Summary

The AI capex bubble ends badly — “it’s like the telecom bubble all over again,” and Apple may end up looking like the smart one. Green’s tell: VCs “have to portray the view that software is going to be dead because they have to justify how much money they’re going to spend” — run the assumptions on required earnings and power generation and “it just doesn’t work.” But the crash is the entry point: “That’s when you’re going to buy these companies.” Timing caveat as hedged: no clue when it stops, “it will probably go longer than people think,” and “this Anthropic round was kind of like an IPO.”
Best risk-adjusted returns right now are public software names. People hate the sector (Constellation’s chart is “a ski slope”), which is exactly the Buffett setup — Lead Edge bought ByteDance two years ago when everybody hated China, and Alibaba has doubled off its lows at 15x earnings. The core belief: software’s moat was never R&D but distribution and customer success, so “it is the incumbent’s game to lose” — Workday does $10B revenue, $3B FCF, 98-99% gross retention; Exxon isn’t rebuilding its HR software.
Model commoditization is his biggest AI worry: Google, Amazon, and Microsoft have more training data than new model companies, while Google, Facebook, Amazon, and Apple have a structural cost advantage; Chinese models run at a fraction of the cost, locally — “why would you pay that amount for OpenAI tokens or Anthropic tokens when you can just run DeepSeek?” He called investing in OpenAI at $100B “a little insane personally” — while conceding “I should have invested” if it compounds to a trillion in earnings.
The return machine targets 2-5x in 3-7 years per deal, 2-2.25x net per fund (~20 net IRR), 20-position portfolios, no leverage — “We’re like Cal Ripken. Doubles doubles and triples.” For LP retention, consistency matters more than peak returns. Only one total wipeout ever: 85-90% recurring revenue, 50-60% profitable, ~70% of positions in the pref turns zeros into 0.8x’s or 0.1x’s, which “massively helps return.” Fund seven was just raised at $3.5B.
Sell discipline is the underrated edge: a standing disposition committee meets once or twice a month, and “the fastest way to get fired at Lead Edge is have a company and not tell us when there’s a liquidity opportunity.” Toast: 12% of a $290M fund, sold $180M before the IPO at $40-50 in secondaries — stock now ~$30. The 2020-21 reckoning is industry-wide: funds that underwrote “a 4x in 2 years” are making “a 1.6x in 8 years.”
70% of dollars deployed are special sits and secondaries — when the front door (primary) and side door (secondary) are shut, “we’ll go through the basement window with a pickaxe and buy a derivative”: in Zoom, Sequoia would roll over direct secondary buyers, so Lead Edge bought out LPs of the original Chinese funds. “In a world where LPs and GPs are desperate for liquidity, that part of our business is absolutely booming.”
The famous eight criteria are a strike zone, not a crystal ball. Patrick’s pushback: eight-criteria deals show no correlation with outperformance versus five-criteria deals, so aren’t the criteria necessarily predictive? Green says they need not be predictive: it’s a Ted Williams strike zone that turns 9,000 cold calls into 900 workable names, and “our biggest mistakes have honestly been not swinging at the pitches when they were in our strike zone.”

Deep dive

1. Run the firm like a software company — the KPI is 95% LP retention

Green’s blueprint comes from working inside the machines he now competes with: he and partner Brian were Bessemer’s first two cold callers, partner Nima came from Insight — “one of the best technology investment machines on the planet.” The stated goal is generational: build the next TA Associates, General Atlantic, or Sequoia, which requires being “extremely rigorous.”
The single number the firm runs on: “our number one KPI… is what is our gross dollar retention for LPs. We want like 95%” — achievable only through good returns and great client service, across people who come and go.
The funnel starts with ~18 analysts aged 22-24 talking to ~9,000 companies a year, guided by a framework whose real job is triage: “in the investment business, we have one asset. It’s time… how do you guide people to say no quick?”

2. 800 executive LPs are the sourcing, diligence, and distribution weapon

95% of capital comes from world-class execs and entrepreneurs (~800 LPs), deployed across the whole cycle. Sourcing: if an automotive software CEO won’t call back, Rick Wagoner, former CEO of GM, sends the note. Diligence: Ian Read back-channels a company’s Pfizer contract and offers to introduce Biogen’s former CEO. Post-investment: “Toast is looking for intros to these restaurants. Do you know anybody?” — “all these people invest in funds and never get asked for help.”
The origin was defensive, not clever: Green knew returns in tech flow to the top 10% of funds, and asked “why in God’s name is anybody going to take my money? I teach them how to ski.” Had he been P&G’s global head of HR when he called Workday 80 times at Bessemer, the CEO would have engaged. In a market “exponentially more crowded than 15 years ago,” the LP base is the differentiation.

3. Doubles and triples: the return math that avoids zeros

Per deal: 2-5x in 3-7 years (~25 net IRR); per fund: 2-2.25x net with 20 net IRRs, in concentrated ~20-position portfolios. The path to a 3x net fund isn’t a grand slam — it’s a 7-15% position going 8-12x. “We’re like Cal Ripken. Doubles doubles and triples… not Sammy Sosa or Mark McGwire.” For the 95% gross-retention goal, Green says consistency matters more than peak returns.
Downside engineering is the actual secret: only one total loss of capital ever. 85-90% of companies are recurring revenue (“if you invest today and know what revenues are in July, that’s a pretty good way to invest”), 50-60% profitable, ~70% of positions in the pref, almost no debt — so zeros become 0.8x’s or 0.1x’s, which “massively helps return.”
They’ve been called traders by hedge fund types; Green’s reply: “No, no, we’re just trying to actually make money” — versus riding a living-dead company for a decade.

4. Very few firms are good on the sell — Lead Edge built a committee for it

The claim: “a lot of firms do a really, really good job on the buy. Very, very few firms do a very good job on the sell” — private equity does it better than venture growth. A disposition committee mirrors the IC and meets once or twice a month; “the fastest way to get fired at Lead Edge is have a company and not tell us when there’s a liquidity opportunity.” Average hold: 3.5-4 years.
The Toast specimen: 12% of the $290M fund three ($36M in), $180M sold before the IPO — secondaries at $40-50 against a stock now around $30 — with $350-400M expected in total. To “why are you selling? You don’t believe in us?”: none of the co-investors put 12% of their fund in, and “somebody is paying us a price in the secondary markets that we think is just like lunacy.”
His honest deflation of the 2015-2018 returns people get excited about: they were “just multiple expansion and we sold. That’s it.” And the reverse hit everyone in 2020-21 — across all alternatives, funds that underwrote a 4x in 2 years are making “a 1.6x in 8 years,” with huge industry impact still coming.

5. The Lead Edge 8 is a strike zone, not a crystal ball

The criteria: $10M+ revenue and product-market fit, 25%+ growth, 70%+ gross margins (“you trade on multiples for earnings” — revenue multiples are just shorthand), recurring revenue, bottom-line profitability, no customer concentration, and the one that “kept us out of the most trouble”: capital efficiency — are your revenues today greater than your historical cash burn? A one-to-one ratio; $20M of revenue on $10M burned, not $80M.
The funnel math forces looseness: requiring all eight cuts 9,000 companies to 90 — too few for 5-7 deals a year. Requiring five yields ~900, of which 150-175 get diligenced.
Patrick’s pushback — worth keeping: eight-criteria deals show no correlation with outperformance versus five-criteria deals, so aren’t the criteria necessarily predictive? Green says they need not be predictive: it’s Ted Williams’ hitting zone — you can homer on a pitch two inches above the plate, but “if you do that over an entire career, your entire career won’t be very long.” And the confession: “our biggest mistakes have honestly been not swinging at the pitches when they were in our strike zone.”
On price: the shorthand test is whether you’re in the money after 18-24 months of growth — if not, “you’re paying way too high a price.” Assuming 20-25x exit multiples, as in 2020-21 and “all this AI stuff,” is “insanity” — though Toast at 10x revenue worked because revenue went from $10M to $25M growing 150%.

6. Cold calling is “investigative journalism with sales”

What 10,000 calls taught: most things are noise, “more responsive CEOs tend to be better CEOs,” and the life lesson — if you tell an entrepreneur you’ll make the Adobe intro, actually do it. “If you’re known as a firm or a person that actually does what you say you’re going to do, it goes a long way.”
They hire former athletes because the failure bar is calibrated differently: “getting a C or a D on a test is not your biggest failure. Dropping the ball at the Rose Bowl… that’s failure.” The craft is bracketing numbers out of founders — “I saw on LinkedIn you have like 80 employees. So what are you, 10 million revenue? 15?… What are you growing, 150%? — Yeah, not that fast. — What, like 100%? — Yeah, around there.”
AI supercharges the analyst: “you give them the power of knowledge and you can sound super smart.” But sales cycles can be a decade, and the competition (Summit, TA, Insight, Battery) is calling the same founders.

7. Software is the incumbent’s game to lose

The heterodox core: “the competitive advantage of software has never been about R&D.” Microsoft could kill any Lead Edge portfolio company with 500 people and a month — “they just don’t care about the Chamber of Commerce market.” Software is about distribution, sales and marketing, and customer success.
The Workday case: 98-99% gross dollar retention, 10-15% growth at $10B revenue and $3B of free cash flow, customers who spent 3-5 years implementing. “If you think they’re going to start building their own HR software, you’re out of your mind” — thousands of Workday engineers beat “Mitchell Green’s cousin vibe coding his way to build Workday.”
The real disruption risk is self-inflicted: Coupa only existed because SAP bought Ariba and “left it for dead.” Green worries over-levered PE-owned software assets may be pushed toward “Rule of 50” by cutting R&D and sales — making them ripe for disruption, versus independent software companies focused on growth.
The precedent as told: in 1999 everyone thought big-box retail was dead, yet the top-10 US e-commerce list reads Walmart, Home Depot, Lowe’s, Macy’s, Target. Incumbents win — except the over-levered or non-innovating companies (Montgomery Ward, Kmart, Sears).

8. The AI capex bubble ends badly — and that’s when you buy

The market call, unhedged: “Overhyped, overfrothed… I believe this AI CapEx bubble will end badly. It’s like the telecom bubble all over again” — and Apple “may look like the really smart one at the end of the day.” The tell: VCs “have to portray the view that software is going to be dead because they have to justify how much money they’re going to spend” — but map the capital going in against required earnings and power (“where are the nuclear power plants?”) and “it just doesn’t work.” The payoff: “that presents the opportunity. That’s when you’re going to buy these companies.”
The mechanism: models will commoditize. Google, Amazon, and Microsoft have more data to train on than new model companies will ever have; Google, Facebook, Amazon, and Apple have a competitive cost advantage; Chinese and European models cost a fraction to run and run locally — “why would you pay that amount for OpenAI tokens or Anthropic tokens when you can just run DeepSeek?” Timing stays hedged: “it will probably go longer than people think,” and “this Anthropic round was kind of like an IPO.”
Where he is excited: infrastructure software, because “agents appear to consume more resources than actually people” — consumption-based models like ClickHouse (early investor, burned “like nothing”) and Grafana Labs, with Datadog compounding high-20s/30% at scale. “I really struggle with valuations, but the growth rates are like we’ve never seen with very good economics.” Many of these companies don’t fit the machine: “we struggle with — are they going to be 200x’s or 100x’s or zeros.”
On the bigger arc he’s a bull: AI is “the biggest productivity gain of the last 75-100 years… pretty damn close” to electricity, and “the age of entrepreneurism.” He scores every portfolio company on AI readiness — data structure, new AI products shipped, AI revenue — explicitly not whether engineering comp fell: if you budgeted 150 engineers, keep 150; they’re exponentially more productive.

9. Through the basement window with a pickaxe: 70% of dollars are special sits

The house analogy, as told: front door is leading the primary round or buying the business; side door is buying out an early investor or employee; and when both are shut, “we’ll go through the basement window with a pickaxe and buy a derivative.” Less control, less insight — “you trade off price for access.”
The Zoom specimen: no primary (company didn’t need money), no secondary (Sequoia would roll over them — “they’re not dumb. Why would we let these knuckleheads in?”). But day one Zoom was backed by “a bunch of random Chinese people and Chinese funds” whose LPs were 10 years in — so Lead Edge bought the LPs out or stepped into their shoes, with governance strings: call us on any vote, hand over the stock on day 181 after IPO lockup.
Today ~70% of deployment is special sits and secondaries, run by partner Tim Beamer: “in a world where LPs and GPs are desperate for liquidity, that part of our business is absolutely booming” — and “we are a market drawdown away from it exploding.”

10. Culture is tracked like a metric — and the founder keeps score

Culture comes from the top, operationalized: Green sends handwritten thank-you notes to nearly everyone he meets — “guess who also does now? The 22-year-old analyst. And by the way, we track it and report on it.” He interviews every employee annually (an idea from Tom Barrack at Excel KKR — a firm he puts on his machine Rushmore alongside Insight and TA): green/red/yellow your job, tell me what you’d change if you ran Lead Edge. And 23-year-olds get put in front of LPs — “99% of firms on this planet wouldn’t.”
His Rushmore logic: Insight talks to ~30,000 companies a year — “an absolute factory.” IC debates are deliberately combative: “if you sat in the room you would think the three of us hate each other… no, that’s just how we talk” — then buddies right after.
The formative edge is ski racing (500-foot hill, endless repetition — the Lindsey Vonn Buck Hill model), and the line that got him hired in early ‘08 by Scott Booth: “because when things get scary, you’re going to want to buy.” When fall 2008 hit: “this isn’t scary. What’s buy?” He sleeps 4-5 hours, races cars competitively, and on why he still cares after the money: “keep score every day… I want to win.”
The closing note: the kindest thing anyone did — Pete Willmott, former FedEx CEO, backed his failed college company, then told Bessemer “they were insane if they didn’t hire me, cuz I was the most persistent person he ever met.”

How Elon Thinks Market Overview — March 24, 2026