Insight Partners manages roughly $90 billion in capital, and its investment team has spent the past two years watching the venture industry converge on a single trade: write checks to OpenAI and Anthropic, or risk looking foolish. Devin Parekh, a partner at the New York–based growth firm, has decided his firm can live with looking unfashionable. In a TechCrunch interview published September 13, 2026, Parekh laid out why Insight is deliberately keeping its AI portfolio wide rather than stacking capital behind the two frontier labs that have absorbed the bulk of late-stage AI funding — and why losing a hot legal-tech deal to General Catalyst did not shake that conviction.
Why Insight Partners Is Betting Against the AI Lab Herd
OpenAI's valuation has climbed into the hundreds of billions of dollars across successive funding rounds, and Anthropic has raised at valuations reported in the tens of billions, making both among the most expensive private companies on Earth. That is the crowded trade Insight Partners AI investment strategy is declining to chase, at least not exclusively. Instead of concentrating capital in the two names everyone agrees are winners, Parekh describes a portfolio designed to hold exposure across a wider set of AI companies — including stakes in competing labs that would ordinarily be seen as contradictory bets.
The logic is not contrarian for its own sake. Venture capital returns are famously power-law distributed: a small number of investments produce the bulk of a fund's gains, which is precisely the argument concentration advocates make. But the counterargument, grounded in portfolio construction theory, is that concentration works only when you can reliably identify the winner in advance. At the frontier-model layer, nobody can. Model quality leadership changes hands on roughly a quarterly cadence, compute costs and talent flows shift underneath incumbents, and the eventual distribution of value between labs and the applications built on top of them remains unsettled. Spreading stakes across rival labs, Parekh suggests, is a way to own the category rather than a single horse.
Industry data supports the premise that the herd is real. PitchBook and CB Insights both documented that AI captured an extraordinary share of global venture dollars in 2025, with frontier model developers alone absorbing a disproportionate slice of late-stage capital — a concentration that has only widened through 2026. When nearly every growth fund's AI thesis reduces to the same two companies, the marginal dollar buys less differentiation and more correlated risk.
The Legora Loss and What It Revealed About the Market
The deal that got away was Legora, the legal AI startup that General Catalyst won instead. Parekh confirmed the loss publicly, and the way he describes it is instructive: he treats the outcome as evidence about market structure, not a verdict on Insight's approach. In a competitive process for a sought-after asset, someone loses. What matters is whether the losing bid reflects a flawed strategy or simply a price another firm was willing to pay.
Read next Top Technology Trends in 2026 You Need to KnowThe Legora episode illustrates a dynamic familiar from earlier technology cycles. When capital floods a category, competitive processes stop being auctions of value and become auctions of conviction, with the most aggressive bidder setting the clearing price. General Catalyst's win suggests it was willing to underwrite Legora at terms Insight would not match — and there is no shortage of historical precedent for both outcomes. In some cycles the aggressive bidder looks prescient; in others, the losing bidder's discipline looks like prescience in retrospect. Parekh's framing treats the loss as a normal cost of operating a diversified strategy: if you are disciplined on price across many deals, you will lose some you wanted, and you will avoid some you should have avoided.
There is also a signaling dimension. Losing Legora did not push Insight toward the frontier labs, as a momentum-driven firm might have been tempted to do. That consistency — absorbing a visible defeat without changing the thesis — is itself information about how the firm allocates.
Holding Stakes in Rival Labs: A Calculated Hedge
Most growth investors describe their AI exposure as a point of view: a bet that one architecture, one team, or one distribution advantage prevails. Parekh describes Insight's exposure as a spread. Holding positions in labs that compete with one another sounds internally inconsistent until you apply basic portfolio reasoning. If you cannot forecast which lab wins, owning several is not indecision — it is the mathematically defensible position under uncertainty.
The hedge has three properties worth naming. First, it reduces single-name risk. A fund concentrated in one frontier lab is exposed to that lab's financing needs, governance disputes, regulatory exposure, and compute economics. Second, it captures category growth even if the winner is unclear. Frontier model spending has been growing fast enough that broad exposure to the layer can outperform concentrated exposure to the wrong name. Third, and less obviously, it preserves optionality. Companies that compete today often become partners, acquirers, or suppliers tomorrow; a portfolio that spans the landscape can participate in consolidation rather than being stranded by it.
The trade-off is real. Diversification caps upside. If Anthropic or OpenAI delivers the kind of return that the earliest backers of Google or Facebook captured, spreading capital across five or ten labs means capturing only a fraction of it. Parekh appears to accept that trade because the probability-weighted math favors it: a slightly smaller share of a broad outcome beats a total loss on a concentrated one.
The Case for Diversification When Everyone Else Is Concentrating
Modern portfolio theory's core insight, formalized by Harry Markowitz in the 1950s, is that adding imperfectly correlated assets improves risk-adjusted returns even when the individual assets are no better than the alternative. Venture is not public equities, and its returns are far more skewed, but the principle survives translation: when your highest-conviction assets are also the most contested and most richly priced, the marginal risk-adjusted dollar may be better deployed elsewhere.
The practical case is even simpler. Frontier labs consume capital at a rate that would have been unimaginable a decade ago, and that capital comes with strings — structured terms, governance provisions, and in some cases compute commitments. A growth firm that writes one enormous check into a single lab takes on concentrated financial and reputational exposure. A firm that writes many smaller checks across labs, infrastructure, tooling, and applications retains flexibility.
None of this argues that OpenAI and Anthropic are bad investments. They may be excellent ones. The argument Parekh makes is about the shape of a portfolio, not the quality of any single company — and it is a reminder that in venture, the consensus trade is often priced for perfection long before it is proved right.
What This Means for the Broader AI Startup Ecosystem
If more growth firms copy Insight's posture, the consequences ripple well beyond one firm's returns. Founders at non-frontier AI companies — the application layer, infrastructure, and vertical AI businesses like Legora — would find a deeper pool of growth capital than the current herd behavior provides, since capital concentrated at the model layer leaves less for everyone else. Competitive processes might moderate, with fewer firms willing to pay any price to win a marquee asset.
There is a risk on the other side. If diversification becomes the new consensus, discipline becomes an excuse for missing the defining company of the era, and limited partners will ask hard questions about why a fund owned ten labs instead of the one that mattered. That tension — between the math of diversification and the power law of venture — is the central unresolved debate in AI investing, and Insight Partners has chosen its side. Whether that reads as prudence or as a missed fortune will not be clear for years, but it will be clear.
Source: TechCrunch
