Sam Altman's Masterplan or a Gift to Anthropic? Palantir & Shopify Crush Earnings
Summary
GPT-5’s underwhelming launch was framed less as a capability failure than as frontier AI entering its commercial grind. Aaron Levie reportedly found document comparison, redlining and term extraction “materially better,” while another guest saw the broader shift as moving from AGI grandiosity to “grind it out, make it better, build a business.” The hedge remains important: exponential takeoff might arrive, but current evidence shifted toward slower improvement.
GPT-5’s price may matter more than its demos because it gives coding platforms leverage against Anthropic. On one guest’s submitted workload, it appeared 8–10x cheaper than the most expensive alternatives, prompting Cursor to push it into its user base and threatening Anthropic’s cited $6 billion revenue pool. Harry argued Anthropic can close a temporary efficiency gap; Rory’s counter was that nobody prefers monopoly economics becoming an oligopoly: “Use the cheap shit where you can and use the dear stuff where you have to.”
OpenAI can plausibly justify enormous value without achieving AGI if ChatGPT becomes the default paid information service. Rory became more confident at a $500 billion valuation because the company can replace civilization-scale promises with a three-to-five-year path toward business fundamentals: a mass-market $20 subscription, higher-priced tiers and eventually advertising. Using the episode’s figures of roughly 700 million free users and 20–30 million paying subscribers, his simple case reached $1–2 trillion.
Perplexity’s $34.5 billion Chrome offer exposed browsers as the revived distribution layer for AI. Chrome has little standalone revenue, but a buyer could install its AI engine in front of roughly a billion users and monetize even a 1–2% paid conversion; Google itself would remain the best search monetizer if another party owned the browser. Whether funded or executable, the bid also served Perplexity’s need for “constant marketing” in a market where companies outside the top two risk disappearing from consideration.
AI spending is rewarding both native applications and infrastructure companies attached to them. n8n reportedly reached a $3 billion valuation while moving from roughly $40 million ARR toward an expected $80 million, after AI transformed workflow automation from routing work to doing it. Datadog posted a record $260 million of net-new ARR and reportedly receives $240 million annually from OpenAI; RevenueCat, which powers 40% of mobile subscriptions, had already doubled usage this year as AI customers proliferated.
Palantir’s reacceleration may be unprecedented in enterprise software, but its valuation demands nearly flawless compounding. Growth moved from 12% at roughly $2 billion of revenue in 2023 to almost 45% at about $4 billion ARR, while US commercial bookings reached $843 million, up 222%. Its advantage is selling a credible $10 million outcome between a fragile $100,000 startup and a bespoke Accenture project. Yet at roughly 120x revenue, both guests took the under on a $2 trillion market cap within five years.
The defining operating call was that B2B winners will combine growth with far fewer employees. Shopify grew revenue 91% from its 2022 employee peak while cutting headcount from 11,600 to 8,100, reaching roughly $1.3 million of revenue per employee; Alex Karp said Palantir could become 10x larger with 10% fewer employees. One guest’s blunt conclusion—“You don’t need half your company”—was paired with a warning that AI-generated visibility will expose employees who neither know the product nor produce measurable work.
Venture is concentrating into fewer winners, while ownership and labor income may concentrate inside those winners too. Carta’s Q2 2025 data showed record seed valuations alongside fewer rounds, and OpenAI’s $40 billion financing exceeded the roughly $12 billion raised across one firm’s entire quarterly enterprise-B2B opportunity set. The one-person billion-dollar company was dismissed as a literal model, but 20–40-person companies look plausible; the best AI orchestrators and sellers may capture disproportionate equity and compensation.
Deep dive
1. GPT-5’s thud marked a transition from spectacle to software
Rory’s first GPT-5 experience was genuinely poor: it claimed the market had suffered its greatest crash since the tulip era. Yet he separated consumer disappointment from enterprise usefulness, pointing to Aaron Levie’s Box testing of redlining, document comparison and term extraction as evidence that the model was “materially better” where businesses pay.
Another guest welcomed the underwhelming reception because it deflated the “we’re underway to AGI” narrative. His categorical prior remains that exponential takeoff is “AGI rubbish,” though he allowed, “Maybe I’m wrong”; the new evidence merely shifts probability toward progress happening “not nearly as quickly as you think.”
The analogy was Windows 95 or the first iPhone: a transformative launch followed by 10–15 years of incremental improvement. Frontier AI may now be past its zero-to-one moment and entering the less glamorous phase of product quality, conversion, reliability, margins and distribution—“grind it out, make it better, build a business.”
OpenAI’s return to open-source models after more than two years, plus its move from confusing model names toward a unified selector and routing, reinforced that interpretation. Rory called these product-manager decisions: stop imagining oneself as “the next Robert Oppenheimer” and pursue the change that produces another 5% of user conversion.
2. Model economics turned GPT-5 into an attack on Anthropic
One guest’s colleagues still preferred Anthropic’s highest-end coding products, but his submitted workload suggested GPT-5 was 8–10x cheaper than the most expensive token option. For Cursor, whose gross margin had been constrained by Anthropic, a competitive lower-cost supplier was “the best damn thing that ever happened.”
Harry believed Anthropic could close today’s efficiency gap: costs have fallen too quickly over 12 months for the difference to be permanent, so Anthropic can reasonably believe that superior models will win once its efficiency catches up.
Rory’s pushback—worth keeping—was that Anthropic can remain the best product and still be worse off. A monopolist becoming one vendor in an oligopoly faces routing by workload: “Use the cheap shit where you can and use the dear stuff where you have to.” Better performance wins usage, but no longer as easily or profitably.
The panel also disagreed over Sam Altman’s foresight. One guest cast him as an Elon-like marketing mastermind who must have anticipated the thud; Rory noted that OpenAI did not recognize ChatGPT as its defining release while focusing on GPT-4. Directional vision, drive and capital can create fortunate breaks without implying perfect step-by-step clarity.
3. Shipping pressure may explain why GPT-5 arrived unfinished
Harry’s operating theory was prosaic: GPT-5 had not achieved what Altman wanted by July 31, but leaders eventually have to deny the team another week, month or quarter. Amid talent moving between OpenAI, Meta and competitors, Altman may simply have declared August the deadline and decided to “box it up and ship it.”
In that framing, this release might effectively be “4.9” or “5.00”: enough to improve Aaron Levie’s workflows, lower costs and establish a new baseline, but not the imagined Nirvana. The organizational benefit of forcing convergence and shipping could outweigh another indefinite delay.
Harry added a financing incentive: releasing before a proposed $500 billion raise sustains momentum better than continuous waiting. The same logic applied to Replit’s cited $3 billion round—once the window is open, “just get it done.”
4. OpenAI no longer needs AGI to support a giant outcome
Rory said he was more confident investing at $500 billion after GPT-5. Grandiosity was necessary when OpenAI needed $30–40 billion before it could credibly pitch a chat application replacing search; changing humanity and eliminating labor was a more effective fundraising story than “trust me, people will use this.”
The next mandate is conventional: ship, improve the product and converge toward profitability over three to five years. Rory now sees OpenAI and Anthropic as roughly half-trillion-to-trillion-dollar opportunities without AGI, because consumers can pay $20 monthly while some users support more expensive tiers.
For OpenAI specifically, the mass-market destination looks increasingly locked in: become the default place people seek information, then layer subscriptions and potentially advertising onto global reach. Rory’s compressed valuation case was “plus or minus your Google with a subscription business”—enough to reach $1–2 trillion, albeit with noise along the way.
Harry reduced the portfolio decision to his crypto analogy: rather than predict which model wins every changing benchmark, own the foundational asset, Nvidia. Rory accepted the rebuke to his rationalist tendency: Bitcoin’s early use-case stories mostly failed while Bitcoin itself worked, illustrating how overthinking a narrative can obscure the durable core.
5. Chrome is valuable because AI created a new monetization machine
Perplexity’s stated $34.5 billion Chrome offer—later discussed loosely as $38 billion—raised two separate questions: whether its capital was real and whether Google would ever sell. Rory’s answer to the second was unequivocal: Google does not want to sell, and only an exhausted Department of Justice process could force it.
Chrome itself does not charge users, making its value buyer-dependent. Google pays Apple roughly $20 billion annually for Safari search placement; paradoxically, Google would also be the party most able to monetize Chrome if somebody else owned it, because search lets it pay the highest price for default traffic.
Perplexity’s strategic case was clear: replace the laborious task of winning users browser by browser with default access to Chrome’s installed base. Using the conversation’s roughly 700 million free and 20–30 million paid ChatGPT users as a reference, even a 1–2% conversion from a billion-user funnel could support a compelling subscription business.
Execution remains doubtful. Google can litigate “to the end of human time,” and another buyer—Satya Nadella was floated at $39 billion—could emerge. But an AI answer engine would “absolutely kill to own the Chrome user base,” so the bid’s industrial logic survives uncertainty over funding and price.
6. Perplexity’s bid doubled as necessary founder-led marketing
One guest’s broader lesson was that AI requires “constant marketing” and a place among the top two perceived players. Sam Altman, Dario Amodei, Aravind Srinivas and Jensen Huang all stay publicly visible because viral product adoption alone does not keep a company in the market’s consideration set.
Aravind had previously told Harry that acting like a politician or state leader—speaking continuously for the machine—was effectively his job. Even if Perplexity never acquires Chrome, putting itself at the center of the browser conversation can be rational when conventional AdWords, SEO and sponsored posts cannot fuel frontier-level growth.
Harry contrasted that visibility with Cohere and Mistral: he did not claim their models were necessarily inferior, but their relative absence from press and public debate correlated with being perceived as losers. Devin offered another example—a tool loved by two of the guest’s companies for niche uses but rarely discussed.
7. Google may lose monopoly economics without losing the AI market
Rory attacked the timing of US antitrust remedies: regulators spent years targeting Google’s search dominance just as ChatGPT created a credible product threat. His analogy was Microsoft, which faced serious intervention when Google was already ascending; delayed enforcement risks kicking incumbents only after market pressure has begun doing the work.
Chrome has consequently returned as an “accidental gem.” Google originally built it to counter Internet Explorer, open-sourced Chromium and watched much of the browser market standardize around it; after a decade of seeming like mature plumbing, the browser is again the gateway to a monetizable computing layer.
Harry’s “AI for normies” argument used Threads as a warning against judging mass behavior from a technology-insider bubble. Gemini is already good and improving, and Google can put it inside its core distribution channel; many users will accept the first adequate AI answer without changing products.
Rory agreed that Google will retain meaningful share and noted the present paradox: search faces an existential narrative while search revenue still rises. The likely transition is not death but movement from monopoly to oligopoly—less attractive economically, yet still an excellent business.
8. n8n captured the AI inflection that workflow investors missed
n8n’s financing was reportedly led by Accel at a $3 billion valuation, with ARR around $40 million and an expectation of ending the year near $80 million. Founded in 2019 and valued around $300 million relatively recently, it became one of Europe’s most contested growth deals.
Rory had evaluated the company around 2022 or 2023 but underestimated it amid an indistinct field spanning RPA, process mining, low-code, no-code and workflow automation. AI changed the value proposition from deterministically moving tasks between people to “literally do the work,” making the software valuable enough to remove repetitive labor.
One guest described a “Zapier renaissance”: vibe-coded applications created many more systems that needed connecting, while n8n offered the more developer-oriented version. The addressable need felt an order of magnitude larger, and the company’s acceleration appeared concentrated in just the preceding six to nine months.
The investable selection rule was not merely picking workflow automation. Among perhaps 10 exposed vendors, Rory said the winner is usually the founder who recognizes the change first and declares: nobody leaves until the LLM-enabled version ships and reaches 20 customers by Friday. n8n evidently supplied both the exposure and that urgency.
9. Paying up works only when the return case survives the brand story
Harry saw Accel’s wins in Lovable and n8n partly as a response to Index making other European firms feel left behind. The Facebook precedent supported aggression: Accel once paid what peers considered an absurd roughly $500 million valuation, yet “if you pick right, no price is wrong.”
Rory resisted competitive retaliation as an investment rationale. Growth investors can legitimately underwrite a three-to-five-times base case, knowing power-law outcomes occasionally turn that modest forecast into 100x; trouble begins when “we need to be relevant” quietly substitutes for an actual return model.
He compared brand-driven investing to the Catholic-school account of sin: start with a small compromise and it becomes easier to keep doing deals that cannot return capital. He conceded that relevance sometimes may be worth purchasing, but wanted managers to say honestly when they were squinting toward 3x rather than disguising marketing as underwriting.
Harry’s $2–3 million “YOLO” sleeve complicated the objection: it was created for access and brand, yet had reached roughly 7x as a pool. Small positions in companies such as Perplexity also created durable founder relationships; the limitation is that even a 10–12x sleeve may return only a small percentage of the overall fund.
10. Early DPI is useful, but liquidity windows matter more than optics
One guest argued that small, fast outcomes can put credible points on the board when raising the next fund. In his 2017 fund, several nominally large early exits returned roughly 20% quickly—useful when core holdings may remain illiquid for 15, 20 or even more years.
The trade-off is selling a future winner merely to manufacture DPI. In one example, Horsley Bridge advised a guest not to sell a company at a $1 billion value because the resulting cash still was not enough of the fund; venture outcomes remain driven by preserving exceptional upside, not optimizing each interim fundraising slide.
The discussion’s conclusion from Horsley Bridge was that venture is difficult unless managers exploit brief periods of hyper-liquidity. The right approach is to lean in while recognizing that the market “will fall off a cliff,” then become disciplined enough to lean out before it closes.
Rory’s broader point was that fundraising creates pressure to show cash returns even when the absolute contribution is small. A high multiple on 10–20% of a fund can demonstrate progress, though it does not replace a fund return.
11. Datadog showed why predicting the Street is a losing exercise
Datadog produced its best net-new ARR quarter, adding $260 million, yet its shares fell roughly 10% and traded below their pre-results level after initially rising after hours. Harry could not reconcile the operational result with the market response.
Rory’s answer came from public-company boardrooms: executives would review their known results, predict the next stock move and be wrong at least half the time. Investors compare results not only with published expectations but with private estimates of what everybody else expected—“a deranged madman” whose reaction is “utterly unknowable.”
The recommended operating response was to ignore daily interpretation. A board should celebrate signing a huge ARR contract with the most exciting company in the market; possible price renegotiation two years later is secondary to the revenue obtained now.
That applied to OpenAI’s reportedly $240 million annual Datadog spend. It is a concentration risk and OpenAI may negotiate lower pricing, but it is first a gift: the customer might instead double as its own usage grows, particularly if replacing Datadog would consume scarce engineering attention.
12. The AI economy rewards every layer attached to its bill of materials
RevenueCat, which Harry and one guest back, powers roughly 40% of mobile subscriptions and had already doubled usage this year—not because it became an AI company, but because AI application builders use its infrastructure.
The “bill of materials” test was to imagine building a frontier model: chips, data centers, power, engineers, monitoring and the surrounding software required to keep everything operating. Nvidia captures the obvious spend, but Datadog and other secondary suppliers can co-attach to the same capital wave.
The macro mismatch was striking: the panel said AI capex was contributing nearly all current GDP growth while application-level productivity and revenue remained harder to see. For B2B vendors, the practical call was to pursue the customers actually expanding budgets; legacy buyers such as John Deere were described as flat or spending slightly less.
Concentration does introduce fragility because only a few buyers operate at frontier scale and each can demand pricing concessions. The panel nevertheless rejected treating a massive AI customer as a curse: operators should prefer receiving the money now and manage the renewal problem if it arrives.
13. Public-market SaaS winners are stronger than the “SaaS is dead” thesis
Harry grouped AppLovin, HubSpot, Shopify, Datadog and Palantir as “bigish tech”—below the Magnificent Seven but benefiting from AI, scale and renewed operating discipline. Rory’s distinction was crucial: public category leaders can defend themselves, grow and produce cash, even if funding a new “Shopify” at $1 million of revenue is unattractive.
Their AI exposure differs. Palantir sits directly in the implementation wave; Datadog receives a major infrastructure benefit through OpenAI; Shopify and HubSpot use AI but are not comparably direct beneficiaries. All retain market leadership, strong founder-led execution or established scale that a small entrant cannot easily recreate.
Monday.com supplied the valuation warning. A good quarter and still-elite near-30% growth were insufficient when forward expectations softened, sending the stock down roughly 30% to about 8.4x ARR. One guest said he bought aggressively, while another called the multiple “soul crushing” for private portfolios priced above it with inferior metrics.
Historically, the median SaaS company traded near 6.3x revenue while growing almost 30%; today’s median was described as roughly 6x with nearer 20% growth but greater profitability. Once growth disappears, even good software can compress toward four-to-five-times revenue—hence the instruction to remain “growth bigots” when underwriting private rounds.
14. Palantir owns the scarce middle between software and consulting
Palantir’s reported reacceleration was the central public-equity fact: from roughly 12% growth at $2 billion of revenue in 2023 to almost 45% at approximately $4 billion ARR. Rory’s historical base rate was sobering—only about one-third of companies reaccelerate for one year and roughly one in nine or 10 sustain it for two.
US commercial bookings reached $843 million, up 222%, alongside a record number of contracts worth $5 million or more. The company has become a major implementation channel for both corporate and defense AI, while its long government history gives executives confidence that a complex project can survive deployment.
Rory framed the competitive wedge through buyer risk. A startup offers a more elegant $100,000 tool but requires substantial customer effort; Accenture proposes building the solution from scratch. Palantir can tell a Fortune 100 CEO, “Give us the $10 million, we’ll get this puppy done,” using its platform and forward-deployed engineers while retaining roughly 50% margins.
The price incorporates much of that promise: roughly 120x revenue, with one cited calculation requiring five years of 40–50% growth merely to approach Google’s present multiple. The panel invoked Scott McNealy’s warning about Sun at 10x revenue, yet acknowledged three exceptional tailwinds—enterprise AI, defense spending and support from the administration.
15. Ruthless efficiency is becoming the defining B2B operating model
Alex Karp said a Palantir 10x larger—roughly $40 billion—could employ 10% fewer people than today. Rory would not place that forecast directly into a valuation model, but Palantir’s current “Rule of 94” and roughly 50% operating margin show that headcount need not track revenue conventionally.
Shopify supplied the harder evidence. From its 2022 peak of 11,600 employees, revenue rose 91% to roughly $11 billion while headcount fell to 8,100, or about 30% fewer people and $1.3 million of revenue per employee. Harry projected that it could reach 7,000 employees while becoming “200% bigger.”
One guest’s rhetoric was intentionally severe: “You don’t need half your company.” He grouped Tobi Lütke, Mark Zuckerberg and Karp as ruthless leaders, arguing that B2B companies unwilling to follow will lose; the contrast was the 2020–22 period of remote work, multiple jobs and managerial coddling.
The underlying mechanism was not layoffs alone but product fluency and automated measurement. An AI can already know every Shopify feature better than a weak SDR, so employees who neither master the product nor produce clearly attributable work face a deteriorating claim on their roles.
16. AI transparency may create fear without managers deliberately creating it
The discussion’s specimen was Momentum.io, which synthesizes Gong, Granola and other sales data into real-time operating insight. A guest said that every company where he introduced it saw somebody on the sales team quit on day one—including one person that afternoon—because “the gig was up.”
Harry asked whether uncertainty should be cultivated to force people toward irreplaceable work. The guests said overt threats are dated and unnecessary: AI systems will increasingly reveal missed code, absent content, weak calls and shallow product knowledge, causing underperformers either to step up, leave voluntarily or be moved out.
Rory initially distinguished ordinary employment from exceptional upside. Teachers and interchangeable workers should not live in terror; highly paid employees in entrepreneurial companies—including someone earning $100 million over four years at Meta—should recognize that extraordinary compensation carries extraordinary performance risk.
Harry’s concern centered on non-specialist workers aged 23–30 and seasoned B2B executives. He said his ability to place a known executive with one email had fallen to roughly 10% because employers now reject candidates lacking urgency, office commitment or AI-tool proficiency.
17. The labor debate ended with a reluctant change of mind
Rory first argued that a young worker losing a $100,000 role might find another at $95,000 or $80,000 and should adapt; fear becomes qualitatively different at 55, when a displaced worker may have no comparable path for the next decade. Harry, whose family had lost everything, emphasized downside rather than theoretical mobility.
Harry then revised his own framing as the examples accumulated. He said many $100,000–$200,000 technology workers may find that their next-best use pays 70–80% of their current income, not merely 5–20% less: “Unfortunately, more people have to be scared.” He still rejected fear as socially desirable, but accepted it as an increasingly accurate response to economic exposure.
Harry and Rory agreed that employers are not queuing for generalists who resist changed working norms or cannot use AI tools. The adjustment may be survivable for young graduates, but survival is different from retaining the equity upside, status and income of a role at a compounding technology winner.
18. The one-person billion-dollar company is a metaphor; 30 people is plausible
One guest rejected the literal one-person company as “idiotic in the extreme.” A billion-dollar business still needs customer conversations, accounts, taxes, legal review and protection against one founder being incapacitated; outsourcing those functions merely means other people perform them outside the payroll.
Rory accepted the bus-factor objection but predicted many billion-dollar software businesses with 20, 30 or 40 core employees. A small engineering group, AI agents, self-service distribution and outsourced accounting or legal work can deliberately trade some human-assisted revenue for extreme ownership and operating leverage.
His analogy was packaged software: write it once, place it on a CD-ROM and sell a million copies. AI could create a renaissance of that leverage, particularly in PLG or SMB products that do not require Palantir-style forward deployment.
The discussion offered a live example rather than a literal one-person company. One guest said an AI BDR had set up three meetings in two days for six-figure sponsorships; another said his team spends two hours a day orchestrating AIs that “don’t quit.” A London event could still require temporary workers to scan badges.
19. AI orchestration will concentrate equity and compensation
Lean employment does not mean zero human work; it changes who receives equity. Temporary contractors may run events, accounting firms may close books and lawyers may review filings, while ownership remains concentrated among the small permanent team operating the system.
One guest predicted the best AI-fueled salesperson could earn $10 million rather than today’s roughly $1 million, supported by 10 representatives instead of managing 200. Fewer people would capture more upside, while displaced workers rotate through intermittent, lower-value work.
Harry identified “chief orchestration officer” as the emerging career path, and one guest said the role could already be worth $500,000. The scarce skill becomes making multiple agents work together reliably—not merely prompting once, but supervising output, routing context, correcting failure and converting automated activity into commercial results.
20. Venture’s record valuations conceal unprecedented concentration
Carta’s Q2 2025 data showed the highest seed valuations on record, but one guest stressed that fewer seed rounds were completed than 12 or 24 months earlier. Rising medians therefore describe concentrated competition for selected winners, not a universally stronger financing market.
At late stage, the statistics have become almost meaningless without exclusions. One firm estimated roughly $12 billion raised across its entire quarterly enterprise-B2B sweet spot, while OpenAI alone raised $40 billion; Meta’s Scale AI transaction and giant Anthropic and xAI rounds similarly distort industry totals.
Some concentration should persist. Frontier models and defense companies are more capital-intensive than SaaS, while businesses remaining private at $10 billion scale need billions of balance-sheet liquidity even when profitable. The market is unlikely to return to a world where nearly everything is an A, B or C and no round exceeds $100 million.
That structure revived Harry’s view of mega-funds. If a manager can deploy $1 billion into an OpenAI-scale winner and plausibly earn 10x, enormous outcomes can overcome enormous fund size; LPs wanting model-company exposure may have no diversified alternative. Only LPs closing the capital valve would force a full return to the older venture regime.
21. The closing bets preserved valuation discipline amid AI enthusiasm
On Palantir reaching a $2 trillion market cap within five years from roughly $450 billion, both guests chose the under. Rory noted that “AI is bigger than cloud,” but Salesforce’s roughly $222 billion value at $40 billion of revenue made valuation gravity impossible to ignore.
Rory refused to predict whether Stripe would announce an IPO before June 1, 2027. The company is cash-flow positive at huge scale and private liquidity already exists, so only a materially cheaper public cost of capital might compel action; without the Collisons’ stated preference, it remains an “idiosyncratic bet.”
Rory leaned toward xAI eventually suing Apple if Apple remains aligned with ChatGPT. The call rested less on legal merits than on Elon Musk’s propensity to fight, the personal and commercial nature of his OpenAI dispute, and the likelihood that adjacent partners get pulled into an existential contest through discovery and executive depositions.