USV’s New $900M Fund — Fred Wilson, Rebecca Kaden & Mike Mignano (Sourcery)

Sourcery with Molly O’Shea · YouTube · published 2026-10-08 · ~85 minutes

An office tour of USV’s New York space with Mike Mignano, then a sit-down with Fred Wilson, Rebecca Kaden and Mignano on the close of USV’s newest fund cycle. Transcript supplied by Kyle. Auto-caption name errors are corrected where the referent is unambiguous (“a bridge” → Abridge, “Docttronic” → Doctronic, “Sunno” → Suno, “Bin Putnham” → Brynn Putnam, “Jared Hec” → Jared Hecht); uncertain ones are marked [?].

One-line: USV raised $900M — by far its largest fund cycle — because rounds got bigger, faster and more compute-heavy, but the partners argue the strategy is still bottoms-up and thesis-driven, and that single-digit-billion outcomes are enough to return a fund of this size.

Summary

The announcement. USV closed $900M across its new set of funds, the most it has ever raised in one cycle. The partners frame it as continuity — thesis-driven, early-stage, long-hold — paired with a new strategy, because “the outcomes are bigger, the swings are bigger” and rounds are “bigger and more expensive and faster” (Kaden). Why not $1B? Mignano: the number came bottoms-up from the model (company count, check sizes, stages, ownership targets), not as a vanity milestone.

Fund construction. Roughly 30–40 companies, a mix of seed and Series A, with more seeds than USV has historically done. Series A checks of $10–30M, wanting to lead or co-lead and “make real bets” (Kaden), noting a typical Series A now “starts with 10.” Up to a third to a half of seed deals may be “proprietary seeds” — a continuum from in-house incubation (Supertake, an agent that invests in a user’s idea through Robinhood or Coinbase and rebalances it) to partner-sourced ideas handed to a founder with USV writing the first two checks, alongside venture partners Scott Belsky and Jared Hecht, plus more pure founder bets than a thesis-driven firm has usually made.

Why more capital when building is cheaper. Three answers. (1) Building is cheap, so there are far more startups and breaking through costs more in marketing, positioning and go-to-market. (2) Compute is now a cost line that “wasn’t really there before,” and startups are subsidizing inference to reach critical mass — Mignano compares it to the mobile/post-internet era of subsidized consumer usage, and Kaden notes a startup can’t charge for what ChatGPT and Gemini give away. (3) USV’s portfolio is more varied than ever — energy and full-stack physical AI/robotics carry different cost structures and sometimes different kinds of capital. Wilson’s historical anchor: Twitter raised $5M, then $25M, then $100M in successive years; “multiply by 10” and you get today’s round sizes.

The 2003 analogy. Wilson compares now to 2003: internet infrastructure built (Google, Amazon, Yahoo, eBay) but the big applications (Facebook, Twitter, YouTube, iPhone) not yet. AI infrastructure is likewise largely built, and the application layer is starting to explode.

Four theses. Physical AI, AI applications, AI infrastructure, and energy — and USV is “loud” only about these, so founders in them find their way to the firm. Inside them:

• Obliterate, don’t automate — the application-layer thesis, taken from Wilson’s 2014 blog post of the same name: AI lets end users access intelligence previously gatekept by institutions, restructuring whole markets rather than automating incumbents. Wilson: “Don’t hire a lawyer, use an AI lawyer.” USV says 20–30% of its own legal work no longer touches a human lawyer; it trained its own model on past term sheets. Example: Doctronic, an AI doctor that can write prescriptions in Utah.
• The Rebel Alliance — the open-source / open-weight / open-harness stack for developers who won’t build on a closed lab that may later compete with them (the Facebook-platform precedent). Coinbase’s CTO said on stage at the USV CEO summit that Coinbase moved off Claude to its own coding-agent harness built on the open-source Pi, from portfolio company Arendelle [?]; Shopify did similarly.
• Energy — 20–30% of the portfolio, framed as “a very levered bet on AI.” Radiant Nuclear (backed 2021) is the example of a bet made before energy-for-AI was obvious.
• Trust and security — consumer security will change once people hand agents their email (password resets become a master key). Portfolio company Common Fabric builds a sandbox for this; Kaden expects on-device/private models and biometrics may finally get the user demand earlier privacy products lacked.

Agents and network effects. Kaden contrasts two strategies for consumer agents: race for scale and distribution, or — more like the Twitter era — find a network effect by turning single-player behaviors multiplayer. Wilson expects each service to offer its own agent (Amazon, Shopify, Etsy) but users to want a personal agent that negotiates with all of them, with a standard eventually replacing today’s agents-pretending-to-be-the-user (he notes Amazon blocking Muse [?]). Mignano’s hot take on the tour: the UI isn’t going away — agents will shop and file taxes for us, but won’t watch Netflix for us.

Diligence. Mignano reprises his “ARR is a fake metric” clip: annual, pay-to-try subscriptions let a company book revenue from people who don’t like the product, so USV looks at true usage retention. Kaden: “trial behavior is at an absolute high and engagement behavior is at an absolute low.” Team matters more than ever because hiring is brutally competitive and market narratives aggregate faster, crowning winners well before the business justifies it. For consumer products, the “magic” is when the instinctive feel from using the product matches the cohort data.

Never a straight line. USV has run 10+ early-stage funds, each producing two or three great companies — Twitter, Etsy and Indeed in the first; MongoDB and Twilio in the second; then Stripe, Coinbase, Duolingo, Abridge. Abridge started as a consumer product for patients recording appointments; the real customer was the doctor. Shiv Rao made all its money on one product for five or six years, then launched two more lines in the last 12 months — Wilson’s favourite example of a great company becoming a potential 100x. Duolingo’s first product was document translation. Founders fail by overthinking, micromanaging, struggling to recruit, or not burning boats early enough.

Return math. Wilson: USV doesn’t need to underwrite to a trillion; with its fund size and ownership, winners in the single-digit billions produce “really really healthy returns,” and most of its biggest winners have been around $10B. Kaden adds that value now aggregates to the top one or two players faster than before, so USV is less willing to back a #3 or #4. Every investment requires believing the TAM will expand (crypto at the Coinbase investment, nuclear today).

LPs and small funds. Kaden says many LPs asked “are you sure you shouldn’t raise more?” Benefits of staying relatively small: easier to return the fund and reach carry, pickier selection, and — Mignano — aligned incentives, where very large funds “potentially distort the incentives.”

Incumbents. Quoting Brad Burnham: the best time to bet against an incumbent is when doing what you do threatens its core business. Law firms can’t give away what their people bill for; Google is the exception that can cannibalize search with Gemini because it has YouTube, Waymo, Gmail and Android. Kaden’s harder version of the question: not “why can’t incumbents do it” but “why can’t anyone” — when building is cheap, advantage comes back to network effects, speed, team and product decisions.

Highlights

Verbatim from the transcript, in order.

• (01:40) “We we don’t we don’t raise uh funds as vanity metrics, you know, it’s not like we got to hit a billion or we got to hit two billion.” — Mike Mignano
• (14:38) “The market has evolved. Uh we think it’s um the outcomes are bigger, the swings are bigger.” — Rebecca Kaden
• (15:33) “The infrastructure was there. We had Google, we had Amazon, we had Yahoo, we had eBay, but we didn’t yet have Facebook, we didn’t yet have Twitter, we didn’t yet have YouTube, we didn’t yet have iPhone.” — Fred Wilson
• (18:14) “We don’t know when models stall. We don’t know how low compute goes.” — Rebecca Kaden
• (21:20) “Twitter, the original financing was 5 million and then a year later it was 25 million and then a year later it was hundred million.” — Fred Wilson
• (22:38) “We ended up, you know, winning that argument and we were right about that and that’s really how that model emerged that companies used, you know, even to this day to get to network effect.” — Fred Wilson, on arguing against early monetization at Facebook and Twitter
• (23:49) “It’s not that we don’t need network effects anymore. It’s just that we need to get there and we haven’t seen a new way.” — Rebecca Kaden
• (29:34) “You cannot like things about the market, but you’re unlikely to change the things about the market.” — Rebecca Kaden
• (30:23) “Founders who are working in those areas have lots of places to go to but they don’t come to us because we’re not loud about those things. But the things that we are loud about founders find their way to us.” — Fred Wilson
• (32:11) “I was a little surprised how many of them said, ‘Are you sure you shouldn’t raise more?’” — Rebecca Kaden
• (33:13) “When the funds are really big, it it it potentially distorts the incentives a bit.” — Mike Mignano
• (39:44) “The market is designating winners, right? And um telling these stories in a way where value aggregates often in a way that way way way precedes where the business is.” — Rebecca Kaden
• (42:28) “Trial behavior is at an absolute high … and engagement behavior is at an absolute low.” — Rebecca Kaden
• (45:30) “It’s usually two or three, maybe four companies per fund end up being the stars of that fund.” — Fred Wilson
• (49:32) “What you learn you’re backing is not that. You’re backing kind of a a team, a product sense, and a direction.” — Rebecca Kaden, on Abridge
• (53:08) “You can bet on spaces that are more on the come or less obvious because you have a prepared mind and perspective that you’re bringing to it of something you believe.” — Rebecca Kaden
• (1:01:21) “If you really want just a simple way to bet on AI and not have to think about all this other stuff, just bet on energy.” — Fred Wilson
• (1:04:15) “Don’t hire a lawyer, use an AI lawyer. Don’t hire an accountant, use an AI accountant. Don’t hire a doctor, use an AI doctor.” — Fred Wilson
• (1:05:18) “Your AI lawyer is as bad as it’s ever going to be.”
• (1:06:46) “The best time to bet against an incumbent is when them doing what you want to do fundamentally threatens their current business.” — Rebecca Kaden, paraphrasing Brad Burnham
• (1:07:47) “It’s not only why can’t incumbents do it. It’s why can’t anyone do it.” — Rebecca Kaden
• (1:18:23) “I actually think value is aggregating to top players faster than they ever have before.” — Rebecca Kaden
• (1:19:30) “I don’t think a trillion is what we need to underwrite to at USV. I think if if our biggest winners are in the singledigit billions given our fund size and our ownership, we can produce really really healthy returns for our partners.” — Fred Wilson

Connections

• Fund size as strategy. This is the live test of the “Puritans” thesis in The Puritans of Venture Capital and The Siren Song of Raising a Venture Fund, which cite USV as the canonical small-fund firm, and of Venture Capital’s Fourth Turning’s “your fund size is your strategy” framing (see Fund Size). The Bifurcation of Capital is Inevitable already says the cottage keepers are “getting rarer, not stronger.” USV’s answer is that $900M is a bigger number with the same bottoms-up logic and single-digit-billion return math, not a move toward the agglomerators.
• Thesis-Driven Investing — the partners’ defence of theses as a “prepared mind,” with the cost stated plainly: you can be wrong on the person and on the whole thesis. The firm’s research-agent experiment USV’s The Librarian is the operating side of the same idea.
• The Rebel Alliance — Nick Grossman’s post naming the open-source AI thesis; here it is promoted to a full fund theme, with Coinbase and Shopify as adoption evidence.
• Network Effects and AI Agents — Kaden’s single-player-to-multiplayer framing and Wilson’s personal-agent-negotiates-with-service-agents model.
• The Age of Incumbents — the counter-argument. Burnham’s rule (attack where the incumbent can’t follow) against Kyle’s case that incumbents have seized the AI moment; Google cannibalizing search is the example that cuts both ways.
• Power Law — two to four stars per fund, and Kaden’s claim that value now goes to the #1–2 player faster than ever.
• Portfolio mentioned: Abridge, Radiant Nuclear, Coinbase, Duolingo, Twilio, MongoDB, Etsy, Granola, plus Twitter, Indeed, Stripe, Suno (Mikey Shulman), Stedi [?] (Zach Kanter [?]), Doctronic, Common Fabric, Board (Brynn Putnam) and the in-house Supertake.
• Books from the USV library tour: The Messy Middle by Scott Belsky and Shoe Dog.
• Host and show: Molly O’Shea, Sourcery. USV partners: Fred Wilson, Rebecca Kaden, Mike Mignano, Albert Wenger (not present), Brad Burnham (quoted).