Is DPI The Only Thing That Matters? with Sam Lessin, Jason Lemkin & Rory O’Driscoll

Summary

• Note on the episode: the feed title promises an Omaze storytelling interview, but the captions are actually a 20VC roundtable — Harry Stebbings with Jason (likely Jason Lemkin of SaaStr), Rory (likely Rory O’Driscoll of Scale), and Sam (likely Sam Lessin). The spine of the episode: DPI is the only score, most companies “don’t matter,” and AI is repricing every layer of software.
• The panel reluctantly sides with Chamath’s “you can’t eat IRR. You can only eat net DPI.” Sam’s sharpening: there are two different games both called venture — actually making people money (a DPI game) and asset gathering, which “is actually a better business” but one he has “no respect for.” Sam’s proof that liquidity gets capital: Thoma Bravo raised a record $34bn fund after $30bn of distributions last year, in a quarter when nobody else could raise $5bn.
• Sam’s pushback is the best counterpoint: calling TVPI meaningless is “vaguely right, but not useful” — venture consciously buries money for 5-7 years, so TVPI is a loose proxy with real signal in it. Sam’s machine-learning example compares funds at 2x versus 8x TVPI at year three or four. His alternative scoreboard ignores marks entirely: “just give me the list of four things” that are credible fund returners.
• The middle of VC is hollowing out: SVB pegs $200-500M funds as the squeezed zone, Sam calls the billion-dollar fund “the death zone,” and Rory’s implication for founders is blunt — if nobody’s left to write the $20-30M Series A, “get to know folks with five or 10 billion funds before demo day.”
• On selling: “you cannot sell the things that matter” — but Harry argues Chime’s seed holders would have wanted to sell into the $25bn round (they’re absent from the S-1 principal-stockholder table), and one panelist would sell Revolut and Chime right now: both sit near 15% market share and “CAC only goes up.” Harry’s counter: Revolut is “quite likely” at least $100bn, likely $150bn — selling at 25 forfeits $125bn of gains.
• The “mattering” fight: one panelist says Chime and Box don’t matter, most companies don’t, and even OpenAI only “might matter” in 20 years; the rebuttal — “if Chime is not an important company… I’ve never done anything important in my life.” The panel’s resolution: market cap is a rough proxy for mattering, because “the capitalist system works.”
• From the Mary Meeker report: ChatGPT hit 800M users in 17 months, the big six spent $212bn of capex (“they would break Mr. Buffett’s heart… capex hogs”), token costs collapsed 99.7% in two years, and the “$600 billion question is where are the apps.” Rory’s overlay: OpenAI’s revenue tracks Google exactly 20 years later, but projections assume it becomes twice the Google of its day — merely Google-good means missing next year’s number by ~40%, and “public company investors are just mean VCs on steroids.”
• MCP is an existential threat to SaaS: agents abstract the app away (“I could become a pipe overnight,” per the Mangomint founder), value accrues to the system of work not the system of record, and the “AI slow roll” is the #1 thing killing B2B. Quickfire bets: the Jony Ive device family gets a screen, Meta ships a closed model (“I wouldn’t bet against Zuck”), and Jason puts 50% on Elon out as Tesla CEO before 2027 — “he can’t scale… with the 58 kids and the 11 companies.”

Deep dive

1. Chamath is right, annoyingly — you can only eat net DPI

• The trigger is a Chamath tweet — “TVPI’s a vanity metric. You can’t eat IRR. You can only eat net DPI” — and Sam’s reaction sets the tone: “I hate agreeing with Chamath on principle, but I agree with him.” His framing splits the industry in two: one game is “actually making people money — finding companies early, making the right bets, paying the right prices and selling… a DPI game.” The other is asset gathering — and here’s the honest part: “the asset gathering game is actually a better business” if you’re just in it for money. “I just am an intellectual snob and I have no respect for asset gatherers.”
• Sam’s market evidence: Thoma Bravo just raised a record $34bn PE fund in a quarter when “no one raised a $5 billion PE fund because of lack of liquidity” — and they had $30bn in distributions last year. “Not paper markups, but distributions.” If you have liquidity in today’s world, you get the capital.
• Sam’s pushback — worth keeping in full: the tweet is “a very typical Chamath comment. On first glance it sounds smart, but on deeper analysis it’s vaguely right, but not useful.” Venture consciously buries money for 5-7 years, so TVPI is “a loose proxy for performance” — and he’d bet that a machine-learning pass over funds at 2x vs 8x TVPI at year three or four would find real signal in ultimate outcomes. “To state it means nothing is a gross exaggeration.”
• Sam’s alternative scoreboard, plus the LP mechanics: marks are “all marketing” — funds hold everything at “the high water mark they can possibly come up with” — so early on, “all I care about is give me five names that matter,” and later, did you make me money. The marketing exists because “institutional LPs are people too” — some junior person who wrote the check wants promotion before year seven, “because everything in life is about get laid or get paid.”

2. The hollowed-out middle and the billion-dollar death zone

• An SVB analysis says mid-size VC is getting hollowed out — Sam checked and the squeezed zone was $200M-$500M funds, prompting Rory’s self-deprecation as a $900M manager: “now I’m a small behemoth, which is even more degrading… a low rent conglomerate.” Sam’s sizing rule from the other end: ~$200M early-stage funds are “not a great fee business, but a great DPI business,” while “the billion dollar zone is like the death zone”; at $10bn you’re playing a different asset-gathering and asset-deployment game.
• Rory doesn’t fully buy the data but concedes the mechanism: you must be “sized for the stage you’re playing at” — his construction is $20-30M checks into $30-40M rounds, 30 positions — and the conglomerates have “made everything harder, including it’s harder for us to make money.” A $100-200M deployer is more inured to $10bn funds than a $500-900M one.
• Rory’s founder-side implication: if the middle disappears, the $20-30M Series A — “not even a large series A today” — comes from maybe ten mega-fund names writing 40 checks a year. “Founders just better get to know folks with five or 10 billion funds before demo day or they’re all going bankrupt.”
• Harry resolves the seeming paradox of scarce-and-abundant capital: “every generation there’s only a few companies that matter.” Jason’s follow-on is blunt: “most of the money at series A is completely wasted” — mega funds win by deploying huge sums “at almost any price,” seed survives as a multiples game with private-to-private exits. Rory’s rebuttal: “there are more wins than the biggest win” — Hinge Health just returned $400M; 20% of that is $80M, and “every single one of you will cash the check… $80 million is still real money in America.”

3. Selling is the harder skill — and one panelist would sell Revolut and Chime today

• Rory’s rules of selling: rule one, “you cannot sell the things that matter”; rule two, be honest when the thesis breaks (he cites being early in Albert and Astra — “thesis broke”). The asymmetry he flags: “you get hundreds of shots to buy… you have so fewer shots at learning to sell that it actually takes much longer to learn to sell well” — yet it matters just as much.
• Chime is the case study. Harry argues an early investor would have wanted to sell into the $25bn last round — “is it an infinity company? … I did my job, which was to fund it early.” Harry reads the S-1 the same way: no seed names on the principal-stockholder table, “I literally only see DST and Crosslink” — the seed guys likely sold near 25 and “did their job,” versus a roughly $14bn market cap. Rory’s defense of the mid-tier: Menlo’s Series B at roughly $200-300M will still book 10-20x — “proof that you can make perfectly shrewd A-to-C bets sub a billion pre and make very good money in a non-‘generational’ company.”
• The live call: “I would probably sell Revolut and Chime right now” — both around 15% market share, and “market share is a limiter” once you’re past the easy growth. The supporting law: “CAC only goes up… the argument that you’re going to get good at something and all of a sudden it’s going to get cheaper is just wrong” — with the partial Chime exception that word of mouth became its #1 customer source only since 2022.
• Harry’s counter and the unresolved question: Revolut “will quite likely be at least a hundred billion dollar company and likely 150” — sell at 25 and you miss $125bn in a European neobank. The question remains whether, as a venture capitalist, you should have an opinion about public markets — “once it’s public, the game should be over” — something the speaker admits he’s “gone back and forth on.”

4. Does Chime matter? The fight over what “important” means

• One panelist, having praised the team: “is Chime an important company? I just don’t see it… they provided some banking products to underbanked people. It’s not bad. But is it important?” The rebuttal takes it personally: “if Chime is not an important company at all, then I’ve never done anything important in my life… I’ve helped thousands of founders build companies from scratch, but I probably don’t matter if Chime doesn’t matter.” The panel doubles down: “the default is that most of us don’t matter… the job is to find the things that really matter.”
• The list of what clears the bar: Microsoft, Facebook, Google, Bitcoin — “I personally think Solana matters. I think Venmo might matter.” On OpenAI, only a hedge: “will we look back in 20 years and say OpenAI was a fundamentally important company? I think it might matter.” The self-definition: “the most insulting thing you could ever call me is a market participant” — being a cog in efficient capitalism is “a recipe for doing fine. It’s just not my goal.”
• The Box roast crystallizes it. One panelist says Aaron Levie’s $5-6bn company “doesn’t matter” — though another told Levie at IPO he deeply respected that “you just ground it out” — and suspects selling to Citrix in 2008 for ~$100M plus retention, plus his Stripe and Gusto shares, would have netted more than going public. Jason’s sarcastic reductio: “he’s wasted his whole f*ing life is basically what you’re saying.”
• The resolution — the investable version: “market cap is a rough proxy for mattering… the capitalist system works and the things that matter the most are valued the most.” Balfour is quoted — “nothing matters a lot and very little matters at all” — and the only-do-what-matters creed “tends to some kind of nihilism.” The closer ties it to AI: companies that were “kind of meh” now see an opening — “the race to matter… that’s kind of the big game to play.”

5. Meeker’s numbers: capex hogs and the $600bn question

• Jason’s first takeaway from the Mary Meeker AI report: ChatGPT went 0 to 800M users in 17 months — Netflix took 15x longer, TikTok 5x. His point isn’t the number, it’s the epistemics: “I don’t think as humans we can process the rate of change of AI… whatever you think AI could do three months ago, it’s super dated” — aimed squarely at B2B people still citing hallucinations.
• The spend side: the big six put up $212bn of capex, and Rory’s framing is the episode’s best line on it — the hyperscalers “have taken very good cash-efficient businesses and they would break Mr. Buffett’s heart because they’ve turned them into capex hogs.” The saving grace is timing: free cash flow itself grew, so the hit was only ~10% — “a little like ‘98-‘99.” The open wound: “the $600 billion question is, where are the apps?” — nobody would sign up cold for $600bn of capex against low-tens-of-billions in revenue and hundreds of billions of losses.
• Rory’s missing variable is time, not direction: “I totally believe all the apps will come… if businesses adopt in two years, it’ll all be fine. If businesses take four or five years, you’re having the costs but not yet the revenues.” The revenue is arriving fast — OpenAI projected at $25-30bn by end of next year, Anthropic $1bn to $3bn in five months.
• Is there a ceiling on investor patience? Rory: yes — “public company investors are just mean VCs on steroids. We turn on a dime from ‘I can’t believe you’re not spending more’ to ‘what the freak do you mean you’re spending so much money.’” Precedent: Meta’s VR spend drew pressure and Zuckerberg belt-tightened. Harry on Altman’s counter-move: Stargate talk of $500bn “just to get going,” up to $5trn — “he’s socializing the whole market… you need all the big participants to say we’re going all in.”

6. OpenAI is priced as twice the Google of its day

• Harry’s overlay, the sharpest analysis of the episode: OpenAI’s revenue matches Google’s exactly 20 years apart — 2022 mirrors 2002, then roughly $1.3bn vs $1.4bn, then $3bn vs $3bn. “It was the Google of its day until now” — but the forward projections now assume OpenAI becomes “twice the Google of its day for the next four years.” Maybe it will — “it is a more impressive piece of technology when you use it” — but maybe not.
• The tradeable kicker: “if all OpenAI is is just as good as Google, then it’s going to miss its number next year by about 40%.” Expectations are set so “anything less than freaking amazing could feel like a fail” — the analogy is Amazon in 2001-02, when a panicky market read minor growing pains as going bust. The scary-moment trigger to watch: OpenAI missing a growth plan, “two rough quarters for whatever reason… would set off a mini panic” — which “could actually be an interesting opportunity to invest.”

7. Tokens down 99.7%, China closing, and the AI slow roll killing B2B

• Meeker’s China point, via Harry: DeepSeek delivered 93% of the performance of OpenAI’s o3 Mini at a fraction of the cost; Alibaba’s model outperforms; Ernie (likely Baidu’s) runs at 2% the cost of GPT-4.5. “We’ve stopped talking about safety and I think we’ve stopped talking about China.” Rory’s read: ChatGPT will be “the Apple-level quality product,” but Android-tier rivals “will just keep pricing honest” — “the undeniable fact is we ain’t a monopoly anymore. There are four or five companies the other side of the pond cranking them out.”
• Token costs collapsed 99.7% in two years, and Jason’s rant at those who missed it is verbatim gold: the “ill-fitting blazer and jeans SaaS guys” and “the mediocre VP of engineering who hasn’t coded in 5 years” telling him AI is too expensive — “find a new job… I tell the CEO to fire that guy.” Rory’s AI board member repeats the operating rule: “you can’t do it today? Build the product — by the time you’ve built it, the stuff will be cheaper. And if it doesn’t work, it’ll work six months from now.”
• Jason names the killer: “I call it the AI slow roll, and I think it’s the number one thing killing B2B companies” — limited Q4 release, wider rollout next year if it goes well: “you’re just going to be slaughtered.” He wants the Windsurf posture — Varun told Harry “startups beat incumbents because of existential dread,” having run three different companies in 18 months — and “I don’t see enough existential dread in B2B startups… I want to walk into a board meeting and see a little bit of shaking.”

8. MCP: be the decider, not the tracker

• The demographic point flagged for B2B: the next 32% of the world comes online AI-first — voice agents, agent-driven interfaces, natural-language interaction. “Kids do not know what a file is. It does not matter.” One panelist’s son, who codes daily, saw the SaaS ticketing tool at SaaStr and said “wow, this is the first time I’ve seen software like this” — the next generation “will never even understand what a Salesforce UI is.”
• The structural claim: “in front of every system of record, there’s going to be some system of work… and it will gradually displace, make irrelevant or replace the system of record if they allow that to happen.” Salesforce instances will exist “at scale long after I’m dead, but the value accretion will all go to the technology that’s helping me do my job, not the technology that’s keeping score on how well my job’s been done.”
• The confession and case study: a panelist had rolled his eyes at Sacks’ agents-as-data-transporters framing as “too nerdy… too Microsofty” — until MCP applications started shipping. His portfolio company Mangomint (best-in-class spa/doctor’s-office SaaS, coming up on $25M) got the wake-up call from its own founder: if booking abstracts through ChatGPT or Claude, “I could become a pipe overnight.” Today the friction is API keys; “when those keys go away… I just don’t think we’re ready for this world.”
• The crisp reduction: “do you want to be the person who keeps track of who’s coming to the spa, or the software that helps decide who’s coming? I want to be the decider, not the tracker” — tracking alone is inherently less valuable. The investor cut: incumbents compound on installed bases for a decade, but “our job is to find the companies that get the new dollar” — and “every company’s market share is up for grabs when there’s a platform shift. The very biggest ones don’t go to zero. They just slow down.”

9. YC at 50-60 post, pre-revenue — admire the game, do the math

• Jason’s field report: YC is pushing the AI companies — ~70% of the batch — to $50-60M post with essentially no revenue, up from the 15 and 20 that once seemed crazy. His posture: “you got to see YC as a business… admire the game that’s being played on the field” — the value proposition of turning two unknowns into something valued at 20-50 pre is why founders keep coming. His own YC outcome: he’s ended up owning “about a little less than half” of what he would have, partially fixed by buying more in later rounds (Revenue Cat, Algolia).
• His pricing discipline, stated as a teaching rule: for seed to make sense the next round should be 3x to justify the risk; for A or B it’s 2x “or you should just wait.” At 60 post, the A has to land around 200 “for the math to really pencil.” Jason adds the fund-size trap: “the low ownership crushes you” — 3% positions don’t return real funds — and correction is a lagging indicator: “it’ll take 8 years to figure out can you pay 60 pre for an AI Y Combinator startup and make money on average” — if 20 gave you a 3x, paying 60 cuts returns proportionally. “But not for YC. And the founders get 6 million instead of 2.”
• Jason also punctures two VC conceits: “take less money at a lower valuation, it’s less risk” ignores that 3x the money at the same dilution “derisks your life as a founder” (assuming you don’t burn it); and SAFEs have quietly normalized messy pricing — notes convert at “11,000 different prices… nobody cares, as long as Rory gets his ownership in the Series B.”

10. The window is open — Circle, the database land-grab, and three bets

• On the IPO/M&A wave (Chime, Groww in India, Shein moving from London to Hong Kong, Circle at $8bn, Salesforce-Informatica at $8bn), Rory’s organizing principle is a friend’s old line: “price clears all markets” — you can’t get out at the 25 you last paid, but “there is a price at which most decent companies can get public… it’s a degree of realism creeping in. It’s totally healthy. Normal IPOs are taking place; normal service has been resumed.” Circle he likes as “the boring version of crypto — crypto but safe”: a ~$44bn money-market-style float earning 4-5%, just under $2bn revenue shared with distribution partners, a couple hundred million of profit — interesting but “bounded in terms of value because there’s no magic explosion thing.”
• Jason’s pick: Snowflake buying nearly-pre-revenue Postgres shop Crunchy Data for ~$250M, sixty days after Databricks paid $1bn for Neon — “pretty interesting: within 60 days they both became database companies,” because their agents need to work efficiently across every data structure. Rory zooms out: “the AI plumbing business is a great place to be” — see ClickHouse’s explosive growth and a late-stage round around $6bn.
• Quickfire bet one — will the OpenAI Jony Ive device have a screen? Jason’s thesis: it’s not one device but everything — pendant, phone app, AirPods, ring, glasses, watch — “eventually it will be all of it… it’ll have audio, it’ll have screen.” Rory takes the under on 5M units in the first full year; Jason takes the over (it might be $50 — “they have enough money to subsidize it”), stakes settled in devices and dinner.
• Bet two, Meta shipping a non-open-source model this year: Rory says no absent DC intervention; Jason notes Llama is “already not fully open” and Zuck is “hyper-competitive… I wouldn’t bet against Zuck.” Bet three, Elon out as Tesla CEO before 2027 (market odds ~32%): Jason goes 50% — Musk thought about handing the role to JB Straubel once, “he’s got to find my Gwynne [Shotwell]… he can’t scale as a human being as CEO of Tesla and Neuralink and SpaceX… with the 58 kids and the 11 companies” — though “he didn’t even pick someone very good for Twitter.” Rory hedges: not the base case, “but you can never discount a rage quit,” plus the Delaware comp-package overhang and the brand upside of “distance between their customer base and a fairly controversial person.”