Benchmark vs a16z: Why Stage Specific Firms Win

Summary

• Megafunds have already won round one — by having the money. Jason’s math: the top multi-stage firms control 50-60% of venture capital, and “if you have all the money and you do all the deals, you get all the wins. It’s just math.” Seed and Series A become loss-leader products — “milk at the grocery store… we’re going to upsell you all the strawberries you can buy, baby” — and for mid-size firms the next 3-5 years will be “a pain in the ass” competing against walls of capital.
• The whole focus-vs-scale dilemma sits in one dataset: Benchmark did ~63 Series A’s with a 10% hit rate on $5B companies over ~15 years; Dre did 454 with a 2% hit rate — but 10 absolute hits to Benchmark’s six. “That’s all you need to know.” Layered on Josh Coppelman’s venture arrogance analysis, the sobering conclusion: no firm has ever achieved the market share the megafund models require — the strategy only works if companies compound from $100B to $400B in private markets, and “right now there’s two: SpaceX and OpenAI. If you have four or five more of those, the math works for everybody.”
• Jason bet $100,000 that AI unemploys more tech workers in 12 months than anyone expects. He didn’t believe it 90 days ago; after running his own AI over 130,000 conversations he’s “100% convinced” — he’s already cut five of his own team, and “as soon as AI is even 80% as good as a human, they’ll all be gone.” His deeper claim: “no one wants to work” — the LinkedIn open-to-work circle “means I need 300K and I’ll do three meetings a week.”
• Rory’s counter: GDP and productivity growth have been ~2% since the industrial revolution and he’ll “lean into” 2% for 20 more years — deep research is not the steam engine, and Klarna will end up with “roughly the same profits as other lenders.” But he concedes tech is now the economy’s largest sector, so AI adoption “might be boomier or quicker” than the PC/internet — massive productivity in tech, Baumol’s cost disease everywhere else. Both agree the hands-off-keyboard middle manager “is going to be gone any year.”
• Decagon at 100x ARR ($15M at $1.5B) exposes the industry’s core vice: Rory’s steelman is that genAI lifts support resolution rates from 30-35% to 60-70% with the clearest ROI in AI — “not wholly crazy” — but his confession carries the episode: “We love new shit with option value over old shit with intrinsic value… We are upside junkies,” and buyers may be pricing a one-in-a-hundred outcome “as if it’s one in two.”
• The killed-vs-maimed framework is the tradeable warning for SaaS: AI doesn’t need to kill incumbents, just maim them — knife-cut churn, downgrades, and pricing pressure that drop growth from 50% to 30% and take companies off the IPO track. HubSpot is 50% more productive with Cursor and now builds more features than it can push to production — “think about that when you think you have a stable state.”
• Windsurf’s potentially confirmed $3B acquisition by OpenAI already feels small (“I’m anesthetized to these numbers now that aren’t in the tens of billions”), and the exit wave is just starting: 800-1,000 unicorns exist, a couple hundred go public, and the rest get “squeezed into other companies over and over again.” For anyone being eclipsed: “take the effing offer” — and per Coppelman, this is “the game on the field for the next five years… quit bitching and play it.”

Deep dive

Note: this episode is the 20VC roundtable — Harry Stebbings with recurring guests Jason (Lemkin, likely) and Rory (O’Driscoll of Scale, likely).

1. Windsurf’s potentially confirmed OpenAI acquisition — and $3B now feels small

• The news broke half an hour before recording, and Jason’s reaction is the tell on where the market’s head is: “I thought three billion was a lot. When I saw the tweets again, I thought it’s not that much… I’m anesthetized to these numbers now that aren’t in the tens of billions” — every junior OpenAI engineer is raising at $10B pre. Rory: the deal “makes total sense for OpenAI” — 1% of market cap to play in one of AI’s largest use cases, with the developer constituency that loves it.
• On whether Cursor shareholders who just paid $10B should worry: “The time to have checked your manhood was before you turned down the big number.” Rory’s logic tree from experience: when you’re number two and the adjacent acquirer squeezes, fold — otherwise they buy number one and you’re done; number one can tell itself it’s the independent winner with other acquirers. “It takes real courage to turn down whatever was offered.”
• The meta-point on the country: half-pivot into “a Visual Studio kind of fork” two years ago, crank like crazy, sell for $3B — “all you have to do is just get it right and Bob’s your uncle.”

2. Around the M&A table, everyone runs their eternal cap table

• Jason on why post-deal advice is “wildly divergent”: the investor who just put in at $10B is zen (it was a risk factor in the prospectus); the $1B-round lead thinks not selling is the worst idea ever; the seed folks who sold half in secondary are “kumbaya.” Rory’s advice to CEOs facing an offer: “understand those numbers for everyone and understand where people are coming from” — everyone talks their book, and not always in the way logic predicts.
• The vibe has shifted as funds got bigger: Andrea (likely Andreessen) invested at 13 in Loom, and they got their 1x back in a year — “that’s cool guys, let’s do the next one together” — at the growth stage. Rory’s sharper example is Thrive on Instagram: a 2x in days, not the target return, but “that is the sound bite that leads almost every Wall Street Journal description of them” — sometimes it’s about perception and what it does for the firm at that moment.

3. Let your winners run — the final double doubles the fund

• Jason’s confession of the seed-investor pathology: once you have your first 10x in a double-digit-millions fund, “you become risk-averse… you lie awake at night the night before the IPO.” He took money off the table early on a deal that could have gone from 2x to 10x of that position — “which to state the obvious is a 20x. And that’s a big difference.”
• Rory’s rule is military doctrine: “reinforce success, starve failure.” If Sequoia just marked you up late-stage, that should smell like the deal — a 2x from there, which he says turns your 10x into a 30x, and letting winners run “is one of the key parts of making the math work.”
• Harry channels Brian Singerman’s “value of the final double”: 6 to 12 billion in company trajectory can be 12 months of work and doubles portfolio returns. Rory agrees and goes further: almost everything about PE is better at making money than venture — the only edge is that once every five years you find yourself with 10% of something compounding toward hundreds of billions, “so much easier than grunting it out from a million to 5 million… and now you’ve turned 10 million into 25 million of value. Whoop-de-doo.”

4. Harry’s flip: multi-stage megafunds win the decade — Jason: step one, they already have

• Harry’s changed mind, stated as a thesis: IPOs have shifted to private markets, sovereign wealth funds have already chosen their five-to-seven mega-partners, trillion-dollar outcomes are normalizing, and megafund cost of capital is so low “they can shit on me and you and do 10 on 100 for a seed round — as they’ve done twice in the last year to me.” Conclusion: multi-stage wins the next 10 years.
• Jason’s three-part response: first, “step one in winning is if you’ve got $7 billion to invest from 2024 on” — with 50-60% of the capital, “if you have all the money and you do all the deals, you get all the wins. It’s just math.” Second, whether they can invest it profitably enough for LPs to re-up in three, five, and ten years is genuinely unclear. Third, either way the next five years are “a pain in the ass” for mid-size firms — the mafia dynamic: “that’s a really nice Series A you got there. Be a shame if it got broken.”
• The bundling mechanics: seed and Series A are now loss leaders — “milk at the grocery store… you wait till you see the Series C and D.” Harry pushes it further: Lightspeed and General Catalyst do more pre-seeds than anything, and “you won’t see the A if you don’t do the pre-seed.” Rory admits Scale is wrestling with exactly this — he lost a deal months ago where the winner’s edge was a seed relationship — but won’t fake it: “if you’re going to say you’re doing it, you got to do it.”
• A reinforcing macro worry: Trump’s threat to Harvard’s tax-exempt status. Rory won’t play pundit, but the venture impact is direct — even if it never lands, precautionary cash planning at endowments squeezes “the LP of choice for the small early innovative funds… the big will get bigger and it’ll be harder to be new.” And don’t kill the golden goose: venture’s non-negotiable ingredient is a well-funded university system generating the graduates and research “that kickstarted the whole thing.”

5. Founders don’t care about your strategy — and no one does real inception work

• Harry’s zoom-out insight: a founder wants “money, lots of it, with the minimum amount of hassle and perhaps the maximum amount of help” — there is no founder-side forcing function on investment discipline. “Someone who’s loosey-goosey drunk with money is their best friend.” Bundling only stops if its returns prove subpar versus specialists, “and even if that’s true, it’s going to take five to seven years to become obvious.”
• Rory’s pushback: founders in processes ask him how many deals he does a year and want to hear two-to-three, not twelve — they want to feel the love. Firms doing 20-30 deals a year get a news-flow advantage — “there’s always something good in the portfolio” — and return quality is an investor-LP problem, not a founder one.
• Jason now folds pre-flop on hot seeds: when the email says 10 at 100, “I just email them back: I can’t compete. I don’t take the meeting.” And he’s scathing on what passes for inception investing — real inception is finding the kid at the carriage house who didn’t go to college, Stripe, or YC and spending months before writing 800K. “That’s too much work. No one works that hard these days… It’s easier to just pay 30% more than Harry.”

6. Benchmark vs Dre: the entire megafund dilemma in one dataset

• The Series A analysis by “Rottman at DST” (the analysis was printed and read for two or three hours): Benchmark did ~63 Series A’s with a 10% hit rate on $5B companies across ~15 years; Dre did 454 with a 2% hit rate — but scored 10 absolute hits to Benchmark’s six. “In those numbers is exactly the dilemma of the mega fund versus the focus fund… That’s all you need to know.” (Harry had remembered Benchmark at 33%; Rory corrects it — 10%, with the other 18 firms bunched at 1-3%.)
• The implication Rory draws: scale up and “your quality does slip, but the aggregate numbers keep going up” — the only question is which strategy makes more money, and whether the volume strategy clears the return threshold. A caveat stands too: those were 2013-2018 deals — “there’s nothing in common with today. Nothing.”
• Cross this with Coppelman’s venture arrogance score and it darkens: if a fund model assumes getting into 20% of all good deals in 2026, note that the most aggressive firm in a less competitive era got 10% and everyone else got less. “The conclusion is no one has achieved the market share that it would require to make this math work.” The only route left: stuffing huge money into late-stage compounders held private for longer — probably lower returns, “but probably over the hurdle rate. That’s what the bet is.” The duration kicker Harry flags from the piece: a 2x in 10 years is similar IRR to a 4x in 17 — “time value of money is a bitch.”
• So it won’t be won on Series A share. Rory: the deciding variable is whether companies compound from $100B to $400B in private markets before going public — “right now there’s two: SpaceX and OpenAI. If you have four or five more of those, the math works for everybody — provided you’re in them.”

7. Jason’s $100K bet: half of tech knowledge work disappears

• Grounding first, from Jason: US GDP ~$30B, the stock market ~2x that at roughly $60-65B, and history says “roughly a trillion dollars per decade of new value created, plus or minus” — so stop counting trillion-dollar exits; “$100 billion is the mental high end of good.” Though he can’t resist the venture flex: of the six trillion-dollar companies, all but Berkshire were VC-founded — “to the extent trillion dollars is possible, it’s only possible in venture.”
• Jason’s conversion is the episode’s most vivid change of mind: he didn’t believe half the tech labor force gets replaced “90 days ago. Now I’m 100% convinced” — after running his own AI through 130,000 conversations. The wager: “I will bet you $100,000 that more of these people are unemployed 12 months [from now] than you think.” His own firm cut five people in 90 days — “it’s better… and they don’t complain about the job. As soon as AI is even 80% as good as a human, they’ll all be gone.”
• The deeper claim isn’t opex, it’s culture: “No one wants to work. That’s the problem, Rory, that VCs are missing.” The specimen: a marketing manager he’s known for years, out of work six months — “I need to make at least 300K and I just want to attend meetings.” The LinkedIn open-to-work circle “means I’m unwilling to work. I need 300K and I’ll do three meetings a week.”
• He now believes what the likely Vinod Khosla has said for years and he once laughed at: half these people will be gone, “there will be no jobs for them,” taxes go up to pay for them. His test case is Benioff’s plan to repurpose 6,000 Salesforce support staff into sales: “how do you repurpose 6,000 people from AI?”

8. Rory’s counter: 2% growth since the industrial revolution — but tech adopts fast

• The boring-nerd macro case: “the past is the best predictor of the future” — GDP and productivity growth have run ~2% since the dawn of the industrial revolution and he’s “willing to lean into” 2% for 20 more years. Deep research is cool, but against the steam engine pumping out the mine or electricity, “at best the inventions are equivalent.” No mass unemployment in 12-24 months; Klarna talked bold but will end up with “roughly the same profits as other lenders.”
• Harry’s rebuttal — worth keeping: PC and internet adoption required installation, fiber, hardware, implementation. “We now press deep research on a model we were already using and we can get rid of three researchers. There is no installation.” Another speaker’s real concession follows: tech is now the economy’s largest sector, so the early-adopter base is structurally bigger — “it might be boomier or quicker,” with massive productivity gains at Salesforce and “Baumol’s cost disease” of inefficiency in health, education and government. All three converge on one casualty: “if you’re a hands-off-keyboard middle manager, you’re going to be gone any year.”
• On the labor mismatch, one speaker gives Peter Thiel credit from 10-15 years ago: college return is fine for STEM and strong students, but “the marginal return on the marginal entrant is profoundly negative” — unmarketable skills plus $150K of debt, chasing soft jobs that are vanishing. One speaker: elite college is already quasi-UBI (Harvard and Stanford free below ~$200K income). The consolation: “if you’ve got three really smart friends at STEM and can crank out a Visual Studio fork, you too can have $3 billion in 24 months. There’s always going to be room at the top, baby.”

9. Decagon at 100x: “we are upside junkies”

• Decagon raised at $1.5B on $15M ARR — 100x, 2021-style. Rory’s steelman: customer service is AI’s #3 use case with the clearest ROI — his references say genAI moves resolution rates from 30-35% to 60-70%, meaning most calls handled without humans in a massive market. “It wasn’t wholly crazy” — though he admits Scale debated joining the hundred people pleading and concluded “that is not a Scale deal.”
• Jason’s con case, delivered as a gauntlet: Intercom is “fucking brilliant,” took 17 years and a lot of money to reach a not-huge ~$2B value, and is one of ten players. “You want to go against Brett Taylor [at Sierra], and Neil Mehta bankrolling him? And then the 50 others coming out of YC… Seriously?”
• One speaker’s explanation of the mechanism is the episode’s best self-diagnosis: “We love new shit with option value over old shit with intrinsic value… We are upside junkies.” Something at 400M growing 20% is bounded — buy at 6x, sell at 6x; something going 3M to 25M lets you story further growth to $300M and a $6B future. The catch: “the probability of that working might only be one in a hundred and they’re pricing as if it’s one in two.”

10. Killed vs maimed: AI maims leaders off the IPO track — take the offer

• Jason’s moats-versus-momentum worry, from the likely Gorgias board where he sees all vendors’ data: Gorgias is objectively best “but the gaps are narrow,” disruption now arrives “literally every five weeks,” and someone at Windsurf claims no moat beyond “working harder than everybody else in speed.” One speaker’s counter: market windows — “momentum begets its own moat”; once two or three players hit critical mass and become the safe choice, “windows shut.” Another speaker’s proof: “I don’t think Salesforce was winning in 2010 because it was the best CRM. It was winning because it was the default option.” Harry’s endgame: enterprise apps settle into an oligopoly of three-to-four — “if you’re wrong in that vision, these assets aren’t worth 10 times revenues, they’re only worth five, and everyone is so horribly wrong my head hurts.”
• One speaker’s sharper frame — killed versus maimed: “AI can maim leaders even if it doesn’t kill them” — knife-cut churn, downgrades, and half-price renewals that drop you from 50% growth to 30% and off the IPO track. “That can destroy venture investing.” Founders should treat elevated churn as “a canary in a coal mine,” not let the CRO wave it off. The proof point: HubSpot’s Yaman Rangan said they’re 50% more productive with Cursor, measured by code commits, and for the first time ever are building more features than they can push into production.
• The exit reality check: Census — once so hot Sequoia put in part of $80M raised — quietly acquired by Fiverr (as spoken; likely Fivetran), leaving Rory-the-grouchy-seed-guy with “eight shares in a decacorn that will never IPO.” Rory shrugs: there are 800-1,000 unicorns, a couple hundred get public, and the rest “are going to have to be squeezed into other companies — over and over again.” Even Lacework, once neck-and-neck with Wiz, sold for nickels of enterprise value — though Rory notes investors likely recovered 50-70 cents on the dollar of the cash still on the balance sheet.
• Hence the advice to the eclipsed: SevenRooms got $1.2B from DoorDash and OLO is shopping itself — “take the effing offer.” And Deliveroo to DoorDash at $2.9B versus a UK public value of $1.4-1.5B draws the kicker: “it may be that you guys are just crap at valuing tech companies.” The closing note channels Coppelman: this capital glut “is the game on the field for the next five years… quit bitching and play it.”