AI turns growth equity into a manager-selection test

The argument that put growth equity into insurance general accounts was arithmetic: loss ratios in the strategy have averaged 9% since 2014, closer to buyout at 8% than to late-stage venture at 19%, while upside ran near venture capital's. That combination of buyout-shaped losses and venture-shaped gains made a strategy with no obvious income profile legible inside a liability-matched book, and for a general account hunting private-asset yield without a venture-style tail it was the pitch in full.

A review published September 23 by Insurance AUM Journal, 'The Next Test for Growth Equity,' keeps the arithmetic and moves the argument: growth equity remains attractive, but the opportunity set is getting more selective as AI shortens product cycles and compresses the period during which product-market fit remains differentiated. That shift puts underwriting weight onto asset quality and onto a manager's capacity to judge how long a competitive advantage holds.

Both halves of that shift matter to an insurer, and they pull against each other. The review argues downside protection remains built into the strategy, with low leverage, preferred equity and disciplined entry pricing supporting a risk-return profile distinct from both venture capital and buyouts, while AI widens what winners can do: faster growth, better operating leverage, larger addressable markets than the same businesses could have claimed previously. Read together, that produces a wider distribution resting on a similar floor, and a wider distribution is the thing a category average is worst at describing.

That distance shows up first in entry price: StepStone SPI data covering more than 7,500 growth equity transactions from 2010 through 2025 show companies backed by smaller growth funds still trade at meaningful discounts to those backed by larger vehicles. The review attaches a caveat that carries more weight than the finding: the discount should not be read in isolation from the durability of the growth it prices. A cheaper entry into a business whose product-market fit is precisely what AI is eroding buys risk at a lower price, which is not the same as buying less of it.

Downstream, buyers appear to have reached the same filter on their own: growth equity deals are on pace for their highest annual count since 2021, even as software platform buyouts run roughly 20% below the 2018–2025 average. Buyout capital is still financing exits, the review notes, but it is aimed at companies that pair strong growth with a durable competitive position. More activity in the strategy and less in the adjacent one converge on whether an advantage lasts, and on the evidence here the market is repricing durability rather than growth.

The part of the review a general account ought to read twice is that manager selection matters more than it did over the prior decade, and outcomes vary more across managers. Telling temporary product momentum apart from lasting advantage through technical judgment, sector depth and informed networks has become a larger driver of returns than it used to be, and when dispersion widens inside a strategy the category average quits forecasting any particular sleeve; that is the change an allocation process has to absorb. Technical judgment about one company's moat is harder to write into an investment policy statement than a target range, which is why the range tends to get written instead.

There is also a mechanical problem with reading 9% forward, because a figure averaged since 2014 pools vintages underwritten while the software playbook was stable and product-market fit decayed on a familiar schedule. Vintages formed now are entering a different regime, and an average spanning both will flatter the weaker one. That is an argument about the arithmetic rather than something the review asserts; what the review does assert is narrower and harder to argue with: separating a manager's skill from a favorable cycle has become a return in itself.

For an investment committee, the mechanism is straightforward: when AI compresses the window in which a product stays differentiated, the return on a growth portfolio rests less on finding growing companies and more on finding companies whose growth is not about to be copied, repriced or made obsolete. That is a judgment about competitive half-life, made company by company, and no fund-size band or sector allocation performs it. It is also why a smaller-fund discount can be genuine and still beside the point: the vehicle is cheaper, but the judgment it demands is not smaller.

The review's own timeline sharpens the stakes. Its February 2025 installment made the case from a decade of absolute and risk-adjusted returns, and the author now notes the timing of that argument: ChatGPT had already captured the industry's attention and enterprise adoption of generative AI was accelerating, while the consequences for software economics remained largely theoretical. Nineteen months later, the same publication reports investors questioning whether the segment can operate as it has, with sponsor-led software platform buyouts having slowed.

AI-related issuance has crossed 15% of investment-grade debt, and as this publication has argued, that makes power, data-center and private-placement concentration a general-account problem before it becomes a ratings issue. Growth equity is a second door into that room. The concentration it carries lives in a manager's judgment about how long a software advantage survives, and it does not fit the credit toolkit: a rating, a limit, a watchlist. Sizing that judgment is a different exercise from sizing an issuer, and the firms that keep governing the sleeve as diversification are the ones most likely to be surprised by it.

Nine percent was a decade of category arithmetic. The next decade gets written manager by manager, and the general accounts still underwriting the average will be the last to learn which half of the range they own.