Insight Partners’ Deven Parekh on why the firm is diversifying while everyoneelse bets the farm on OpenAI and Anthropic

Deven Parekh has co-run the heavyweight investment firm Insight Partners for 26years. Unlike many VCs who are loud on X and seem to live on podcasts, Parekhand Insight Partners tend to lay low.In this sit-down with TechCrunch at its StrictlyVC event[https://www.flickr.com/photos/techcrunch/] on Thursday night in New York,Parekh was refreshingly candid about some of the firm’s wins (it has led andco-led numerous rounds in Databricks, for example, and owns stakes in OpenAI andAnthropic); the deals it hasn’t won, including buzzy AI legal-tech companyLegora; conflicts of interest in venture investing; and why Insight has stuck toa diversified strategy even as VCs have piled into the frontier AI labs.This interview has been edited for length and clarity.There’s a researcher who’s become the big story of the week[https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/]— I just asked Keith Rabois about this earlier tonight [and he was dismissive ofthe researcher]. Do you think that concerns about AI risk amount to hysteria, ordo you have real concerns?Keith and I probably don’t agree on politics, but we agree on this. Sure,there’s a risk some non-state actor gets access to an open source model andcreates a biological weapon. But there’s an even higher probability we get amassive decrease in the time it takes to develop new drugs and cure diseases.I’ll take that bet.I’m on the board of NYU Langone — what AI is already doing with patient data isamazing. We can look at 50 million patient records and tell someone walking infor something unrelated that they have a 25% chance of a heart attack. Net-net,I think this is highly positive.There are risks, sure, just like there are risks with next-generation dronewarfare. Every generation has new risks, and somehow, over time, the world stillraises living standards. We’re going to need AI to scale healthcare — thepopulation is aging and there aren’t enough medical professionals to go around.Insight has $90 billion in assets under management but seems comparatively quietcompared to firms of similar size. Is that purposeful?Every venture capitalist thinks they’re an expert on everything now —epidemiology during COVID, geopolitics during the Iran war. I’m not sure we’reall experts on everything. Our attitude has been: Let the portfolio do thetalking. We’re investing in founders and companies. We have to communicateenough that people know who we are, but our performance should speak for itself— and that’s driven by the portfolio, not by us being loud.You do early-stage, growth, buyouts, and presumably secondaries. What’s thesplit?It’s temporal, not fixed — we invest globally, so there’s no set geographic orstrategy allocation. Look at our last seven funds and you’d see differentpercentages of early-stage, growth, and buyout in each. Buyouts aren’t greatright now — rates are high, debt markets aren’t receptive to software, exitmultiples have come down. We haven’t done a major buyout since 2024.On the venture side, valuations are rising at a pace we saw before, in 2021 —and that didn’t end well. Normally, a follow-on round means more data, so youpay a higher price for lower risk. Right now, rounds move so fast there’s almostno incremental data, so you’re paying more without reducing risk. The logicalresponse is to go earlier. With a scale fund, you can make smaller bets — writea $20-25 million check instead of $500 million — and double down on the winners.That’s where our returns have disproportionately come from. With Wiz, we wrote aSeries A and kept writing checks, so our gain was much larger than if we’dstopped at the first check. And if Wiz hadn’t worked out, it would have barelydented a fund our size.As a global investor, what percentage of your deals are regional versusconcentrated somewhere like the Bay Area?Talent has gone flat globally. We competed for Legora — my partner Jeff Horingflew to [Stockholm] to pitch the company, because that’s where the founder was.We lost that one to General Catalyst.That said, AI infrastructure talent is genuinely concentrated in San Francisco —my 23-year-old son, also a VC, is moving there because he says you can’t investin AI without being there. But talent density varies by vertical: Ramp isfinancial services, and that talent is concentrated in New York. So vertical AIinvesting can be more geographically diverse than pure AI infrastructure.Why did you lose Legora to General Catalyst?I don’t know the specific reason, but I think they sold their value propositionbetter than we sold ours that time. There are plenty of examples where it wentthe other way. It’s a big world; we don’t need to win every deal.You’re invested in rival companies — OpenAI and Anthropic. That was once tabooin VC. Did that cause any anguish inside the firm? Did you worry about whatfounders would take away from this?The internal debate was more about whether we should have gotten into earlierrounds. It’s very stage-dependent. Khosla did OpenAI’s Series A, and there’s noway they could have then invested in Anthropic, and if we’d done Anthropic’sSeries A, we likely couldn’t have done OpenAI either. Once you’re at a laterstage, off the board, not driving governance, you’re just buying a great stock.We saw OpenAI as the dominant consumer play and Anthropic as having a clearenterprise strategy; that’s shifting in real time. As these companies needed toraise $30-$100 billion, they stopped being able to dictate exclusivity. Thatsaid, at the Series A/B stage, we do have information-sharing restrictions andwe don’t invest in directly competing companies, though some founders aresensitive even to 2% revenue overlap.Are you getting more aggressive on physical AI?Physical intelligence companies are still largely science projects. It’s notthat they won’t become real businesses, but you’re making a bet on when roboticsadoption happens, layered on top of a bet on whether it happens at all. We’rewatching, but we’re not there yet.OpenAI and Anthropic raised roughly half of all VC dollars in the first half ofthis year. Do you think LPs worry about concentration risk?We’re not overly concentrated, so it’s not an issue for us. But I’m an LP inother funds, and I know two funds right now — raising their entire fund in amonth — whose pitch is literally “35-40% of this fund is going into one of thosetwo companies.” I’m not saying OpenAI and Anthropic won’t do well. But thisbusiness has always rewarded diversification over a long horizon. We’re on fund13, so we have to think in terms of 10 funds, not one.In this particular moment, if 25% of our fund were in Anthropic, our returnswould look better. But data over time doesn’t support excessive concentration,and most LPs don’t want that exposure either — though firms like Founders Fundand Thrive have done very well running concentrated strategies. There are alwaysgoing to be exceptions who execute that well.Secondaries are attractive right now, given how much capital was raised in2021-2023. How are you thinking about these?The bigger issue is a lot of funds raised a lot of money and haven’t returnedany of it to LPs. Many first- and second-time funds won’t raise a next fundbecause they didn’t prioritize liquidity. I tell fund managers I advise: ifAnthropic’s going to triple from here, fine — take your basis out anyway. LPswant to know you can turn positions into cash; that’s the job.We were guilty of this early on, too. As one of the biggest LPs in most of ourown funds, we’d think, “Why sell if it could double again?” But LPs don’t getpaid that way. Over the last two years we’ve returned more than $20 billion toLPs through strategic sales and IPOs, with a few billion more coming. DPI[https://carta.com/learn/private-funds/management/fund-performance/dpi/]matters, even on fund 13. Secondaries are really a liquidity mechanism, oftenfor early venture investors more than employees. Nobody complains about a 10xthat stays a 10x, but if it drops to 5x, people ask why you didn’t sell.VC Elad Gil has argued there’s a narrow window[https://techcrunch.com/2026/04/19/the-12-month-window/] — maybe six to 12months — where a company’s valuation will never be higher, and founders shouldsell into it. Do you have that conversation with your founders?We’re always having that conversation, though founders listen to me about asmuch as my kids do. It’s case by case, but when a founder gets an offer at afrothy valuation, I ask them what happens when the market corrects, because itwill, even if I can’t tell you when. If I could time it, I’d be on an islandmanaging my portfolio, not talking to you. You don’t have to sell everything;de-risk 10 or 20%.Right now valuations are rising so fast people assume the trend continues, butyou can’t compound $40 billion at 50% every two months for two years withoutbecoming the world economy. That math doesn’t work.Anthropic will likely file to go public soon, with OpenAI presumably behind it.What does that IPO mean for the industry?Anthropic is already larger than Salesforce and it’s four years old — the factthat they can go public doesn’t necessarily mean much for everyone else. You’llhave three companies — SpaceX, Anthropic, OpenAI — going public within six toeight months, each north of a trillion dollars in market cap, and the marketabsorbed SpaceX just fine. The real question is when the next tier of companiesgoes public, and what bar that sets. If you’re a public-market investor watchingsomething go from zero to $65 billion in four years, “double, double, triple,triple” no longer looks that exciting by comparison. But that 10x growth ratecan’t continue forever. Eventually even these companies become normal-growthcompanies, and you need public markets for that. I think we’ll see more of theseIPOs over the next 18 months.With so much capital locked up, will all this LP money finally flowing backsustain the frenzy?We all do this in our personal lives — stay out of an expensive market until wecan’t stand it anymore, and pile in right when we should be pulling back. LPs dothe same thing at a macro level; everyone wanted in before 2021, pulled backafter, and now the same LPs are piling back in. That boom-bust cycle is hard toavoid. Venture-growth funds of $6 to $10 billion used to be rare; now they’recommon.How long do you give a company with a bad cap structure before deciding whetherto double down or walk away?It varies enormously. Wonderful [an enterprise AI agent platform] was createdless than two years ago; we did two rounds and it’s now at a $5 billionvaluation — a very fast double-down. On the other hand, some 2021 investmentswent nowhere for three or four years before finding product-market fit. That’spart of why we do portfolio reviews — we recently went through 300 portfoliocompanies over three days, checking not just on the big positions but lookingfor the ones showing an inflection point worth doubling down on, buyingsecondary in, or in some cases pulling back from.Our best example is Armis, a security company. We lost the initial deal toSequoia, but my partner kept the relationship alive with a $5 million check outof an $11 billion fund. Eighteen months later, we bought out the entire captable, including Sequoia, for a nine-figure check, and sold it to ServiceNowthis year for $7 billion. Sometimes you make money with small checks, sometimeswith big ones. The goal is finding the best founders in the best markets.Update: An earlier version of this piece did not mention that Keith Rabois wasasked about former Anthropic researcher Jacob Coxon’s comments before Parekhwas. Rabois was asked first; Parekh was then asked a version of the samequestion.