Insight Partners’ Devin Parekh on why the firm is diversifying while everyone else bets the farm on OpenAI and Anthropic

Devin Parekh has co-run the heavyweight concern steadfast Insight Partners for 26 years. Unlike galore VCs who are large connected X and look to unrecorded connected podcasts, Parekh and Insight Partners thin to laic low.

In this sit-down with TechCrunch astatine its StrictlyVC event connected Thursday nighttime successful New York, Parekh was refreshingly candid astir immoderate of the firm’s wins (it has led and co-led galore rounds successful Databricks, for example, and owns stakes successful OpenAI and Anthropic); the deals it hasn’t won, including buzzy AI legal-tech institution Legora; conflicts of involvement successful task investing; and wherefore Insight has stuck to a diversified strategy adjacent arsenic VCs person piled into the frontier AI labs.

This interrogation has been edited for magnitude and clarity.

There’s a researcher who’s become the large communicative of the week — bash you deliberation that concerns astir AI hazard magnitude to hysteria, oregon bash you person existent concerns?

Sure, there’s a hazard immoderate non-state histrion gets entree to an open-source exemplary and creates a biologic weapon. But there’s an adjacent higher probability we get a monolithic alteration successful the clip it takes to make caller drugs and cure diseases. I’ll instrumentality that bet.

I’m connected the committee of NYU Langone — what AI is already doing with diligent information is amazing. We tin look astatine 50 cardinal diligent records and archer idiosyncratic walking successful for thing unrelated that they person a 25% accidental of a bosom attack. Net-net, I deliberation this is highly positive.

There are risks, sure, conscionable similar determination are risks with next-generation drone warfare. Every procreation has caller risks, and somehow, implicit time, the satellite inactive raises surviving standards. We’re going to request AI to standard healthcare — the colonisation is aging and determination aren’t capable aesculapian professionals to spell around.

Insight has $90 cardinal successful assets nether absorption but seems comparatively quiescent compared to firms of akin size. Is that purposeful?

Every task capitalist thinks they’re an adept connected everything present — epidemiology during COVID, geopolitics during the Iran war. I’m not definite we’re each experts connected everything. Our cognition has been: Let the portfolio bash the talking. We’re investing successful founders and companies. We person to pass capable that radical cognize who we are, but our show should talk for itself — and that’s driven by the portfolio, not by america being loud.

You bash early-stage, growth, buyouts, and presumably secondaries. What’s the split?

It’s temporal, not fixed — we put globally, truthful there’s nary acceptable geographic oregon strategy allocation. Look astatine our past 7 funds and you’d spot antithetic percentages of early-stage, growth, and buyout successful each. Buyouts aren’t large close present — rates are high, indebtedness markets aren’t receptive to software, exit multiples person travel down. We haven’t done a large buyout since 2024.

On the task side, valuations are rising astatine a gait we saw before, successful 2021 — and that didn’t extremity well. Normally, a follow-on circular means much data, truthful you wage a higher terms for little risk. Right now, rounds determination truthful accelerated there’s astir nary incremental data, truthful you’re paying much without reducing risk. The logical effect is to spell earlier. With a standard fund, you tin marque smaller bets — constitute a $20–25 cardinal cheque alternatively of $500 cardinal — and treble down connected the winners. That’s wherever our returns person disproportionately travel from. With Wiz, we wrote a Series A and kept penning checks, truthful our summation was overmuch larger than if we’d stopped astatine the archetypal check. And if Wiz hadn’t worked out, it would person hardly dented a money our size.

As a planetary investor, what percent of your deals are determination versus concentrated determination similar the Bay Area?

Talent has gone level globally. We competed for Legora — my spouse Jeff Horing flew to [Stockholm] to transportation the company, due to the fact that that’s wherever the laminitis was. We mislaid that 1 to General Catalyst.

That said, AI infrastructure endowment is genuinely concentrated successful San Francisco — my 23-year-old son, besides a VC, is moving determination due to the fact that helium says you can’t put successful AI without being there. But endowment density varies by vertical: Ramp is fiscal services, and that endowment is concentrated successful New York. So vertical AI investing tin beryllium much geographically divers than axenic AI infrastructure.

Why did you suffer Legora to General Catalyst?

I don’t cognize the circumstantial reason, but I deliberation they sold their worth proposition amended than we sold ours that time. There are plentifulness of examples wherever it went the different way. It’s a large world; we don’t request to triumph each deal.

You’re invested successful rival companies — OpenAI and Anthropic. That was erstwhile taboo successful VC. Did that origin immoderate anguish wrong the firm? Did you interest astir what founders would instrumentality distant from this?

The interior statement was much astir whether we should person gotten into earlier rounds. It’s precise stage-dependent. Khosla did OpenAI’s Series A, and there’s nary mode they could person past invested successful Anthropic, and if we’d done Anthropic’s Series A, we apt couldn’t person done OpenAI either. Once you’re astatine a aboriginal stage, disconnected the board, not driving governance, you’re conscionable buying a large stock.

We saw OpenAI arsenic the ascendant user play and Anthropic arsenic having a wide endeavor strategy; that’s shifting successful existent time. As these companies needed to rise $30–$100 billion, they stopped being capable to dictate exclusivity. That said, astatine the Series A/B stage, we bash person information-sharing restrictions and we don’t put successful straight competing companies, though immoderate founders are delicate adjacent to 2% gross overlap.

Are you getting much assertive connected carnal AI?

Physical quality companies are inactive mostly subject projects. It’s not that they won’t go existent businesses, but you’re making a stake connected erstwhile robotics adoption happens, layered connected apical of a stake connected whether it happens astatine all. We’re watching, but we’re not determination yet. My lad thinks it’s the hottest abstraction astir and that I’m brainsick to disregard it, which is precisely what I’d expect from a 23-year-old.

OpenAI and Anthropic raised astir fractional of each VC dollars successful the archetypal fractional of this year. Do you deliberation LPs interest astir attraction risk?

We’re not overly concentrated, truthful it’s not an contented for us. But I’m an LP successful different funds, and I cognize 2 funds close present — raising their full money successful a period — whose transportation is virtually “35–40% of this money is going into 1 of those 2 companies.” I’m not saying OpenAI and Anthropic won’t bash well. But this concern has ever rewarded diversification implicit a agelong horizon. We’re connected money 13, truthful we person to deliberation successful presumption of 10 funds, not one.

In this peculiar moment, if 25% of our money were successful Anthropic, our returns would look better. But information implicit clip doesn’t enactment excessive concentration, and astir LPs don’t privation that vulnerability either — though firms similar Founders Fund and Thrive person done precise good moving concentrated strategies. There are ever going to beryllium exceptions who execute that well.

Secondaries are charismatic close now, fixed however overmuch superior was raised successful 2021–2023. How are you reasoning astir these?

The bigger contented is simply a batch of funds raised a batch of wealth and haven’t returned immoderate of it to LPs. Many first- and second-time funds won’t rise a adjacent money due to the fact that they didn’t prioritize liquidity. I archer money managers I advise: if Anthropic’s going to triple from here, good — instrumentality your ground retired anyway. LPs privation to cognize you tin crook positions into cash; that’s the job.

We were blameworthy of this aboriginal on, too. As 1 of the biggest LPs successful astir of our ain funds, we’d think, “Why merchantability if it could treble again?” But LPs don’t get paid that way. Over the past 2 years we’ve returned much than $20 cardinal to LPs done strategical income and IPOs, with a fewer cardinal much coming. DPI matters, adjacent connected money 13. Secondaries are truly a liquidity mechanism, often for aboriginal task investors much than employees. Nobody complains astir a 10x that stays a 10x, but if it drops to 5x, radical inquire wherefore you didn’t sell.

VC Elad Gill has argued there’s a narrow window — possibly 6 to 12 months — wherever a company’s valuation volition ne'er beryllium higher, and founders should merchantability into it. Do you person that speech with your founders?

We’re ever having that conversation, though founders perceive to maine astir arsenic overmuch arsenic my kids do. It’s lawsuit by case, but erstwhile a laminitis gets an connection astatine a frothy valuation, I inquire them what happens erstwhile the marketplace corrects, due to the fact that it will, adjacent if I can’t archer you when. If I could clip it, I’d beryllium connected an land managing my portfolio, not talking to you. You don’t person to merchantability everything; de-risk 10 oregon 20%.

Right present valuations are rising truthful accelerated radical presume the inclination continues, but you can’t compound $40 cardinal astatine 50% each 2 months for 2 years without becoming the satellite economy. That mathematics doesn’t work.

Anthropic volition apt record to spell nationalist soon, with OpenAI presumably down it. What does that IPO mean for the industry?

Anthropic is already larger than Salesforce and it’s 4 years aged — the information that they tin spell nationalist doesn’t needfully mean overmuch for everyone else. You’ll person 3 companies — SpaceX, Anthropic, OpenAI — going nationalist wrong six to 8 months, each northbound of a trillion dollars successful marketplace cap, and the marketplace absorbed SpaceX conscionable fine. The existent question is erstwhile the adjacent tier of companies goes public, and what barroom that sets. If you’re a public-market capitalist watching thing spell from zero to $65 cardinal successful 4 years, “double, double, triple, triple” nary longer looks that breathtaking by comparison. But that 10x maturation complaint can’t proceed forever. Eventually adjacent these companies go normal-growth companies, and you request nationalist markets for that. I deliberation we’ll spot much of these IPOs implicit the adjacent 18 months.

With truthful overmuch superior locked up, volition each this LP wealth yet flowing backmost prolong the frenzy?

We each bash this successful our idiosyncratic lives — enactment retired of an costly marketplace until we can’t basal it anymore, and heap successful close erstwhile we should beryllium pulling back. LPs bash the aforesaid happening astatine a macro level; everyone wanted successful earlier 2021, pulled backmost after, and present the aforesaid LPs are piling backmost in. That boom-bust rhythm is hard to avoid. Venture-growth funds of $6 to $10 cardinal utilized to beryllium rare; present they’re common.

How agelong bash you springiness a institution with a atrocious headdress operation earlier deciding whether to treble down oregon locomotion away?

It varies enormously. Wonderful [an endeavor AI cause platform] was created little than 2 years ago; we did 2 rounds and it’s present astatine a $5 cardinal valuation — a precise accelerated double-down. On the different hand, immoderate 2021 investments went obscurity for 3 oregon 4 years earlier uncovering product-market fit. That’s portion of wherefore we bash portfolio reviews — we precocious went done 300 portfolio companies implicit 3 days, checking not conscionable connected the large positions but looking for the ones showing an inflection constituent worthy doubling down on, buying secondary in, oregon successful immoderate cases pulling backmost from.

Our champion illustration is Armis, a information company. We mislaid the archetypal woody to Sequoia, but my spouse kept the narration live with a $5 cardinal cheque retired of an $11 cardinal fund. Eighteen months later, we bought retired the full headdress table, including Sequoia, for a nine-figure check, and sold it to ServiceNow this twelvemonth for $7 billion. Sometimes you marque wealth with tiny checks, sometimes with large ones. The extremity is uncovering the champion founders successful the champion markets.