Is a $4.5BN Exit Enough in VC? & Harvey Raises $150M & Why Google is a Buy and Amazon is a Sell
Summary
- Navan’s IPO anchors mature software at roughly 6–7x NTM revenue when growth settles near 30%. Jason Lemkin framed Navan’s listing alongside Dev Ittycheria’s departure from MongoDB as the end of the SaaS 2.0 era. Navan survived COVID, reached $700 million-plus in revenue growing 32%, and was worth about $6 billion on IPO day before falling toward a $4.8–$4.9 billion market cap. For Jason, the message is brutal: new investments must be “Harvey or better.”
- An IPO headline is not cash, and a $4.5 billion exit can still disappoint a large venture fund. Typical lockups last six months, while full distribution may take 18–30 months; Lightspeed’s roughly $257 million position became about $1 billion, but that sub-4x blended return includes early dollars likely up 20x-plus and later dollars potentially underwater. “Locked-in value” 18 months after listing is more meaningful than day-one marks.
- Venture’s exit bar has risen while its time-to-liquidity has lengthened. Rory O’Driscoll’s rough model moves a seed journey from eight years and a 20% completion rate to 12 years and perhaps 10%, with IPO readiness now requiring roughly $400–$500 million of revenue. Concentrated funds therefore need believable $10 billion-plus outcomes; investors with more optionality can afford to “turn the next card.”
- Harvey’s $8 billion valuation works only if legal AI taps labor budgets, not merely the old software TAM. Reported metrics included $150 million ARR, 98% GRR and a 170% expansion figure; Jason inferred roughly $400 million forward ARR, or 20x. A future $24 billion value at 7x would require about $3 billion of revenue, making task automation and spend per lawyer the decisive underwriting questions.
- Founder-friendly fundraising has structurally compressed venture ownership. Chad Peets said his last three investments landed around 6–8% despite believing he needs double-digit stakes in two winners per fund. Harry cited a The Information report using MaC as an example of a 10% ownership outcome versus a customary 20% target, while Chad separately used Benchmark as an example of a premier firm getting only 10%. The inversion is clean: “A founder’s optimized fundraising is a VC’s below-ownership target.”
- OpenAI’s financing question is legitimate at both company and macro scale, regardless of Sam Altman’s dismissive response. Against roughly $12 billion of revenue, Brad Gerstner asked how plans requiring about $1.1 trillion could be funded; “sell your shares” supplied no answer. Harry extrapolated that, at a 50% gross margin, cumulative revenue may need to exceed $2.2 trillion, while Chad argued for rigorous board-level guardrails.
- Current cloud demand contradicts claims that the AI buildout is already breaking, but leadership has shifted. AWS reaccelerated to 20% growth while Google and Microsoft remained in the mid-to-high 30s; Chad called Google underappreciated and Amazon overappreciated because Google has models, TPUs, Search and applications. Meta’s dilemma is the reverse: a core business growing about 20% is funding roughly $70 billion a year of AI investment without an attached revenue engine.
- Incumbent software companies must capture some AI-driven reacceleration or accept mature-company multiples and eventual consolidation. Twilio reached 15% growth as voice-AI usage surged, while MongoDB moved from 13% to 24%; that can mean the difference between 6–7x revenue with a forward story and a 3x private-equity sale. Jason’s changed view is categorical: “Agents are better than mediocre humans,” and Frank Slootman gave the example of a $10,000-a-year agent replacing a $40,000 worker.
Deep dive
1. Navan made 30% growth look mature, not magical
Jason Lemkin read Navan’s listing alongside Dev Ittycheria’s departure from MongoDB as “the end of the SaaS 2.0 era.” A company with $700 million-plus of revenue and 32% growth was worth about $6 billion on IPO day, then traded toward $17 a share and a $4.8–$4.9 billion market cap.
Rory’s counterweight matters: Navan survived a near-death experience when COVID shut travel, investors financed the recovery, and management still delivered a multibillion-dollar public company. Two years from now, he argued, the debut volatility may be “all in the noise.”
The drop also challenged Bill Gurley’s claim that IPO allocations are “free money.” Buyers demand discounts on winners because they cannot know whether they are getting Figma’s first-day surge or Navan’s roughly 20% fall; from the issuer’s perspective, Navan may actually have priced at a temporary high.
2. Day-one IPO wealth can take 30 months to become cash
Rory put the standard lockup at six months and expected a position in a company like Navan to take at least 18 months to sell or distribute. His firm’s better performance measure was therefore the “locked-in value” 18 months after IPO, not the mark generated during the first trading session.
Jason learned a still more conservative rule: wait through the six-month lockup, then distribute ratably over 24 months. That implies roughly 30 months before most carry and LP proceeds arrive—potentially pushing Navan liquidity into 2028 or 2029.
About $200 million reportedly changed hands in the offering, including roughly $50 million for the founders, which Jason preferred to early-stage secondary sales. His reading, offered with uncertainty, was that most large holders retained their stakes while smaller investors supplied much of the secondary.
The Figma analogy captures the psychology: an investor might briefly be marked up $4 billion, ultimately realize a very successful $2 billion, and still feel disappointed. “The economic significance is limited” until the lockup, trading liquidity and distribution schedule have played out.
3. A $4.5 billion exit no longer guarantees venture-fund success
Lightspeed’s roughly $257 million investment was worth about $1 billion at listing, a blended return just under 4x; Oren Ze’ev reportedly invested about $150 million across vehicles and SPVs and ended up with roughly $1 billion. Early rounds may have generated 20–30x, but aggressive follow-ons diluted that multiple while putting far more dollars to work.
The later money carries price-compression risk rather than necessarily catastrophic loss. Rory believed investors from Navan’s last private round, around $9 billion, were down about 50% at the current mark; he nevertheless stressed that this is only today’s price, invoking Facebook’s initially weak listing and later appreciation.
Harry’s intentionally horrible question was whether “a $4.5 billion exit [is] good enough today.” It can return an early-stage fund, but for a $1.5–$2 billion vehicle it may represent only one-third of 1x after 12 years of work.
Jason and Rory separated the businesses: early funds seek one company that can return the vehicle; large late-stage funds expect many 3–5x outcomes, low loss ratios and perhaps 2–2.5x overall. The latter is “moving money at scale,” not a portfolio built around a small number of giant hits.
4. Seed underwriting now starts with a believable $10 billion outcome
Rory’s public-market anchor was 6–7x NTM revenue for a mature software or transaction business growing around 30%, “the 10-year Treasury equivalent of SaaS.” A startup growing 5x or 10x deserves different treatment today, but once it decelerates to 30%, “there’s no magic”: it converges toward the same multiple.
Jason had just invested at a $50 million post-money valuation and calculated that, after dilution, a 100x outcome would require the company to become substantially more valuable than Navan. Against OpenAI’s projection of more than $100 billion of revenue in 2027 and Anthropic’s cited $70 billion in 2028, he joked that he no longer wanted meetings with “mortal founders.”
Harry resisted rejecting companies simply because the $10 billion case is not visible on day one: value can accrue incrementally, and investors sometimes need to “turn the next card.” Jason conceded that optionality works when the initial check is small; when a first check consumes 4–5% of the fund, he has much less room for discovery.
Rory’s rough reset moves the seed journey from eight to 12 years and the share reaching the end from 20% to perhaps 10%, while IPO readiness may now require $400–$500 million of revenue. Clever but bounded markets lose the “magic pixie dust” of an IPO and leave only an M&A path.
5. Harvey’s valuation is ultimately a labor-budget wager
Harry relayed $150 million ARR, a 40% DAU/MAU ratio, 98% GRR and a 170% expansion figure for Harvey’s $150 million financing at an $8 billion valuation. Jason regarded daily usage as table stakes for a legal tool, but called the retention and expansion metrics outstanding.
Jason’s valuation calculator inferred approximately $400 million of forward ARR—not GAAP revenue—making the round about 20x. Raising nine figures for perhaps 1–2% was exceptionally attractive to existing shareholders because the effective dilution was tiny.
Chad’s terminal math was less forgiving: a 3x from the round means $24 billion; at a mature 7x multiple, that requires roughly $3 billion of revenue. With about one million US lawyers, approximately half in-house and half external, Harvey must become the clear category leader and command thousands of dollars per lawyer annually.
The key distinction is automating tasks rather than necessarily eliminating whole jobs. Legal software historically addressed a constrained software budget; Harvey’s larger opportunity appears only if demonstrable productivity converts some human-labor spend into software spend.
6. Optimized fundraising has smashed ownership rules
Harry cited a The Information report about Benchmark lowering its ownership requirement, using MaC as the example of a 10% outcome versus a customary 20% target. Chad separately said his last three deals delivered only 6–8%, even though his fund math says he needs double-digit ownership in two winners; refusing those allocations could mean refusing the best companies.
Chad saw two opposite paths to the same result. A capital-efficient breakout may sell only 10%, while a foundation-model company can consume billions yet still give a $100 million investor only a few points. Both extremes can produce exceptional returns, so traditional ownership heuristics have been “smashed to pieces.”
ICONIQ’s reported data offered a nuance: top companies may burn large absolute amounts while maintaining low burn multiples because revenue grows faster. That lets founders sequence smaller financings at rising valuations—the exact mechanism behind Chad’s inversion, “A founder’s optimized fundraising is a VC’s below-ownership target.”
YC has effectively productized low dilution through rounds such as “3 on 30,” “4 on 40” and “2.5 on 25,” alongside advice to sell roughly 10% around Demo Day and another 10% later at 3–5x the valuation. If angels already own 4%, a new lead may have only 5–6% available.
7. OpenAI’s trillion-dollar question cannot be answered with liquidity
Brad Gerstner’s question was straightforward: with about $12 billion of revenue, how will OpenAI finance close to $1 trillion of CapEx commitments over five years? Chad treated Sam Altman’s offer to find buyers for dissatisfied shareholders as a tired, snarky moment that revealed personality but nothing about the funding plan.
Chad said a founder had once given him a similar warning, and Harry argued that this kind of response represented an egregious escape from fiduciary duty. Chad agreed that “sell your shares” is not an acceptable board-level answer. Because the broader US economy now depends heavily on continued AI CapEx, the question requires serious scrutiny.
Harry inferred that the sharp reaction exposed some stress around honoring approximately $1.1 trillion of commitments. At a 50% gross margin, he reasoned that cumulative revenue may need to exceed $2.2 trillion, while Chad noted that the precise burden depends on timing and that annual revenue still needs to reach the many hundreds of billions.
Chad’s darker framing was that Altman has become the poster child for the entire buildout: if it unravels, history will want a villain. A board’s job is not to be “a rabbit,” but to test cash flows, identify which 2029–2030 commitments might be deferred and keep an optimistic trillion-dollar plan from becoming the defining story of an AI crash.
8. Cloud demand remains real, but AWS no longer owns the category
AWS reaccelerated from roughly 13% to 20% growth, while Google and Microsoft cloud remained in the mid-to-high 30s. Microsoft described itself as capacity-constrained; Amazon found capacity for a new OpenAI agreement. Chad’s present-tense conclusion was that anyone predicting collapse must argue it will happen later, because “if you have it, you can sell it” now.
Harry called the OpenAI announcement partly “AI performance theater”: Amazon looked like the fifth or sixth compute supplier after Microsoft, Oracle, Google and others. Chad’s answer was pragmatic—being late is still better than having no agreement—but both agreed AWS had surrendered its old dominance by adapting too slowly to AI-centric compute.
Amazon’s broader quarter included roughly 11% total growth and Harry’s cited estimate that shopping assistant Rufus contributed $10 billion of sales. Shopify complicated the victory lap by delivering 32% revenue and GMV growth, suggesting powerful commerce tailwinds were lifting both sides.
Chad preferred Google, calling it underappreciated relative to Amazon because it combines consumer AI, improving Search, TPUs, cloud and an application layer. Harry noted that the stock had already appreciated about 53%; Chad acknowledged that Google had assembled the assets needed to compete—even if its former search monopoly would plainly have preferred no disruption.
9. Meta is funding AI without an obvious place to sell it
Meta’s core business still grew about 20% and generated substantial cash, despite Harry’s cited $15 billion fine. The market’s objection was the commitment to roughly $70 billion of annual AI expenditure and still larger spending in 2026, which pushed the shares down by double digits.
Chad contrasted Meta with Google, Microsoft and Amazon, which can sell AI through enterprise platforms, and with ChatGPT, which has an AI-native consumer destination. Meta had neither an attached enterprise revenue stream nor a clearly incremental engagement product to justify the outlay.
Zuckerberg’s implicit response was founder control: “I refer you to the articles of incorporation. I control this company.” Chad could not yet see an efficient plan—particularly amid reported internal conflict—but acknowledged that Zuckerberg built Facebook and may be pursuing something public investors cannot yet see.
10. Mature software needs AI reacceleration, not an AI label
Twilio’s 15% revenue growth produced a roughly 15–20% stock bounce, yet Chad still saw a bounded $20 billion company trading near 4–5x revenue. The contrast is Palantir at the cited 123x: cash generation alone does not restore the “area of magic.”
Harry focused on the acceleration, not the ceiling. Twilio moved from single digits to 15% as voice AI grew 60% and its top 10 voice-AI startups expanded 10x; MongoDB moved from 13% to 24% over five quarters. Infrastructure companies including Datadog and Cloudflare were similarly attaching themselves to AI spend.
Chad accepted the amendment: even if an incumbent cannot become Harvey, moving from 15% to 25% growth can separate a 6–7x public multiple with a forward story from a 3x sale to private equity. “You better find a way to matter in this world.”
Jason admitted changing his mind after dismissing premature claims that agents would replace every worker: the products are now good enough that “agents are better than mediocre humans.” Frank’s concrete example was a $10,000-a-year agent replacing a $40,000 worker while producing a better result; the open question is how quickly that economics diffuses.
11. AI demand rewards speed, but finite markets and regulation still bite
Jason said SaaStr’s agent directory reached roughly 12,000 monthly views and sent millions of dollars of business to vendors such as Artisan and Qualified. He called demand for software that genuinely replaces labor “insatiable,” warning Salesforce and HubSpot to monetize products such as Agentforce in 2026; his harsh prescription was, “Fire yourself or half your team,” though the others endorsed the urgency more than the wording.
Harry cited OpenEvidence growing to $300,000 in one year, versus roughly 10 years for Doximity. Chad’s caution was pure TAM math: there are not suddenly more doctors, so professional AI markets equal the number of professionals multiplied by the work each product can automate. Individual adoption may happen in a year; corporate adoption could still take five or six.
That rapid pull makes Series A unusually attractive: YC, Neo and South Park Commons are supplying an enormous seed funnel, and the architectural direction—AI performing more enterprise work—is clearer than at the end of SaaS. The cost is tougher competition and compressed decision time; Chad cited Andreessen Horowitz as his most recent loss, while Harry named Chamath Palihapitiya.
On Kalshi at $5 billion versus Polymarket at $9 billion, Harry raised Kalshi’s possible nationwide-compliance advantage conditionally, while Chad preferred Kalshi if its regulatory position was genuinely safer. Chad thought the larger threat would come from sports leagues confronting proposition-bet cheating and insider influence—prediction markets let someone “put their finger on the scales”—but his simplest investment conclusion was: “I wish I was in one of them.”