Perception Companions’ Devin Parekh on why the agency is diversifying whereas everybody else bets the farm on OpenAI and Anthropic

Devin Parekh has co-run the heavyweight funding agency Perception Companions for 26 years. Not like many VCs who’re loud on X and appear to stay on podcasts, Parekh and Perception Companions have a tendency to put low.

On this sit-down with TechCrunch at its StrictlyVC event on Thursday night time in New York, Parekh was refreshingly candid about among the agency’s wins (it has led and co-led quite a few rounds in Databricks, for instance, and owns stakes in OpenAI and Anthropic); the offers it hasn’t received, together with buzzy AI legal-tech firm Legora; conflicts of curiosity in enterprise investing; and why Perception has caught to a diversified technique at the same time as VCs have piled into the frontier AI labs.

This interview has been edited for size and readability.

There’s a researcher who’s become the big story of the week — do you suppose that issues about AI danger quantity to hysteria, or do you’ve gotten actual issues?

Certain, there’s a danger some non-state actor will get entry to an open-source mannequin and creates a organic weapon. However there’s an excellent greater chance we get a large lower within the time it takes to develop new medicine and treatment illnesses. I’ll take that guess.

I’m on the board of NYU Langone — what AI is already doing with affected person information is superb. We will have a look at 50 million affected person data and inform somebody strolling in for one thing unrelated that they’ve a 25% likelihood of a coronary heart assault. Web-net, I feel that is extremely optimistic.

There are dangers, positive, similar to there are dangers with next-generation drone warfare. Each era has new dangers, and in some way, over time, the world nonetheless raises dwelling requirements. We’re going to wish AI to scale healthcare — the inhabitants is growing older and there aren’t sufficient medical professionals to go round.

Perception has $90 billion in property underneath administration however appears comparatively quiet in comparison with corporations of comparable dimension. Is that purposeful?

Each enterprise capitalist thinks they’re an professional on all the things now — epidemiology throughout COVID, geopolitics in the course of the Iran battle. I’m unsure we’re all specialists on all the things. Our angle has been: Let the portfolio do the speaking. We’re investing in founders and corporations. We have now to speak sufficient that folks know who we’re, however our efficiency ought to converse for itself — and that’s pushed by the portfolio, not by us being loud.

You do early-stage, progress, buyouts, and presumably secondaries. What’s the break up?

It’s temporal, not fastened — we make investments globally, so there’s no set geographic or technique allocation. Take a look at our final seven funds and also you’d see completely different percentages of early-stage, progress, and buyout in every. Buyouts aren’t nice proper now — charges are excessive, debt markets aren’t receptive to software program, exit multiples have come down. We haven’t performed a serious buyout since 2024.

On the enterprise facet, valuations are rising at a tempo we noticed earlier than, in 2021 — and that didn’t finish effectively. Usually, a follow-on spherical means extra information, so that you pay the next worth for decrease danger. Proper now, rounds transfer so quick there’s nearly no incremental information, so that you’re paying extra with out decreasing danger. The logical response is to go earlier. With a scale fund, you can also make smaller bets — write a $20–25 million examine as an alternative of $500 million — and double down on the winners. That’s the place our returns have disproportionately come from. With Wiz, we wrote a Sequence A and stored writing checks, so our achieve was a lot bigger than if we’d stopped on the first examine. And if Wiz hadn’t labored out, it could have barely dented a fund our dimension.

As a world investor, what proportion of your offers are regional versus concentrated someplace just like the Bay Space?

Expertise has gone flat globally. We competed for Legora — my associate Jeff Horing flew to [Stockholm] to pitch the corporate, as a result of that’s the place the founder was. We misplaced that one to Common Catalyst.

That mentioned, AI infrastructure expertise is genuinely concentrated in San Francisco — my 23-year-old son, additionally a VC, is shifting there as a result of he says you’ll be able to’t put money into AI with out being there. However expertise density varies by vertical: Ramp is monetary companies, and that expertise is concentrated in New York. So vertical AI investing might be extra geographically numerous than pure AI infrastructure.

Why did you lose Legora to Common Catalyst?

I don’t know the particular cause, however I feel they offered their worth proposition higher than we offered ours that point. There are many examples the place it went the opposite manner. It’s a giant world; we don’t must win each deal.

You’re invested in rival corporations — OpenAI and Anthropic. That was as soon as taboo in VC. Did that trigger any anguish contained in the agency? Did you are worried about what founders would take away from this?

The inner debate was extra about whether or not we must always have gotten into earlier rounds. It’s very stage-dependent. Khosla did OpenAI’s Sequence A, and there’s no manner they might have then invested in Anthropic, and if we’d performed Anthropic’s Sequence A, we possible couldn’t have performed OpenAI both. When you’re at a later stage, off the board, not driving governance, you’re simply shopping for a terrific inventory.

We noticed OpenAI because the dominant shopper play and Anthropic as having a transparent enterprise technique; that’s shifting in actual time. As these corporations wanted to boost $30–$100 billion, they stopped having the ability to dictate exclusivity. That mentioned, on the Sequence A/B stage, we do have information-sharing restrictions and we don’t put money into instantly competing corporations, although some founders are delicate even to 2% income overlap.

Are you getting extra aggressive on bodily AI?

Bodily intelligence corporations are nonetheless largely science initiatives. It’s not that they received’t develop into actual companies, however you’re having a bet on when robotics adoption occurs, layered on high of a guess on whether or not it occurs in any respect. We’re watching, however we’re not there but. My son thinks it’s the most well liked house round and that I’m loopy to disregard it, which is strictly what I’d anticipate from a 23-year-old.

OpenAI and Anthropic raised roughly half of all VC {dollars} within the first half of this 12 months. Do you suppose LPs fear about focus danger?

We’re not overly concentrated, so it’s not a problem for us. However I’m an LP in different funds, and I do know two funds proper now — elevating their total fund in a month — whose pitch is actually “35–40% of this fund goes into a type of two corporations.” I’m not saying OpenAI and Anthropic received’t do effectively. However this enterprise has at all times rewarded diversification over an extended horizon. We’re on fund 13, so we now have to suppose when it comes to ten funds, not one.

On this specific second, if 25% of our fund have been in Anthropic, our returns would look higher. However information over time doesn’t help extreme focus, and most LPs don’t need that publicity both — although corporations like Founders Fund and Thrive have performed very effectively working concentrated methods. There are at all times going to be exceptions who execute that effectively.

Secondaries are engaging proper now, given how a lot capital was raised in 2021–2023. How are you desirous about these?

The larger challenge is a number of funds raised some huge cash and haven’t returned any of it to LPs. Many first- and second-time funds received’t elevate a subsequent fund as a result of they didn’t prioritize liquidity. I inform fund managers I counsel: if Anthropic’s going to triple from right here, superb — take your foundation out anyway. LPs need to know you’ll be able to flip positions into money; that’s the job.

We have been responsible of this early on, too. As one of many greatest LPs in most of our personal funds, we’d suppose, “Why promote if it may double once more?” However LPs don’t receives a commission that manner. Over the past two years we’ve returned greater than $20 billion to LPs by means of strategic gross sales and IPOs, with a couple of billion extra coming. DPI issues, even on fund 13. Secondaries are actually a liquidity mechanism, typically for early enterprise buyers greater than staff. No person complains a few 10x that stays a 10x, but when it drops to 5x, individuals ask why you didn’t promote.

VC Elad Gill has argued there’s a narrow window — possibly 6 to 12 months — the place an organization’s valuation won’t ever be greater, and founders ought to promote into it. Do you’ve gotten that dialog along with your founders?

We’re at all times having that dialog, although founders hearken to me about as a lot as my children do. It’s case by case, however when a founder will get a proposal at a frothy valuation, I ask them what occurs when the market corrects, as a result of it is going to, even when I can’t inform you when. If I may time it, I’d be on an island managing my portfolio, not speaking to you. You don’t need to promote all the things; de-risk 10 or 20%.

Proper now valuations are rising so quick individuals assume the development continues, however you’ll be able to’t compound $40 billion at 50% each two months for 2 years with out changing into the world financial system. That math doesn’t work.

Anthropic will possible file to go public quickly, with OpenAI presumably behind it. What does that IPO imply for the trade?

Anthropic is already bigger than Salesforce and it’s 4 years outdated — the truth that they will go public doesn’t essentially imply a lot for everybody else. You’ll have three corporations — SpaceX, Anthropic, OpenAI — going public inside six to eight months, every north of a trillion {dollars} in market cap, and the market absorbed SpaceX simply superb. The true query is when the following tier of corporations goes public, and what bar that units. When you’re a public-market investor watching one thing go from zero to $65 billion in 4 years, “double, double, triple, triple” now not seems to be that thrilling by comparability. However that 10x progress price can’t proceed eternally. Ultimately even these corporations develop into normal-growth corporations, and also you want public markets for that. I feel we’ll see extra of those IPOs over the following 18 months.

With a lot capital locked up, will all this LP cash lastly flowing again maintain the frenzy?

All of us do that in our private lives — keep out of an costly market till we will’t stand it anymore, and pile in proper once we must be pulling again. LPs do the identical factor at a macro stage; everybody needed in earlier than 2021, pulled again after, and now the identical LPs are piling again in. That boom-bust cycle is tough to keep away from. Enterprise-growth funds of $6 to $10 billion was once uncommon; now they’re frequent.

How lengthy do you give an organization with a nasty cap construction earlier than deciding whether or not to double down or stroll away?

It varies enormously. Fantastic [an enterprise AI agent platform] was created lower than two years in the past; we did two rounds and it’s now at a $5 billion valuation — a really quick double-down. However, some 2021 investments went nowhere for 3 or 4 years earlier than discovering product-market match. That’s a part of why we do portfolio evaluations — we lately went by means of 300 portfolio corporations over three days, checking not simply on the large positions however in search of those displaying an inflection level value doubling down on, shopping for secondary in, or in some circumstances pulling again from.

Our greatest instance is Armis, a safety firm. We misplaced the preliminary deal to Sequoia, however my associate stored the connection alive with a $5 million take a look at of an $11 billion fund. Eighteen months later, we purchased out your complete cap desk, together with Sequoia, for a nine-figure examine, and offered it to ServiceNow this 12 months for $7 billion. Typically you become profitable with small checks, generally with large ones. The purpose is discovering one of the best founders in one of the best markets.