Summary

Venture learned nothing from 2021. Mitchell Green’s core charge: “I think the venture industry was about to be in for a rude awakening and then AI showed up” — Mitchell’s gloss, “AI was the oxygen that we needed.” The lessons that should have stuck: “entry price matters a lot, don’t overcapitalize companies,” and stock-based comp dilution is massively underestimated (Lead Edge now models 20-30%). Will the tourists wash out? “100%… it might be a slow hole in the canoe, but eventually yes.”
The ByteDance bull case survives a TikTok ban. Lead Edge bought late last year at ~5x earnings growing 25-30%, underwriting the US business at zero (single-digit % of revenue, unprofitable). It’s “China’s first truly global business,” a future “foremost AI company on the planet,” and liquidity comes via Hong Kong like Tencent — versus Facebook at ~$1.5-1.7T on similar earnings with slower growth. “We like to buy stuff when no one else likes to buy things.”
AI infrastructure is 1997 websites; incumbents win the application layer. Prices will plummet the way $50M Sun server websites became $10/month — the DeepSeek-day trade (Nvidia down, software up) was rational. Since the iPhone only three new $100B companies have been built — ByteDance, Pinduoduo, Uber — because “incumbency wins, it’s customer distribution,” and the one-person AI company is “comical at best.”
The model: criteria, not theses. Eight objective criteria, 10,000 cold calls a year, five-plus criteria = ~10% yield → 5-7 deals; less than 10% of the portfolio is Bay Area and 70% of the time Lead Edge is the first institutional money. Specimen: SafeSend, bootstrapped in Ann Arbor, 60% bought in 2021 at ~$130-140M, grown from $13M ARR to $47M revenue, sold to Thomson Reuters — in Silicon Valley “it would have been $500 million.”
Hundreds of 2021-vintage SaaS companies are the living dead — $50-200M revenue, mid-teens growth, effectively already IPO’d on private money. The only exit is Rule of 40 plus 90%+ gross dollar retention, sold to mid-market PE (50-60% of which now buy software). And take the exit: “if you told me today I could take a 7x just to get out of it, I would happily do it.”
DPI over everything: “marks are completely for suckers.” The 3x-net fund some endowments demand is “a complete fallacy”; emerging managers should copy Fabrice Grinda — “invest in the seed or A and sell a bunch in the B or C” — because “you got to stay in business.” The LP diligence question that exposes everyone: how much unlocked public stock did you hold on September 30, 2021, and why didn’t you distribute it?
LPs are more important than founders: “if you do not have LPs, you do not have a business.” Lead Edge made its LP roster (ex-CEOs of Schwab, Kimberly Clark, Colgate) the moat — every intro logged in Salesforce, 97% gross-dollar-retention target on LPs — and Harry agrees it’s “really arrogant” of VCs to deny LPs are customers.
Watchlist extras: AI software gross dollar retention is “just really, really low — actually shocking” (Lovable ~85%, “better than ChatGPT but it’s not 90”); on a 30% drawdown he’d buy Snowflake or Datadog “and put it in a drawer” for 10+ years alongside Microsoft; and MicroStrategy’s debt-funded Bitcoin flywheel “sounds like a house of cards to me.”

Deep dive

1. Criteria beat theses: eight boxes, 10,000 cold calls, zero professors

Green’s formative story is Bessemer, 2005: while the “Shark Tank-esque” partnership waited for pitches, Insight was cold-calling $15M-revenue bootstrapped companies using “22 to 24 year old knuckleheads” — copying Summit and TA. The apprenticeship lesson he kept: “if the company calls you back, the company sucks” — the good ones you call every two days for a month, and you learn quality by talking to 10,000 bad companies over two years. Week by week the partners hardened rules ($10M+ revenue, 50% growth, 70% gross margins) into five criteria; Lead Edge expanded them to the “Lead Edge 8.”
The funnel is mechanical and deliberately objective for a 23-year-old caller: 10,000 companies a year; all eight criteria is a ~1% yield (“too small of a pond”), five-plus criteria yields ~10% — validated over ~70,000 calls in a decade — leading to 150-175 diligences and 5-7 deals.
Harry relays a Spark GP’s jab that this “spreadsheet investing” is outdated in an AI world. Green’s rebuttal: 70% of what Lead Edge buys, Spark has never looked at — less than 10% of the portfolio is Bay Area, 70% of the time they’re the first institutional investor, and “it’s very hard to make money investing at 100 times revenues.”
The anti-thesis rant, verbatim: “every investor has a thesis — and what the f* are we, professors? I’ve never had a thesis… my thesis is just meet six or eight criteria,” plus “people should spend less time tweeting, more time investing.” His one rule for emerging managers: “define what you’re going to do and do exactly that — don’t stray from it at all.” Firm culture matches: fast nos with reasons (“don’t drag it out a month”), and the analyst who wrote “I reject your rejection” got pulled back in and interviewed.

2. AI infrastructure is 1997 websites — incumbents win the application layer

The frame that opens the show: “investing in AI infrastructure today is like investing in websites in 1997.” Then, you’d spend $50M on Sun Microsystems servers; today £10 a month buys something 50x better — “prices are going to plummet.” DeepSeek day was the market being rational: Nvidia fell, software stocks rose — and he finds it “quite funny” that many AI-infra investors hadn’t even heard of DeepSeek.
Who captures the value: incumbents. Gravity’s 10-12 engineers become “30 or 40” via Cursor and Copilot — “you’re going to see this at Salesforce, at Workday.” Since the iPhone (~2006-07), only three companies that didn’t exist before reached $100B — ByteDance, Pinduoduo, and Uber — while Facebook, Google and Microsoft won. “Incumbency wins. It’s customer distribution.”
Hence: “the idea of a single-person AI company I think is comical at best.” Software “is not rocket-science tech people are solving — it’s sales, it’s distribution, GTM, regulations.” And the old fear inverted: “if I had a dollar for every time somebody said to me ‘Microsoft’s just going to do this,’ I would have never invested in any software companies.”
The hedge, kept intact: people “always overestimate [technology] over the near term and underestimate it in the long term.” AI revolutionizes the world over 10-20 years — but not via “a new call-center software company”; it’ll be something nobody could do before, found early by “guys like you or Benchmark or Sequoia.” He also owns the doubt: “we question ourselves now — are we totally wrong on AI?… I could be totally wrong.”

3. Buy the boring: control deals nobody else is calling

Gravity, “the world’s most boring company”: budget-planning software for small local governments — the Palo Alto Police Department has to post a budget online — roughly $10M revenue growing 50-60%, never raised, never had a salesperson. Lead Edge bought it for ~$50M and installed a CEO and CRO. “It will never be an IPO in a million years” — that’s the point, not the problem.
SafeSend is the specimen: a verticalized DocuSign for tax returns, bootstrapped in Ann Arbor, found by cold call. COVID-enabled and — unlike Hopin — it stayed enabled: electronic tax-return signing stuck after COVID. Bought 60% in mid-2021 at ~$130-140M ($90M equity, $20M debt) on ~$13M ARR growing 50-60%; built to ~$47M revenue and sold to Thomson Reuters. Had Benchmark done it as a minority deal in Silicon Valley, “it would have been $500 million.”
Harry’s pushback — these aren’t generation-defining founders, and Benchmark’s Fenton would say so. Green: “100% correct… totally fine.” The game is buying $10-20M-revenue software and exiting at $60-80M — “zero percent chance it’s the next Snowflake.” Exhibit: ExaGrid, a storage business competing with HP and Dell — Lehman Brothers’ stake bought out at a $130M valuation, now $165-170M revenue, 70% gross margin, $26M EBITDA — to be built to $250M revenue and sold at 10x EBITDA for 4x. “Not generational… but it is tech investing making really good returns.”
The same flexibility runs through every structure — his table analogy: a company is a table, and 10% or 60% of it is the same table. Walk past three-criteria houses, and at a six-criteria house go in the front door (primary), the side door (secondary from an early investor), or “the basement window with a pickaxe” — fund a co-invest vehicle, or buy a 20-year-old fund where 90% of NAV is one company.

4. The living dead: Rule of 40 or bust — and mid-market PE is the likely exit

The 2021 cohort effectively completed IPOs in private: where 2015-17 IPOs raised $100-300M, 2021 rounds raised the same — leaving companies with $130M revenue, 18% growth and $130M of cash, against a $3B last-round mark. His prescription, drilled into his own portfolio: “you have to get to Rule of 40, because that’s the only way you’re getting out. A strategic’s not going to just come and buy you; this company’s never going to go public” — and even then “you might be worth five times revenue.”
The route he describes is Rule of 40 plus 90%+ gross dollar retention, sold to mid-market PE. The exit that didn’t exist last cycle: in 2008-10, mid-market PE (Nautic, GTCR, Charlesbank) bought industrials and services, never software. Now 50-60% of them have software sleeves, their portfolios grow GDP-plus so 15% growth “is really fast for them” — and a third of Lead Edge’s exits already go to private equity.
What makes you salable is high gross margins plus 90%+ gross dollar retention (“if you have 70, 75, 80% gross retention — good luck”). And no board-seat romance: “yes we have a pref, so we would get our 1x — but if you told me today I could take a 7x just to get out of it, I would happily do it. There are hundreds of these companies out there.”
On IPOs: the market itself “is actually totally fine” (look at Reddit versus its opening price); the problem is the Grafanas, Databricks and Stripes “have so much cash they don’t need to go public.” Harry’s contrarian claim — in five years most companies that could go public won’t, “being public will be an unfortunate consequence of scale,” which is dangerous for VC and LPs without a mature secondaries market — draws a flat “correct.” Green holds both sides: the quarterly cadence is “a little bit nonsensical” (likely John Collison: if a Goldman analyst is your source of discipline, “you clearly do not have a great company”), yet Benioff would probably call being public a necessary evil, and Stripe-style honesty about staying private is what makes it fine.

5. Sell relentlessly: the disposition committee and the DPI religion

Lead Edge runs a disposition committee — “the exact same thing [as an investment committee], just in reverse”: is there a secondary, an early investor, a crossover fund that would buy our stake; has the market proved too small, have we lost confidence in management. “There are a lot of really good funds that are really good at investing… a lot of people that are not very good at selling.” His corollary for early-stage firms: when the company IPOs, get off the board and sell — “good company and good investment are two very fundamentally different things because of valuation.”
The creed: “never have I regretted selling too early… pigs get slaughtered at the end of the day.” The target is 2-5x in three-to-seven years — Green notes it blends to roughly a 20% net IRR — and from the quickfire: “DPI is the most important thing and marks are completely for suckers.”
On “companies are bought, not sold” — he disagrees from experience, having generated returns by running auctions. But you can’t get bought if strategics don’t know you exist, so build those relationships early — and sandbag: “if you think you’re going to do 50 million of revenue this year up from 30, tell them you’re going to do 40 and then beat the number.” LPs share blame for the industry’s complacency — one of his longtime LPs calls some VCs “pigs at the trough”: “I ate the food, I spent all the money, now give me more money to spend again.”

6. Venture’s math problem: 15-year duration, the 3x fallacy, and staying in business

Harry’s LP arithmetic: at Lead Edge’s duration an allocator can compound ~3x nearly three times over in the 15 years an early-stage fund needs to (hopefully) return 3x once — so how does venture earn its slot? Green doesn’t fight it: “I think it’s very hard. There are too many venture funds” — and it’s “actually shocking” that fund count grew over the last five-to-seven years while exits got longer. The fallacy, as told: an endowment rejected them because “I only invest in 3x net funds,” and the partners left asking “should we go hire that guy to run our money?… please tell me where all these 3x net funds are. It’s a complete fallacy.”
Advice to emerging managers: copy Fabrice Grinda (an LP of his for 15 years) — “invest in the seed or A and sell a bunch in the B or C.” You still ride winners, you return DPI, and you harvest the secondary wave of bloated growth funds and crossovers wanting in.
Harry’s counter, worth keeping: Emergence sold Salesforce at IPO and forwent something like $20-30B; Bessemer sold Shopify at IPO and “lost billions” — if venture is a power-law game, selling the B caps it. Green concedes the power law “but by the way, you got to stay in business”: the fund that reaches 1x DPI fastest raises the next one, and “for every Shopify, go ask him about 1999 and 2000 — or how many billions of dollars were lost in 2020 and 2021 by not distributing positions.”
Who’s actually good: Spectrum Equity — “relentless in thinking about how to get liquidity to LPs,” doing minority sales two years in to get their basis back — and TA Associates, “the best returns in the tech investing world the last 34 years.” Honor roll for staying small: Benchmark across three generations, Kopelman’s First Round, Floodgate. His LP picks: Iconiq or Meritech at growth (“the returns of Iconiq are freaking amazing”), Benchmark or Bessemer at seed/A.

7. Nothing was learned from 2021 — entry price is the whole lesson

The industry indictment: “the venture industry was about to be in for a rude awakening and then AI showed up.” Mitchell’s line — “I always say AI was the oxygen that we needed” — and Green’s verdict: “people didn’t learn a damn thing from ‘20 and ‘21. It’s shocking.” The lessons, when Harry asks for them: “entry price matters a lot; don’t overcapitalize companies”; dilution really matters; and if your fund needs IPOs, back founders who actually want to IPO. Tourists get washed out “100%… it might be a slow hole in the canoe, but eventually yes.”
His own 2021 sins were “overpaying for a couple companies assuming the exit multiple was going to be higher than it actually is.” House math now: software growing 15-30% exits at four to seven times revenue; Lead Edge underwrites 4-8x, “maybe sometimes 10 times at the absolute highest” for 30-40% growers. He credits Iconiq for the reverse trade: underwriting 2015-17 deals to exit at 10-12x, paying ~20% up for the best assets, then watching multiples go to 20-30x — “the best way to 4x your money is 2x the revenue and 2x the multiple. The reverse happens too… 2x your revenues and half your multiple — that’s called a 1x.”
Today’s stupidity: “paying 100 times revenues.” His test: after 12-18 months of growth, “am I kind of in the money, or do I need to grow four or five years until I even get in the money?” Toast circa 2015-17: ~~$25M revenue growing 250% a year, Lead Edge paid 20x (~~$500M) — “pretty expensive,” but a year out they were in at ~10x. Today’s prices: “totally insane.”
Dilution discipline: “most people, including ourselves, over the last 15 years massively underestimated stock-based comp dilution.” Lead Edge now assumes 20-30%, more when investing earlier. Alibaba was presented as a possible exception — a billion dollars of profit at entry and perhaps fewer shares outstanding — while “Uber was totally insane.”

8. Gross dollar retention is the tell — and AI software’s is “shocking”

The case for GDR over NDR: a company that doubled revenue on 200% net retention but 50% gross retention lost half its customers — invisible at $10M, but at $100-300M revenue “it’s a huge hole in the bottom of the bucket” that wrecks sales-and-marketing efficiency as burn rises. Entrepreneurs telling him “I don’t look at gross dollar retention, net’s the only thing that matters” are themselves the red flag.
The AI read: “we look at tons of these AI software companies and the gross dollar retention rates are just really, really low — it’s actually shocking. Pretty much all of them.” Lovable at ~85% is “pretty good, better than ChatGPT — but again, it’s not 90.” The contrast case: a cardiac-monitoring software business with 99% GDR that “will not change the world” but runs capital-efficiently over time.
The other screen — “Warren Buffett would laugh at us because it sounds kind of stupid, but it works”: are revenues today greater than cumulative historical cash burn (a 1:1 ratio or better)? Benchling passed: ~$13M ARR growing north of 100%, priced “in the low threes” — expensive — but it had burned only ~$10M. The famous fail: Snowflake, passed at $500M on “horrible gross margins” — “we were completely wrong. Like 100% wrong.”
From his viral “hierarchy of BS” letters on how CEOs lie: “greatest place to work” on page two of the deck, total contract value charted as revenue, gross profit fudged through COGS. And LPs should diligence the same way — “talk to the companies that failed… how did the partner deal with adversity” — not the easy winners.

9. LPs are more important than founders — the network is the product

The origin logic: two “knuckleheads” who had “never been the global head of HR of anything” asked why any founder would take their money — so they made the LP base the moat. 80-90% of LPs are listed on the website — former CEOs of Charles Schwab, Kimberly Clark, Colgate-Palmolive — and the pitch to a pharma-software founder is: want to meet the former CEO of Pfizer, of Biogen? Intros happen during diligence: “they act as our own version of McKinsey.”
“Everybody says they help; very few people do.” Every intro is BCC’d to an assistant and logged in a Salesforce customized over 15 years; 80-plus current and former portfolio execs are LPs; Duo’s founder DM’d after Harry teased the episode — “now I’m an LP.” An associate flying to Seattle for a wedding is told to stay Monday and meet four LPs, plane ticket paid — “it’s just not their model” at an Index-style fund backed by endowments.
The heresy stated plainly: two customers, “founders, but more importantly LPs — because if you do not have LPs, you do not have a business.” Harry goes further: “I think it’s really arrogant to suggest they are not your customer.” Lead Edge targets 97% gross dollar retention on LPs — quarterly calls, events, real transparency (“the lack of transparency in this industry to LPs is shocking”) — and communication buys forgiveness for a bad vintage, while jerks don’t get re-upped: the take-it-or-leave-it PE fund that spent three and a half years raising “a fraction of the size” of its prior fund.
His killer LP diligence question: ask any manager of 10-plus years, “on September 30th of ‘21, how much unlocked public stock did you have in your portfolio — and why didn’t you distribute it?” “Isn’t your goal just to return money to LPs? Isn’t that the whole job of the business?” On Harry’s point that a16z proved brand beats performance at scale: “we’ll see over the next 20 years… by the way, they’re far richer than I am” — ego hurt? “Nope.” “We stay in our lane.”

10. ByteDance at 5x earnings — and Microsoft in a drawer

The underwrite: Lead Edge was buying ByteDance late last year at ~5x earnings, growing 25-30% a year, with the US business valued at zero — North America is a single-digit percentage of revenue and unprofitable there. When TikTok went dark for a day, “both sides of the aisle came running… please keep it open.” He suspects it won’t be done by April 5th—or whatever the date is—and they’ll probably push it out again, but hedges hard: “there’s four or five people in the world that know what’s going to happen.”
The deeper thesis: ByteDance is “that first truly global business” China has always wanted — Alibaba and Tencent aren’t truly global; Nike, Microsoft and John Deere are — the government “really likes” it, and it will be “one of the foremost AI companies on the planet over the next decade” (in India, banned for years, “nobody’s really able to build a competitor”). Liquidity is Hong Kong, where Tencent lists. The comp: Facebook at roughly $1.5-1.7T on same-size earnings, growing slower; Alibaba and Tencent are 5-8% growers at 13-15x earnings — “you do the math.” Harry’s catch — doesn’t “buy what everyone hates” contradict “the good ones don’t call you back”? — earns a concession: “that’s fair.”
The tension he owns: he’s a ByteDance investor who thinks teen social media is “absolutely horrible.” In China the product is highly regulated — “kids in China go on TikTok to read about science experiments and math projects… in the States I assure you that’s not happening.” He’d follow Australia’s under-16 ban: “social media is the demise of society.” His other worry is income inequality — the factory workers he grew up with in Michigan are worse off while the 0.1% broke away: “that causes revolutions, to be clear.”
Closing public-market calls: Harry suggests buying and holding Microsoft for 10 years — “the pricing power they have is just absolutely incredible”; Mitchell calls Satya an “absolutely incredible CEO” — and on a 30% drawdown, “I would buy Snowflake or buy Datadog and put it in a drawer” for 10-plus years. The other side of the book: MicroStrategy — issuing debt to buy more crypto “just sounds like a house of cards to me,” crypto itself “reminds me a little bit of the Tulip craze” since “if I could go buy a Tesla with crypto it’d be amazing… but you can’t.” He owns no crypto, missed Chainalysis and Coinbase, and admits “I should have bought Bitcoin too — clearly I could have made a fortune.”

Julia Hoggett, CEO @ LSEG plc: The Myths and the Reality of The London Stock Exchange 20VC: Delivery Hero's Niklas Östberg on Losing $200M on Gorillas