Summary

Byron Deeter’s biggest change of mind in twelve months is the headline: “very sincerely we’ve probably added a zero to everything.” Bessemer has been a buyer of Anthropic — reportedly raising at $170 billion — and when Harry asks over/under on a $1 trillion Anthropic within three years, the answer is “over”: a year ago “we thought that where they’re at now would be the exit, not another entry point.”
On margins, Deeter cares about “the margin profile of the future,” not today’s P&L — Snowflake had negative gross margins very late in its life cycle. Foundation models may be commodities, but “the best business in the history of software is sitting there with AWS in what people refer to as a commodity,” and model releases should be read as products: each atomic unit profitable, making it “economically imprudent not to forward invest.”
T2D3 is no longer enough. Bessemer’s State of AI report documents “supernovas” going zero to $100M in 1.5 years — Dario described Anthropic going zero, $10M, over $100M, over $1B in successive years, with Dario saying there’s a chance to cross $10B next year. The old zero-to-$10M-in-18-months term-sheet trigger is “cute now, but it’s off by an order of magnitude.” Efficiency still matters: the Rule of X shows growth worth a 2–2.5x multiplier over efficiency at ~$50M ARR.
Vertical SaaS isn’t dead — AI is its payments moment, the TAM-doubling unlock that Toast and Shopify got from payments. But the honest concession: unlike cloud one, this wave has no business-model or delivery dislocation, so incumbents keep platform, data, and distribution advantages — “that is a reason to be scared” for challengers, even if high-execution challengers still win.
Whether AI moves from the technology budget into the labor budget is “already being answered. It’s not even a debate anymore, Harry. It’s over.” Proof point: a portfolio company swapped human-based interventions for Intercom’s Finn and got 90% automated deflection with NPS up. Harry’s pushback — 23-to-30-year-olds are “about to get hit with a train” — gets the Bessemer-steel-process history lesson in reply, not a concession.
Liquidity is the tradeable macro call: Bessemer’s Cloud 100 spoiler shows over $1 trillion in private market cap in the top 100 cloud/AI companies — “entirely legitimate… tradable public entities in a private wrapper today.” Deeter confesses a change of mind on secondaries: the stigma is wrong, LPs deserve DPI, and the next liquidity wave comes from four buckets — PE, big M&A (SAP, Oracle, IBM “need to buy or they’re going to get crushed”), a reopening IPO window, and secondaries.
On the future of venture, he rejects Harry’s Chanel-vs-Walmart frame: the right analogy is investment banking — Goldman-style full-service platforms (Bessemer: nine offices, tens of billions in AUM) versus boutiques like Benchmark, with the middle getting squeezed. And no fixed sector teams: “you bet your ass they’re going to invest 500 million in semi whether the right answer was 2 billion or zero.”

Deep dive

1. “We’ve probably added a zero to everything”

Asked what he’s changed his mind on most in the last twelve months, Deeter doesn’t hedge: “I thought we understood this next phase we were going into and how big this was going to be and very sincerely we’ve probably added a zero to everything. I think there’s going to be a lot of trillion dollar businesses that are created from this.” Bessemer’s own published memos are the evidence — upside scenarios that treated $1B outcomes as the great success case, for companies that became $10B and $100B businesses.
Harry’s unfair closer — Anthropic over or under $1 trillion within three years — gets a position, not a dodge: “We put our money where our mouth is. We’ve been a buyer. So, over.” The kicker: “A year ago when we were buyers… we thought that where they’re at now would be the exit, not another entry point.”
The flip side he keeps live: “the stakes are way higher than they’ve ever been” and these businesses “in some cases could still go to zero” — hyper power-law outcomes that change the nature of the game. Harry’s confession lands the theme: when Vince at Thrive led the $30B round saying “if it’s a trillion dollar company, we’ll make money,” Harry called it “the weakest investment rationale I’ve ever heard — and now I feel like a total [expletive].”

2. Today’s margins are the wrong question — the future margin profile is the right one

Deeter rewords Harry’s margin question deliberately: he cares “a lot about margins on investments we make today — but the margin profile of the future.” Snowflake ran negative gross margins “very late in their life cycle”; Stripe, Twilio and Shopify all crossed capital-intensive investment horizons. None of Bessemer’s investments are cash-flow based short term, and “very few of them are gross margin based.”
On commoditization he refuses the derogatory reading: sure, foundation models may be commodities the way hyperscalers are — but “the best business in the history of software is sitting there with AWS in what people refer to as a commodity.” Same playbook, and the layers on top “are going to extract phenomenal value because they’re going to deliver phenomenal value.”
His frame for upside-down LLM P&Ls: treat each model release as a product with its own life cycle — “you’re monetizing last year’s training in this year’s revenue line while you’re investing in next year’s model.” Each atomic unit may be highly successful and profitable even when the consolidated P&L looks inverted, so “it would be economically imprudent not to forward invest” when demand and unit economics are proven at each level.

3. The ownership rulebook is changing — small stakes in generational companies

Bessemer now happily takes positions “well below historical venture standards where you aspire to own 20%”: nine figures each into Anthropic, Perplexity and Canva. Harry’s dilution math bites — a friend entered Anthropic around 4, and by the 60 round held only a 3.8x. Deeter’s answer: “3.8x will keep you in business for a long time, but the reason to do it is because you believe it could be a 30x” — with Anthropic reportedly now raising at 170.
The concentration stats behind Jason Lemkin’s question — 40% of venture funding into ten deals — don’t scare him: the top three LLM companies (Anthropic, OpenAI, X) “are going to raise a hundred billion dollars in this six-month period,” yet hundreds of compelling businesses will be built in and around those ecosystems, with “a lot of 10xs and many hundred-Xs.”
Scale perspective, told against himself: he sold his company in 2005 for hundreds of millions, “the top outcome for our software cohort in that vintage… now that’s a seed round for some of these businesses.”

4. Vertical SaaS isn’t dead — AI is its payments moment

Harry admits to feeling “like an old man… looking for the next ServiceTitan” while everyone chases AI. Deeter’s answer: legitimately debatable, but no — and “alpha comes from not only being right but contrarian.” Data models, supply-chain connectivity, and marketplace capabilities matter more than ever. The historical template: payments doubled the TAM and market caps for Shopify, ServiceTitan, and Toast — “I think AI is going to do the same thing,” via ServiceTitan’s technician copilot, MaintainX on the factory floor, Luxury Presence in real estate. “In many ways I love that it’s not as sexy right now.”
The confession worth the episode: “we’re in the challenger business, and AI gives the incumbents advantages that didn’t happen in cloud one.” Cloud one had a business-model dislocation (license→subscription) and a delivery dislocation (on-prem→multi-tenant); this wave is “really the next horizon of cloud” — so incumbents keep platform, data, and massive distribution advantages. “The fast-moving incumbents are absolutely going to make a run… which hurts the challengers, and that is a reason to be scared.” He still bets high-execution challengers win over time.
His conversion story on AI-vs-human service: a portfolio board showed deflections up to ~90% automated on Intercom’s Finn — and NPS went up. He objected that a robot beating a human “makes no sense,” and the answer was faster, more comprehensive responses with links and references: “a better experience… a win-win-win on so many levels.” Customer support is “one of these killer use cases that’s just starting, but it’s going to roll through so many other areas.”

5. The labor-budget debate “is over” — but Harry won’t buy the utopia

Harry channels Rory’s framing — the whole game is whether AI moves from the technology budget into the multi-trillion-dollar market. Deeter flattens it: “That question is already being answered. It’s not even a debate anymore, Harry. It’s over.” Tech is now addressing software, hardware, and services budgets comprehensively — Bessemer is going down accounting, legal, and medical, and a likely Abridge example is “freeing doctors up in their patient interactions.”
On Epic’s fightback (offering transcription): “the game is on.” Epic has enjoyed a “state-endorsed monopoly” and holds the patient-data treasure trove that needs unlocking — he expects “a thousand flowers to bloom” in medical AI, citing Dario’s Machines of Loving Grace claim that “a hundred years of medical research is about to be pulled forward in the next decade.”
The workforce numbers on the table: Shopify grew revenue 91% with a 30% workforce reduction. Deeter pushes back only on “ruthless” — leaders are giving people leverage, and he predicts “the era of the micro business… 10-person companies crossing billion-dollar valuations.” Harry’s counter is the sharpest disagreement of the episode: “It’s a reduction in force, Byron… these 23 to 30 year olds are about to get hit with a train. Do you disagree?”
Deeter’s reply is the firm’s own namesake: newspaper clippings a hundred years ago predicted workforce catastrophe from the Bessemer steel process — then skyscrapers and railroads; ditto what he thinks was 4% of the female workforce running switchboards. “It’s coming, definitely — and at the same time more opportunity is going to be created.” He explicitly rejects the full-abundance utopia too: “I don’t go that far, but I do think we’re going to get some hours back in the day.”

6. T2D3 is no longer enough — “cute now, but it’s off by an order of magnitude”

Has Bessemer misled founders with triple-triple-double-double? “Unfortunately yes.” The new State of AI report maps galaxies: “supernovas” going zero to $100M in 1.5 years, and Dario described Anthropic going “zero, 10 million, over 100 million, over a billion the next year,” with Dario openly saying “there’s a chance to cross 10 billion in the next year.”
The old reflex — zero to $10M in 18 months meant “go deliver the term sheet with a dog and a golden” — gets Harry’s verdict: “that’s cute now, but it’s off by an order of magnitude.” The fatter part of the curve is “shooting stars”: zero to $100M over four years, a quadruple-quadruple cycle, versus the old seven-year centaur chart.
Harry’s trap — so founders should take max money and ignore margin? — gets a flat no. The Rule of X quantifies the trade-off: growth carries “about a 2 to 2.5x multiplier value over efficiency” at mid-stage (~$50M ARR), and “we do still believe that all businesses should be valued as a sum of their future free cash flows.” Many of these AI businesses are scaling in “pretty capital efficient ways.”

7. Scaling, breakthroughs, and sales and marketing matter less

On GPT5 as the start of incrementalism: “It’s going to be fits and spurts” — he concedes “people being underwhelmed with the GPT5 release” — but “the curve is still up into the right without a doubt,” and higher-level reasoning “that does mirror the world’s smartest scientists… is coming in the next 18 months.” Also: “it’s not just an Nvidia world anymore” — Amazon, Google, and AMD chipsets are “becoming quite capable,” and “these scaling laws continue to hold.”
Harry relays Rory’s competitive playbook — “enter a space and just scream the freaking loudest… suck all the air out of the VC room and then deliver,” Harvey as exhibit A. Deeter dissents: “ChatGPT didn’t scream from the mountaintops” — great products get pulled through, and Perplexity is doing the same in answer engines.
The structural claim underneath: “marketing and sales have less of a role in this new economy.” Mark Leslie’s sales learning curve, from the Veritas days, “is no longer applicable because you just can’t possibly throw the bodies at a 0-to-100 growth curve inside of two years.”

8. A trillion dollars in private wrappers — and the liquidity that has to find it

Cloud 100 spoiler: “over a trillion dollars in private market cap among just the top 100 cloud AI companies.” Harry: how much is legitimate versus synthetic hype? “Entirely legitimate… those are incredibly high quality companies that essentially are tradable public entities in a private wrapper today” — though 101–300 may hold “some walking wounded.” Even “Mr. Irrelevant,” number 100 in the NFL-draft analogy, “is an awesome company.”
His confessed change of mind: the stigma on GPs selling — signal risk, “a dirty word” — “is wrong.” His first IPO, Cornerstone OnDemand, went out at $50M ARR and ~$700M market cap; now Canva sits at $40B+, Anthropic $70B+, and “there’s an argument that they should be handed off to later-stage investors and hedge funds.” LPs “deserve” DPI, and for emerging funds secondaries are “economic necessity in a healthy ecosystem.” (On Sequoia’s evergreen structure: merits, but “public management’s a different beast” and LPs get paid to allocate.)
On the IPO reopening after Figma, CoreWeave, Circle, and Bullish popped: “God, I hope so, Harry” — a healthier IPO market late this year and into next, though he’s “not convinced” public markets return to premium multiples. Canva “should and I think they will” go public — great CFO hired in Kelly, hundreds of millions invested, “one of our largest investments in our firm history” — but “the urgency level there is moderate.”
The next liquidity wave comes in four buckets: PE (these businesses are high gross margin, “run pretty inefficiently”), big M&A returning as the FTC rationalizes — “companies like SAP and Oracle and IBM need to buy or they’re going to get crushed” — IPOs, and the wild card of secondaries. On Tiger: “incredibly mixed… a lot of happy disruption and big thinking” alongside “reckless overfunding and non-governance that we’re going to have to clean up for quite a while.” Harry’s contrarian take: top of the pref stack in $50–60M companies at $300–400M prices, “they’re not going to lose money on those deals.”

9. Bessemer’s craft — paying up, doubling down, and the anti-portfolio

Discipline, reframed: “we’re certainly not value investors. We pay market clearing prices” — but Bessemer walks when unit economics don’t pencil. The famous scar: Tesla, on his personal anti-portfolio — he couldn’t see roadster unit economics working, “and to be clear, they didn’t”; without the DOE loan Tesla wouldn’t have existed. “What I missed was that Elon’s a force of nature… and that’s the challenge I put back on myself: how to break this notion of short-term discipline for the long-term horizon.”
The double-down problem is psychological: sitting on a 10x mark and having to “reset everything and put in 200 million that now I need to dig out and get a return on again… it’s intimidating.” Bessemer built process against the inertia — the “century team” re-underwrites the best deals fresh, asking whether another 10x is ahead, “to break that mental trap of being comfortable with success and being afraid to really back up the truck.” Harry’s parallel confession: refusing to pay 5x four months after a seed “is a very dangerous mind.”
The costliest pattern: TAM misestimation. Bessemer owned 28% of Shopify and Twilio at IPO but far less of Procore and ServiceTitan — “we underestimated TAM and weren’t sure these could be $50 billion businesses because we didn’t yet unlock the payments expander… and got weak-kneed investing into the billions.” Harry’s gloss: market-size misreading “is the single greatest reason why great investments are not made.” On Shopify, a distinction: “we distributed early. We didn’t necessarily sell” — a lot of LPs sold by mandate, and money was left on the table.
Scenario analysis is required in every investment memo, and Bessemer publishes them — “it’s kind of embarrassing… why is it that every memo I’m reading solves to a 3x?” The real tiebreaker is the partner “pounding the table that the high end and more is possible.” Against Lemkin’s rule (world-class founder + visible 3x, skip the TAM math), Deeter holds out for “exciting adjacencies” — “the three-dimensional cube of segments and products and users that can flex over time… playing in a big enough pond.”

10. The future of the firm: banking-style bifurcation, no fixed sector air cover, and surviving the trough

Chanel vs Walmart? “Do I agree with the Walmart analogy? No. But do I agree that scale is important? Yes.” The better map is investment banking: Goldman/Morgan/JPMorgan-style full-service platforms — Bessemer’s path, with nine offices, tens of billions in AUM, multi-stage — alongside boutiques like Benchmark. A bimodal curve where “this maturation is going to make it tough in the middle.”
On thematic funds he mostly sides with Harry, with a governance edge: “we don’t hire sector-specific investors and we don’t give you air cover if your sector goes out of favor… If you carve out 500 million for semiconductors, you bet your ass they’re going to invest 500 million in semi whether the right answer was 2 billion or zero.” Instead: partners pitch “road maps” at every offsite, competing for the incremental dollar across sector, stage, and geography — “if you don’t constantly reinvent yourself, you don’t have a future.”
His own trough, told straight: first roadmap was RFID — “a zero trillion dollar market… a total dud” — and “my first three investments were all very bad. My next two ended up being billion-dollar IPOs” (Cornerstone OnDemand and likely Eloqua). Senior partner Phil Hardman preached patience — “don’t shoot out of the gun crazy big checks so if you go 0 for 3, you’re done” — and Robin Vasan’s warning proved doubly right: “your first investment is going to suck. It always does… but this one really sucks.” The systemic lesson: investors who got checkbooks in 2019–20 bought at peak prices and “got washed out… we may never know if they could have been great.” Per the cited data, roughly eight of the top ten firms repeat in the next cycle.
Quick-fire signal: best sourcer is [likely Jeremy Levine] (“the insights to see what others don’t”); best picker David Cowan — “finds great people in weird places doing bizarre things,” Rocket Lab’s Peter Beck in New Zealand among them. His LP picks — likely Pete Sonsini’s seed-stage pick, likely Law Ventures, with Perplexity and Databricks among its examples; First Round/Uncork early; Meritech and Sequoia at growth — draw Harry’s needle: “none of the new guard… you didn’t mention Sarah Guo.” Deeter’s reply: “I think quantitatively history is a positive” — those firms are getting stronger through generational transition, and the upstarts are “less proven.”

Open AI Insider on GPT-5, AGI & the Great AI Race w/ Kevin Weil & Dave Blundin Guinea Value's Jingshu Zhang on $EDU