Summary

Burn multiple remains useful, but AI has broken the comparability that once made it feel definitive. ICONIQ’s sub-$100M ARR cohort shows AI-native businesses at -126% free-cash-flow margins versus -56% for non-AI, yet their extraordinary growth can still produce better burn multiples. Rory O’Driscoll’s warning: reported ARR, hidden churn, changing gross margins, CapEx and cash balances can invalidate the ratio—“you could have a great burn multiple and you still could be out of cash on Friday.”
For merely good software companies, fundraising is now an availability decision rather than a valuation-optimization exercise. Rory says VCs price deals “on hope” or “on the multiples”; a $15M-revenue company with reasonable growth can be “of zero value to a VC” when it lacks a credible path to a major IPO. Jason Lemkin’s advice to a $15M ARR company with good burn and a limited TAM is to take an offer at 250 rather than wait for a better price—triple-triple-double-double is no longer a funding guarantee.
Kingmaker financing is powerful because the first prestigious check attracts a wall of follow-on money. Elad Gil argued that proximity to a Valley kingmaker can deter funding for challengers; Harry said it should be treated as a factor, not a binary veto. A company that raises $20M from a top firm may quickly pull in another $60M, forcing challengers to prove they can beat not merely a competitor but “the wall of money.” Jason’s operating rule is simpler: “If you’re not number one, don’t spend like you’re number one.”
AI’s technological direction can be right while venture entry prices and public SaaS benchmarks are dangerously generous. Either AI drives a profound transfer from labor budgets and companies such as OpenAI reach $200B-$300B of revenue quickly, or current valuations are wrong “by an order of magnitude.” Figma may be down 63% from its IPO-day peak, but at $53 it still traded around 26× revenue; the deeper risk is that 30%-growth public software now commands roughly 15-20× versus a historical 6-7× NTM benchmark.
OpenAI can win the AI race without fulfilling every dollar of its trillion-dollar infrastructure vision. The proposed 125× energy expansion would require more energy supply than India’s current capacity, while one 10-gigawatt NVIDIA build alone would consume more power than New York City; David Friedberg expects financing, construction and adoption constraints to force forecasts lower. Chamath Palihapitiya says Sam Altman is “willing” the trillion into existence, but the system could stop at $400B or $600B—and David Friedberg’s distinction is crucial: “We foolishly thought it wasn’t a metaphor. We thought it was a PO.”
Fivetran buying dbt is the kind of portfolio consolidation required to clear venture’s liquidity backlog. With roughly $400M and $100M of reported ARR respectively, the adjacent products could form an IPO-scale business; against 600-700 unicorns and only about 15 IPOs year to date, combining assets is increasingly unavoidable. The ownership trade-off depends on the structure: 20% of one company can become roughly 8% of a combined company, but Jason argues that “20% of something that’s not going public is not nearly as interesting as 8% of something that is.”
AI has broken tech PE’s assumptions that products remain stable and seat-based economics are predictable. Jason’s Pipedrive example captures the old world: it could take four years to ship a mobile app, while business software barely changed from 2008 to 2023. Now product-market fit can disappear as models advance, and 12 AI agents may require only two Salesforce seats—making both product durability and NRR far less dependable.

Deep dive

1. Burn multiple survived, but its hidden assumptions did not

Jason pulled the sharpest finding from ICONIQ’s software report: AI-native companies below $100M ARR averaged -126% free-cash-flow margins, versus -56% for non-AI businesses, yet grew so quickly that their burn multiples were lower. The afterburners are expensive, but they reach “Mach 10” faster.
His concrete case was Lovable: imagine putting $200M into it while it burns $6M monthly. Most VCs would recoil, but if it adds $300M of ARR, the capital consumed per dollar of new ARR remains low—and venture gets extraordinary revenue leverage.
Rory restated the arithmetic: spend $2 to add $1 of ARR, and the burn multiple is two; if that revenue receives a 10× valuation, $2 of burn created $10 of market value. “All ratios are wrong, but some ratios are at times useful.”
The buried assumptions now matter more than the output. ARR may not be durable; hypergrowth can conceal churn from a much smaller prior-year base; gross margins may be moving; and model companies carry CapEx the ratio misses. Rory’s cross-check is delta GAAP revenue and GAAP run-rate recognition, but his decisive practical test is cash: under six months of runway beats any elegant ratio.

2. Good companies should take the round before investors change their minds

Harry described a stark market of haves and have-nots: portfolio companies with good burn multiples and strong growth are fundraising into indifference. The founders followed the old scorecard correctly, only to discover that the scorecard no longer guarantees attention.
Rory’s framework was unsentimental: “There’s only two ways of pricing a deal. You price a deal on hope, or you price a deal on the multiples.” At $400M of revenue, fundamentals can support a valuation; at $15M with ordinary growth, a perfectly sound business may offer a VC no meaningful upside option.
Jason’s specimen was a hypothetical $15M ARR, AI-enhanced mug company with good burn and a limited TAM. If Scale offers to invest at 250 on that deal, take it; advice to wait and optimize price because triple-triple-double-double makes the company “golden” is “terrible, terrible advice in 2025.”
Rory noted that roughly 10 of the year’s 15 IPOs had little AI story, so non-AI outcomes plainly remain possible. The difficulty is underwriting seven or eight years of compounding from $10M while capital fixates on AI; founders should close reasonable rounds, conserve cash and earn the right to prove investors wrong.

3. Kingmakers create brands—and walls of follow-on capital

Elad Gil argued that venture investors are unusually reluctant to fund companies near Valley kingmakers such as Harvey or Abridge. Harry agreed that raising aggressively can deter competitors, especially when customers are themselves venture-backed, but said kingmaker status belongs in the risk analysis rather than operating as a binary veto.
The effect is weaker outside Valley-centric markets: oil-and-gas buyers “barely can tell the Sequoias from their KPs.” Kingmaker status matters, but it is not dispositive.
Brand is unusually valuable while nervous AI buyers know they must purchase something but cannot evaluate vendors. Jason’s example: Bolt won a large deal against Lovable because both were known, but Lovable did not call back; the responsive humans became the trusted choice. “Who the hell to buy?” is itself a distribution problem.
The discussion then added the “wall of money”: $20M from a celebrated investor can draw another $60M within months. Money alone cannot create winners—SoftBank “proved the negative”—but challengers now need enough differentiation to overcome $80M, not just a prestigious logo.

4. AI can be real while venture underwriting is still wrong

Harry posed the central fork: are valuations such as $5B-$10B for relatively small-revenue companies, alongside billion-dollar pre-revenue seed rounds, genuinely absurd, or will software’s migration into human-labor budgets make today’s skepticism look small-minded? Jason’s answer was that one of two outcomes will prevail: either AI creates profound productivity changes and companies such as OpenAI reach $200B-$300B of revenue quickly, or current valuations are wrong “by an order of magnitude.”
Elad said the B2B AI wave is still very early and that his team has replaced 11 people with AI agents. Jason separately said venture again feels “almost risk-free,” as if the industry were back in 2021.
David Sacks separated two independent ways a fund loses. A high loss ratio means it selected companies without product-market fit; excessive valuation means it selected winners but paid too much to earn a return. Venture must get both picking and price right, and current funds may be weakening both disciplines simultaneously.
Harry worried that Alexandr Wang’s roughly $14B outcome became a precedent for $30B and $10B-$20B founder-linked valuations elsewhere. David Sacks invoked Irving Fisher’s 1929 claim that stocks had reached “a permanently higher plateau”; whenever investors hear “permanent,” they should be suspicious.

5. Public SaaS remains expensive beneath the post-IPO wreckage

Figma had fallen 63% from its IPO-day peak to $53, leaving locked-up VCs watching the retracement. Jason’s counterweight was that Figma still traded around 26× revenue, while leading public B2B companies averaged roughly 30% growth and commanded about 20× ARR. “The markets are too generous today” is a plausible interpretation.
David Sacks’s mental benchmark was the pre-2019 public-software median: approximately 30% growth at six or seven times NTM revenue. Today that same growth can receive 15-20×; if the “bedrock price” reverts to seven or eight, every rung of the private valuation ladder moves down with it.
Klarna dipping below its IPO price and StubHub selling off hard will make new-issue buyers demand more protection. If established public comps trade at 10×, the unfamiliar IPO previously had to price at eight; buyers may now demand seven. The window stays open, but sellers unwilling to accept that discount may postpone.

6. EA’s $55 billion take-private puts real leverage behind a hits business

David Sacks called EA’s $55B transaction the largest LBO in history, carrying $18B of leverage. Industrial buyers routinely tolerate roughly six times EBITDA, but recurring-revenue manufacturing economics differ from a hits-driven games company; the debt is therefore neither unprecedented in isolation nor trivial in context.
Silver Lake’s presence tempered his concern. He pointed to its Airbnb investment and especially the Dell-EMC sequence—the Dell take-private was “a work of genius”—and summarized the deal as “pretty smart money at the helm” of a very large, unusually risky bet.

7. OpenAI can win AI without consuming the full trillion-dollar vision

Harry framed the physical scale: OpenAI plans to expand energy capacity 125× in eight years, eventually requiring more power than India currently supplies, while potentially seeking $1T for data centers alone. David Friedberg located the causal chain precisely: ambition drives compute demand, and compute drives energy demand.
The evidence keeps pulling David in opposite directions. Claude can reportedly code for 30 hours without human intervention, while Jason Calacanis personally reached three hours—technical progress supports exponential forecasts. Finance, power construction, data rollout, enterprise adoption and software sales suggest the implicit four-to-five-year adoption curve is too optimistic. David expects downward revisions.
Jason’s image was a country dotted with “AI cities”: the announced 10-gigawatt NVIDIA build would need more power than New York City, yet employ only hundreds while producing the equivalent of billions of digital minds. He believes Altman might solve fusion and intervening constraints, while conceding the scale is hard to comprehend.
Chamath treated $1T as an ambition rather than a minimum viable outcome. If only GPT-5 and Claude 4 or 5 were available, progress would continue; $400B-$600B could suffice, and GPUs could last six years rather than three. Altman is “willing” as much of this into existence as possible.
David Friedberg’s synthesis preserved both sides: “Whatever the prize is for being the best company in AI, OpenAI’s gonna get that prize.” Yet an experienced CFO would not pre-spend a CEO’s most aggressive forecast; if NVIDIA and Oracle are valued as though every metaphor were a purchase order, slower—but still extraordinary—50%-60% growth creates the painful ripple effects.

8. Meta has earned another roll, not a presumption of success

Harry disclosed Meta as his largest public position but said his confidence in its AI strategy had “dwindled and dwindled,” citing Alexandr Wang, Yann LeCun’s treatment and the organization of competing teams. OpenAI, Anthropic, Microsoft and Google sharpen the opportunity cost of disorder.
Jason would not certify the strategy as weak, but called Zuckerberg a poor communicator beside Altman: investors cannot see where the thick-glasses vision leads. Zuckerberg did provide one Altman-level statement—he would rather burn $20B of operating income and fail than allow Meta to become irrelevant.
Brad Gerstner distinguished permission from probability: prior success means Zuckerberg “earned the right to roll again,” not that this roll is correct. Brad said he would take a bet that the spending produces little meaningful revenue and resembles Meta VR more than Instagram or WhatsApp; a 50% hit rate can still generate exceptional DPI when the winners swamp the failures.
Brad framed the existential mechanism as attention, not marginally better ad targeting. Two hours spent in ChatGPT are two hours not spent on Facebook, so Meta must invent a major product again after two decades optimizing its original engine. “We’re just gonna make shit until somehow we get people to come back and play with us.”

9. ChatGPT commerce must reach billions or remain an experiment

Brad Gerstner said purchasing inside ChatGPT is part of an inevitable set of experiments: Google and OpenAI are exploring commerce protocols, and free users eventually must be monetized. With infrastructure costs approaching the scale discussed, “you can sell them shit, or you can sell advertising to them.”
Jason’s hedge was materiality. Joint announcements may generate attention, but commerce could be an integration rather than a top-five OpenAI initiative; it likely needs roughly $2B of revenue next year just to move the needle. Existing Instagram and Pinterest behavior also suggests discovery does not automatically become checkout.
Brad recalled that users accepted ads on Facebook and Instagram more readily than native purchasing, making advertising the easier-looking route. Jason said the apps leadership must create several multi-billion-dollar streams within two or three years, while Brad emphasized that “there’s a thousand billions in a trillion,” exposing how little a merely successful feature contributes against the CapEx vision.

10. Fivetran–dbt is the consolidation venture portfolios need

Jason called Fivetran and dbt a “damn smart obvious combo” from a distance: Fivetran last reported around $400M ARR and dbt around $100M, implying more than $500M combined before accounting for subsequent growth. Their adjacent data products should not leave investors asking why the businesses belong together.
The portfolio arithmetic makes consolidation unavoidable. The panel cited roughly 600-700 unicorns and about 15 companies through the IPO gate year to date; at around 20 IPOs for the year, the backlog represents approximately 30 years. Consolidation is part of the work required to make more companies IPO-able.
Jason observed that Andreessen Horowitz being the lead or near-lead in both companies makes alignment easier. His ownership thought experiment explains the conflict: owning 20% of one asset can become roughly 8% of a merged company, while owning both assets can make the spreadsheet case easier and reduce internal resistance.
Jason argued that “20% of something that’s not going public is not nearly as interesting as 8% of something that is.” With an IPO threshold around $300M-$400M, value behaves less like a smooth continuum than “electron states”: below critical mass lies a painful private exit; above it lies public-market liquidity. Rory countered that, if the upside is sufficiently large, owning 20% of the combined company can still be the better economic outcome, even if the partnership’s incentives make the decision difficult.
The greater danger is turning 20% of a well-run company into 8% of a failed integration. The CEO must believe the assets obviously belong together; investor-engineered portfolio mashups invite disaster. Jason criticized the version that combines underinvested or shrinking properties, while Harry pushed back that a Salesloft–Clari-style combination can make intuitive product sense. The partner and CEO fit are decisive.

11. AI has broken tech PE’s two quiet assumptions

Harry asked whether tech PE now receives too little upside for rising displacement risk, especially when other firms can deploy more capital at richer multiples. Rory advised against copying late-stage growth investors: firms rarely transplant successfully into a different discipline. The correct response is rigorous AI-downside underwriting, not abandoning control investing.
Jason identified the historical subsidy PE may have underappreciated: business-software products barely changed from 2008 to 2023. Pipedrive took four years to launch a mobile app and still produced a $1B-$1.5B cash exit. A 140% NRR model could function as “spreadsheet glue,” while a buyer could even acquire Marketo and fire everyone.
That stability has vanished on both sides of the table. Incumbent products face technical obsolescence, while post-LLM startups can lock in and out of product-market fit as foundation models change. A product that felt definitive one year earlier can suddenly look obsolete, making rapid growth compensation for much higher instability.
Seat economics compound the risk. Jason runs 12 agents but needs only two Salesforce seats; six agents may resemble six workers without purchasing six licenses. Agentforce could eventually make Salesforce more money, but until pricing shifts, agents attack the unit PE traditionally modeled—even if seats never literally reach zero.
Jeff Lawson’s distinction stayed with Jason: Twilio’s API layer could benefit from the AI boom, while seat models are exposed. Rory closed with Accenture’s corporate epitaph for workers it could not retrain: “We are exiting on a compressed timeline”—a polished formulation of AI turning both human and software seats variable.

12. CEOs retain political rights, but business damage can become dispositive

Jason Lemkin resisted saying CEOs surrender personal political expression, while arguing that companies should stick to their missions and avoid culture wars. He invoked the University of Chicago principles and Eisenhower in 1952, when both parties courted him because his political affiliation was unknown.
David Sacks’s tactical experience was bleaker. He has privately warned roughly 10 executives that a post landed differently than intended; only one ultimately welcomed the feedback. The others understood they might alienate 40% of customers or upset underrepresented employees and effectively answered, “I feel so strongly, I don’t care.”
David Friedberg’s boundary appears when speech causes half the customer base to cancel or key employees to leave; the individual may retain the opinion yet cease to be the right CEO. Harry’s counterweight was the attention cycle: Deel and Rippling’s fight, or even Elon Musk and Donald Trump’s rupture, rapidly left the foreground. “Just keep moving forward—the rear-view mirror, it vanishes so quickly.”

Ro Khanna Calls Out Both Parties: Immigration, Tech & Free Speech Exor: Fiat Crisis to Ferrari Glory - [Business Breakdowns, EP.229]