Summary

Navan’s debut anchors mature, roughly 30%-growth software or transaction businesses at about 6–7× NTM revenue, rather than hypergrowth AI-company multiples. At $700 million-plus revenue and 32% growth, its market cap fell from roughly $6 billion on IPO day toward $4.8–$4.9 billion. Jason Lemkin called it “the end of the SaaS 2.0 era,” while Rory O’Driscoll stressed that surviving COVID and reaching the public market remains an excellent outcome.
An IPO headline valuation is neither cash in the bank nor proof that allocations are “free money.” Navan priced in the middle of its range, then traded near $17, demonstrating Rory’s counter to Bill Gurley: investors demand upside on winners because occasional 20% drops compensate for it. The normal six-month lockup and gradual selling mean the better measure of realized value is often the market cap 18 months later—or roughly 30 months under Jason’s ratable-distribution model.
The venture exit bar has risen enough to make a $4.5 billion outcome genuinely debatable for large funds. Jason’s $50 million post-money seed investment must become “way better than Navan” to approach 100× after dilution, while Rory said investors may now need roughly $400–$500 million of revenue for an IPO. A seed journey has stretched from perhaps eight years with 20% reaching the end to 12 years with only 10%, forcing concentrated funds to seek believable $10 billion outcomes while larger platforms buy optionality across rounds.
Harvey’s $8 billion valuation is defensible on growth but ultimately rests on whether legal AI becomes a $3 billion revenue market. At $150 million ARR, 98% GRR, 170% NRR and a stated $400 million forward ARR forecast, the round prices Harvey near 20× forward revenue while raising $150 million for minimal dilution. To reach $24 billion at a mature 7× multiple, it must redirect meaningful lawyer-labor spending into software: “It’s not about automating people, it’s about automating tasks.”
OpenAI’s trillion-dollar infrastructure ambition demands a substantive financing answer, not “sell your shares.” David Sacks and Harry treated Brad Gerstner’s question—how roughly $12 billion of current revenue supports about $1–$1.2 trillion of commitments—as legitimate; Harry said it was a board-level issue affecting “the health of the entire US economy.” The speakers’ rough math requires eventual annual revenue in the many hundreds of billions; Harry warned that if the plan breaks, Sam Altman risks becoming the face of the unwind.
Jason’s public-market relative-value call is Google over Amazon, while Meta’s AI spending remains unproven despite a strong core business. AWS reaccelerated from roughly 13% to 20%, but Google and Microsoft cloud remained in the mid-to-high 30s; Jason called Amazon “overappreciated” and Google “underappreciated” because Google has models, TPUs, search, applications and monetization. Meta’s core grew about 20%, yet roughly $70 billion of annual AI spending lacks either an enterprise sales channel or an obvious AI-native consumer payoff.
Every pre-GPT software company must convert AI spending into measurable reacceleration heading into 2026. Twilio moved from single-digit growth to 15%, with voice AI up 60% and its top 10 voice-AI startups up 10×; MongoDB recovered from 13% to 24%. Jason’s uncompromising test was: “Where’s your agent? Where’s your re-acceleration?”—because 6–7× revenue with a forward story is dramatically better than a 3× sale to private equity.
The durable AI opportunity is replacing expensive tasks, but rapid adoption can also exhaust a finite market sooner. David Friedberg endorsed Jason’s lived example and cited cases where a $10,000-a-year agent can outperform a $40,000 worker, while Jason declared, “The age of the co-pilot is behind us.” Harry said OpenEvidence grew to $300,000 in one year versus the 10 years Doximity took, but investors must still calculate professionals × automatable work and ensure there is expansion after the initial adoption wave.

Deep dive

1. Navan’s IPO closes the SaaS 2.0 chapter

Jason paired Navan’s stumble with Dev stepping down from MongoDB: one company returned from 13% growth into the 20s as its CEO handed over, while another reached public markets at $700 million-plus revenue and 32% growth yet struggled around a $5 billion valuation. Together they felt like “the very end of an era.”
Rory’s wider lens was less mournful. Navan survived a near-death experience when COVID stopped travel, investors financed it through the crisis, and CEO Ariel Cohen kept operating; a roughly $4.8–$4.9 billion market cap therefore remains “a great outcome.” Rory said the short-term movement should be noise two years from now.
The offering priced in the middle of its range, slipped on day one and fell harder around day three, reaching roughly $17 per share. Rory used that outcome against Bill Gurley’s “free money” critique: IPO buyers want a discount on Figma-like winners because “every once in a while…this shit goes wrong, and the stock goes down.”
Harry’s pushback on media scorekeeping matters: Oren Zeev’s reported $150 million-to-$1 billion result, Lightspeed’s $257 million-to-$1 billion stake and Andreessen’s $635 million position were marked-to-market values, not immediately spendable proceeds.

2. IPO wealth becomes real slowly—and often at another price

Rory described six months as the typical minimum lockup, followed by a sale or distribution process that can take at least another year. His former “locked-in value” convention valued an exit at the company’s market capitalization 18 months after IPO, a much closer estimate of what investors actually realized than the first-day print.
Jason learned a still slower base case: distribute shares ratably for 24 months after the six-month lockup. Large holders cannot sell everything at once, so LP distributions and carry may not substantially arrive until 30 months after the listing—potentially well into 2028 or 2029.
Navan included almost $200 million of secondary sales, although Jason said he did not think the major institutions sold any, “as near as I can tell.” He approved of the founders taking roughly $50 million at IPO: “I’d rather see them take fifty in the IPO than after Demo Day.”

3. Mature growth has a 6–7× anchor again

Rory’s underwriting baseline is now explicit: a mature SaaS, transaction or comparable decent-margin company growing around 30% is worth approximately 6–7× NTM revenue. He called that “the 10-year Treasury equivalent of SaaS”—the multiple portfolios should use once extraordinary growth normalizes.
That comp does not directly price an AI company growing 5× from $50 million or 10× from $10 million. But Rory’s endpoint is unforgiving: once its growth decelerates to 30%, it will probably trade around the same 7× multiple as other 30%-growth businesses. “There’s no magic there.”
Jason translated that endpoint into seed economics. He had just invested at a $50 million post-money valuation; to make 100× after dilution, the company must become “way better than Navan.” His deliberately abrasive filter—“I don’t even wanna take meetings with mortal founders”—expressed the difficulty of believing every expensive seed can exit north of $10 billion.
Harry resisted rejecting companies before “turning the next card”: value can accrue incrementally, and a future $10 billion company rarely looks inevitable on day one. Jason conceded that diversified funds with small first checks and follow-ons can preserve that optionality; his own 4–5%-of-fund initial positions leave far less room for error.

4. A $4.5 billion exit serves different fund businesses differently

Rory said investors may now have to assume roughly $400–$500 million of revenue as the threshold for an IPO. In his stylized comparison, seed investing has shifted from an eight-year journey where perhaps 20% reach the end to a 12-year journey where perhaps 10% do, raising the required market size and eliminating “clever little small markets” that offer only M&A exits.
Navan also illustrates how firms dilute spectacular early-round multiples by following winners. Rory said he would be willing to bet the earliest dollars produced 20–30× returns, while Lightspeed’s total $257 million produced just under 4×; he believed the late private round at a $9 billion valuation was roughly 50% underwater at the episode’s snapshot, while stressing that this was only a point in time.
Harry argued that a $4.5 billion exit cannot be enough for a $1.5–$2 billion fund if it represents only one-third of the fund. Jason voiced the resulting frustration—“I spent 12 years with these amazing founders… and I only got a third of the way to 1X?” Rory’s answer was that this is the wrong expectation for a growth vehicle: a diversified late-stage strategy seeks many 3–5× wins, low losses and perhaps 2–2.5× net—not one investment returning the fund.
The outcome can simultaneously be excellent for founders, fund-making for early vehicles and merely one successful deployment inside a multibillion-dollar growth platform.

5. Harvey’s valuation is a wager on legal labor becoming software spend

Harry supplied the operating snapshot behind Harvey’s $150 million raise at an $8 billion valuation: $150 million ARR, 40% DAU/MAU, 98% GRR and 170% NRR. Rory called the daily usage table stakes for a legal workflow, while highlighting the exceptional retention and expansion; against the company’s stated $400 million forward ARR forecast, the price is about 20×.
Jason loves “getting nine figures for 1 or 2% of the company” because the dilution is almost immaterial to existing holders, though the price raises the eventual exit bar. The round is the inverse image of Navan: investors are paying for extraordinary forward growth, not applying mature-company multiples to present revenue.
Rory credited Harvey with quickly establishing brand and Am Law presence in a field where software historically produced few major outcomes. LLMs fit legal work because both manipulate language; Harvey leads, with Lago clearly second, but market leadership alone does not settle the valuation.
A 3× return implies a $24 billion company; at a mature 7× multiple, that requires roughly $3 billion of revenue. With about one million US lawyers, split approximately between in-house and external practice, the underwriting question is whether task automation can command thousands of dollars per lawyer and shift enough human-labor budget into software.

6. Founder-efficient fundraising structurally compresses VC ownership

Jason historically believed he needed double-digit ownership in two winners per fund, yet his latest three investments landed around 6–8%. In a hot round selling only 10%, his founder-friendly limit is to request the largest allocation and invest every available dollar; demanding “me or nothing” once backfired and is not his style.
Rory agreed ownership is falling across stages: his target is about 10–11% in late Series A or B, while even Benchmark reportedly obtained only 10% of Mercor rather than its traditional 20%. “What are you gonna do?” If a great company needs to sell only 10%, refusing to participate can be the dumbest choice.
Low ownership comes from opposite capital profiles. An efficient company can raise little and retain leverage; a foundation-model company may need billions, yet a $100 million check still buys only a few points. Rory’s real-time conclusion was that old rules were “smashed to pieces”: either extreme can still generate exceptional returns.
Fast growth and low burn multiples let founders sequence rounds—perhaps selling 10%, then 5%, then raising $1.2 billion at an $8 billion valuation. Rory supplied the inverse: “A founder’s optimized fundraising is a VC’s below-ownership target.” YC has institutionalized it through rounds such as $3 million on $30 million or $4 million on $40 million, leaving single-digit allocations for outside VCs.

7. OpenAI’s capital plan is too consequential for a glib answer

David Sacks and Harry called Brad Gerstner’s question entirely legitimate: how does roughly $12 billion of revenue finance about $1 trillion of CapEx over five years? Sam Altman’s response—effectively, sell your shares and he would find a buyer—revealed “a person in a bad moment,” but nothing about the funding mechanism.
Jason remembered receiving that response from a founder once and never criticizing the company again. Harry argued that “fuck off and sell your shares” is not an acceptable answer when the company’s capital plan is at issue and, as he put it, the AI infrastructure boom affects “the health of the entire US economy.”
Harry had seen respected board members refuse to intervene at a failing company because challenging its founder might hurt their reputation. David agreed and called it a disgrace. Harry said boards can become “grin fuckers” around successful founders, when their actual job is to provide experienced guardrails rather than act as cheerleaders.
Harry felt some sympathy for Altman. If the CapEx cycle breaks, the market will quickly choose its poster child; trillion-dollar forecasting errors become economic history rather than forgotten startup mistakes. Jason’s rough 50%-margin math implied more than $1.1 trillion of cumulative revenue to cover $1.1 trillion of commitments, while he described the annual requirement as many hundreds of billions of dollars.

8. AWS reaccelerated, but it no longer owns the compute narrative

AWS growth rose from roughly 13% to 20%, supporting the market’s conclusion that Amazon remains relevant in AI compute. Microsoft and Google were still growing cloud in the mid-to-high 30s. The episode’s evidence pointed to continued demand: available capacity sells, and compute suppliers benefit while that demand remains strong.
Jason discounted Amazon’s OpenAI agreement as “AI performance theater”—effectively finding GPUs after Microsoft, Oracle, Google and others had already committed vastly larger capacity. Rory agreed it was smaller, but argued it was “better than not having one”; the stock’s reaction came primarily from actual AWS reacceleration, not the press release.
The deeper concession was harsher: AWS created and dominated cloud computing, then failed to evolve quickly enough as the category became AI-centric. Shopify’s simultaneous 32% revenue and GMV growth also showed that Amazon’s commerce strength benefits from broad tailwinds rather than uniquely superior execution.
Jason’s relative call was “Google is underappreciated, and Amazon is overappreciated.” Google now combines consumer AI, recovering search growth, TPUs, cloud and a broad application layer; Rory noted that Google was up about 53% from early-year pessimism, but agreed it had assembled nearly every component required to compete.

9. Meta’s cash machine does not yet explain its AI bill

Meta’s core business remained excellent, growing roughly 20% and producing substantial cash despite Harry’s concerns about a $15 billion fine. The market’s objection was the destination of that cash: approximately $70 billion a year going into AI infrastructure without an attached revenue stream.
Rory contrasted Meta with Google, Microsoft and Amazon, which can sell AI through enterprise products, and with ChatGPT, which already has an obvious AI-native consumer application. Meta had neither clear route, leaving investors to ask, “What the hell?” even while its advertising engine performs.
Zuckerberg’s practical answer is founder control: he believes strategic relevance requires the bet and can refer dissenters to the company’s articles of incorporation. Rory saw the selloff as rational, not catastrophic—the market remembers Meta’s 2021–22 spending cycle and is pricing the possibility that another enormous commitment may not work.

10. AI reacceleration separates durable incumbents from PE fodder

Rory initially framed Twilio as a bounded mature company: about $20 billion of market value, $4–$5 billion of revenue, 15% growth and a 4–5× multiple. It can produce cash and rally on execution, but it is not playing Palantir’s game at a cited 123× revenue.
Jason focused on the change in slope. Twilio reaccelerated from single digits to 15%, voice AI grew 60%, and its top 10 voice-AI startups grew 10×; MongoDB moved from 13% growth five quarters earlier to 24%. Those are meaningful gains for large incumbents, not cosmetic AI branding.
Rory adopted that distinction: enough “AI pixie dust” to move growth from 15% toward 25% can support 6–7× revenue and a forward narrative. Missing the spend entirely can lead to a 3× private-equity sale where the business is “smushed” into another asset and disappears.
Jason’s 2026 test is unforgiving: HubSpot and Salesforce have shipped products but must now show growth. Agentforce is “quite good,” with roughly 2,000 people working on it; if it produces no meaningful bump by the middle of next year, Jason said he would fire half the team. “Where’s your agent? Where’s your re-acceleration?”

11. The agent opportunity is now labor replacement, not assistance

Jason changed his mind after seeing real operating data. Earlier claims that every employee would become an agent sounded like venture chatter because the software was not good enough; now, in selected workflows, “agents are better than mediocre humans.” His conclusion: “The age of the co-pilot is behind us.”
David Friedberg believed Jason’s conviction because Jason had directly moved work from people to machines and seen the result. David supplied the concrete economics: in some areas, a $10,000-a-year agent can replace a $40,000 worker and produce a better outcome. The unresolved “million-dollar question” is how quickly that capability diffuses through the economy.
Harvey may partially replace associates who do not want to grind through an IPO prospectus; next year’s product should automate still more. Jason’s broader warning to incumbent platforms was simple: if customers cannot buy the automation from you, “they’re gonna buy it from somebody else.”
Jason’s agent directory reached about 12,000 monthly views without promotion and sent millions of dollars of deals to Artisan and Qualified within months. For products that replace workers “for real, not for pretend,” demand is so intense that vendors cannot onboard all interested customers.

12. Fast adoption sharpens both Series A opportunity and market risk

Harry said OpenEvidence grew to $300,000 in one year, versus about ten years for Doximity. The speakers saw enormous latent demand but added the constraint: there are not suddenly more doctors. Investors must model professionals × automatable work and demand follow-on products after the first user wave saturates.
Individual adoption can happen in a year, as with ChatGPT, Lovable or physicians using research tools; corporate deployment may still take five or six years. Jason therefore advised slower companies to target slower-moving sectors such as retail or manufacturing, where only a small percentage of customers may yet be fully ready.
Jason rejected the claim that Series A has never been harder. Seed supply from YC, Neo, South Park Commons and other accelerators makes its top-of-funnel “the best it’s ever been”; Rory added that the ten-year architectural direction—agentic software reworking enterprise tasks—is clearer than at the exhausted end of SaaS, though capital and competition make winning harder.
In the closing prediction-market call, Rory preferred Kalshi as a user but wished he owned either platform; Jason chose Kalshi only if its all-state compliance advantage is real. Rory saw league integrity, insider knowledge and manipulable proposition bets—not necessarily a new administration—as the likelier constraint when markets can wager on something as specific as a quarterback’s third-quarter miss.

Massive Selloff, Mamdani Wins, & Jonah Ditches His CryptoPunk AI, Attention, and Ownership: Ari Emanuel Explains the Next Era of Entertainment