagues in the public markets, can be both mimetic and generative - they underwrite the business they see today, but also have the unique opportunity to transform them into something new. For all the noise about "value creation", that lever has never been more important.

Over the past year we've been working with the world's largest PE firms to deploy AI into their portfolio companies. Through our work, we've developed unique insights on how AI will impact investing and operating in the new economy. We formalized our insights and built a scoring framework for the whole PE universe, covering 5,425 companies across 147 funds. We measured the two things that actually matter when evaluating AI as an investor; and are now releasing that framework, along with its application to those companies to the world.

The first variable is durability: can this business survive what AI does to its industry? Regulatory moats, switching costs, physical assets: a business that scores high on all five has walls built from materials AI can't breach.

The second is opportunity: how much new value can AI create inside this business from where it is today? How cognitive are the workflows, how rich is the data, how much of the production function can be compressed or replaced?

We can plot both factors on a graph, and divide the universe into four quadrants — split at the median durability and median opportunity — with every company represented by a dot on the (interactive!) graph below.

Applying a consistent methodology across the universe of sponsor-backed companies, it becomes clear that durability and opportunity are inversely correlated.

The inverse correlation between durability and opportunity makes a lot of sense. Physical assets, regulatory moats, capital intensity, and low cognitive-labour share protect businesses from AI disruption but limit the opportunity for AI transformation. Conversely, the attributes that make a business ripe for transformation are the same attributes that make it vulnerable to disruption.

Even more interestingly, durability and opportunity are each correlated with recent public comp price moves. Durability alone explains nearly a third of the variance in 6-month comp returns (r² = 0.29); opportunity, with its negative correlation, explaining 23%. But the simple difference — durability minus opportunity — is more predictive than either score individually (r² = 0.33), and the result is statistically solid across all 5,425 companies.

To take from Graham, we call this difference the Margin of Safety.

Crucially, though, this means that public markets are punishing high-opportunity, low-durability businesses: the very businesses which are best positioned to benefit from AI transformation. Some of these, to be sure, will be donuts. But in both public and private markets there are companies, which we call Hail Marys, which have the opportunity to AI transform and win in the new world. For the Hail Marys, this is their Super Bowl.

The Margin of Safety clearly differentiates between verticals: basic materials, food and beverage, and energy are solid, while HR, legal, and insurance are exposed. These are the same stocks affected by the SaaSpocalypse; some are cooked, but some of them will end up as winners: deserved stars.

Firm-level data

If we look at the simple average performance for the public comps of each portfolio company, and then take a simple average across the portfolio of each fund, the relationship between the Margin of Safety and public performance is even more compelling.

The primary factor explaining this performance is the segement the business performs in. Companies in real world industries have fared well, while companies that are cognitive have sold off. At Antin Infrastructure Partners, the Margin of Safety is +45.7 points. Every company in its 30-company portfolio has durability that far exceeds opportunity. Atlas Holdings sits at +40.3. Wynnchurch Capital at +40.1. These firms own metal fabricators, pipeline operators, power plants, and chemical companies.

Now look at the other end. At Updata Partners, the average portfolio company has a Margin of Safety of (11.4) points. At Stone Point Capital, (10.5). At Long Ridge Equity Partners, (11.3). While there is plenty of opportunity for their portfolio companies to win, their concentration in software and financial services leaves them exposed, with most companies scoring as either a Hail Mary or Cooked.

The chart below gives a granular look at the portfolio of some of the largest funds:

We can extend that out to look at every fund in the dataset, based on how many of each firm's companies fall into each bucket. Thoma Bravo has a lot of stars, but few fortresses; Appian and KPS Capital Partners' portfolio is entirely fortresses.

Why Now?

With public markets fleeing to safety, there are opportunities for forward thinking investors to find value in the Stars and Hail Marys. Even in the most exposed industries - marketplaces, brokers, consultants, software platforms - companies will find a way to win.

We're particularly excited about marketplaces and brokers that can induce demand, consultants that can serve new markets at lower cost and higher volume, and software platforms that cannibalise their services ecosystem.

Some of these businesses will win on durability - regulatory moat, proprietary data, or a refuge in high-end workflows - but the most exciting AI transformation opportunities are here. These companies are now in play.