Insight Partners Spreads Its AI Bets While Rivals Concentrate
Insight Partners' Devin Parekh says the $90 billion firm is holding stakes across rival AI labs rather than concentrating in OpenAI and Anthropic — a contrarian stance in a market built on winner-take-all bets.
Devin Parekh of Insight Partners, which manages $90 billion, said the firm is deliberately keeping its artificial intelligence exposure diversified rather than concentrating capital in OpenAI and Anthropic, and is comfortable holding stakes in competing AI labs even after losing the Legora deal to General Catalyst.
Venture capital in 2026 has developed a gravitational centre. Two names — OpenAI and Anthropic — absorb an outsized share of the private capital flowing into artificial intelligence, and a growing number of large funds have organised their strategies around getting as much exposure to those two as their allocations allow. Insight Partners, which manages $90 billion, is publicly declining to play that game.
Devin Parekh, a managing director at the firm, set out the reasoning in an interview with TechCrunch: Insight is staying deliberately diversified across the AI stack, is relaxed about owning positions in competing model labs, and recently lost the deal for legal AI company Legora to General Catalyst. That last point matters more than a normal competitive loss, because it illustrates exactly what the diversified approach costs — and what it is meant to buy.
Why concentration became the default
The logic behind piling into the two best-funded labs is straightforward and, on its own terms, defensible. Frontier model development is capital-hungry in a way that software historically was not. Training runs, data centre leases and specialised silicon create a cost floor that few private companies can clear. If only a handful of organisations can afford to sit at the frontier, then owning a piece of them looks less like a venture bet and more like buying infrastructure.
The complication is arithmetic. A venture portfolio built around two enormous, late-stage, richly valued private positions does not behave like a venture portfolio. It behaves like a levered index of two assets. Entry prices in the private AI market have risen far enough that the return required to make a fund whole depends on outcomes that look more like public-market blowouts than traditional early-stage multiples. And because these positions are illiquid, there is no easy way to trim.
Parekh's stated position — that Insight is comfortable holding stakes in rival AI labs rather than picking one — is a rejection of that framing. If you cannot know which lab wins, you do not have to. You can own several, and you can also own the layers above and below them.
What losing Legora actually signals
Insight was outbid, or out-manoeuvred, on Legora by General Catalyst. Application-layer AI companies in law, healthcare and financial services have become some of the most fiercely contested assets in private markets, precisely because they sit where model capability converts into recurring revenue from customers who are already used to paying software prices.
Losing a deal like that is the predictable consequence of price discipline. A firm that will not stretch on valuation will lose competitive rounds; that is what discipline means in practice. The more interesting question is whether the discipline holds across a full cycle, because the cost of losing a deal is visible immediately while the benefit of not overpaying shows up years later, quietly, in a fund's net multiple.
Insight's own history is relevant here. The firm built its reputation on growth-stage software — companies with measurable revenue, retention and unit economics — rather than on pre-product research bets. A diversified AI book is closer to that heritage than a two-name concentration would be.
The conflict question nobody has settled
Holding positions in competing labs raises an obvious governance issue. Information flows, board seats, customer introductions and follow-on rights all become more complicated when a firm sits on both sides of a rivalry. In traditional venture, backing direct competitors was close to taboo. In AI, where a single firm's exposure may run from chips to models to applications built on those models, the taboo has eroded, largely because the alternative is sitting out the category entirely.
Parekh's comfort with rival stakes suggests Insight treats the model layer more like a commodity input than a proprietary franchise — something you want exposure to broadly, not something you pick a champion in. That view has a testable implication: if foundation models commoditise, value accrues to the applications and the distribution, and the firms that spread their bets across the stack come out ahead of the ones that paid up for the labs.
The backdrop: public markets are still paying for AI
Information flows, board seats, customer introductions and follow-on rights all become more complicated when a firm sits on both sides of a rivalry.
Private strategy is being set against a public tape that has not yet punished AI enthusiasm. At the most recent close on Friday, 11 September 2026, the Nasdaq 100 tracking fund (NASDAQ: QQQ) finished at $714.88, up 0.87% on the day from a prior close of $708.69, with a session range of $713.63 to $717.63. The S&P 500 proxy (NYSEARCA: SPY) closed at $764.29, up 0.85%, and the Dow tracker (NYSEARCA: DIA) closed at $525.79, up 0.97%.
That matters to private investors for one reason: exit math. Late-stage AI valuations are ultimately underwritten against what public markets will pay. As long as large-cap technology holds its bid, the concentrated bets look reasonable. If the public multiple compresses, the marks on enormous private positions become the problem, and a portfolio spread across many smaller, revenue-generating companies is the easier book to carry.
What to watch from here
- Whether Insight's discipline survives more losses. Legora was one deal. A pattern of losing contested application-layer rounds to firms willing to pay more would test the stated strategy.
- Follow-on behaviour at the labs. Firms holding stakes in multiple competing model companies will eventually face rounds where they must choose whether to defend pro-rata in all of them.
- Application-layer revenue durability. The diversified thesis rests on value accruing above the model layer. Evidence of net revenue retention in AI-native software is the scoreboard.
- Public-market appetite. Any sustained repricing of large-cap technology flows back into private marks with a lag.
Insight is not arguing that OpenAI and Anthropic will fail. It is arguing something narrower and harder to dispute: that a $90 billion pool of capital has no business resolving that question with concentration. Whether that reads as prudence or as a missed generational trade will not be settled by any single round — including the one it lost.
Frequently asked questions
What is Insight Partners' stated AI strategy?
According to managing director Devin Parekh, Insight Partners is deliberately keeping its artificial intelligence exposure diversified rather than concentrating capital in the two best-funded model labs, OpenAI and Anthropic. Parekh also said he is comfortable with the firm holding stakes in competing AI labs at the same time, rather than backing a single champion at the model layer.
How much money does Insight Partners manage?
Insight Partners manages $90 billion in assets. That scale is central to the argument Parekh makes: a capital pool of that size faces different constraints than a small early-stage fund, and concentrating a meaningful share of it in two large, illiquid, highly valued private positions would make the portfolio behave less like venture capital and more like a levered two-asset bet.
What happened with the Legora deal?
Insight Partners lost the investment in Legora, a legal artificial intelligence company, to rival firm General Catalyst. Application-layer AI companies serving law, healthcare and financial services have become among the most competitive assets in private markets, because they convert model capability into recurring software revenue from customers already accustomed to paying for enterprise tools.
Why do some venture firms concentrate on OpenAI and Anthropic?
Frontier model development requires enormous capital for training compute, data centre capacity and specialised chips, which limits how many organisations can operate at the leading edge. Investors who believe that barrier is permanent treat stakes in the largest labs as closer to infrastructure ownership than to conventional venture bets, and size their positions accordingly.
What is the risk of backing competing AI labs at once?
Holding stakes in direct rivals complicates governance: information flow, board representation, customer introductions and follow-on funding rights all become more sensitive. Traditional venture practice discouraged backing direct competitors. In AI, that norm has weakened because a single firm's exposure often spans chips, models and applications, making complete avoidance of overlap impractical.
How did major US market benchmarks close most recently?
At the last trade on Friday, 11 September 2026, the S&P 500 tracking fund SPY closed at $764.29, up 0.85% from a prior close of $757.83. The Nasdaq 100 fund QQQ closed at $714.88, up 0.87%, and the Dow tracker DIA closed at $525.79, up 0.97%. Markets were closed at the time of writing.
Sources
Photo: cottonbro studio · Pexels Licence — source
