Summary

GPT-5’s thud is the bull case, not the bear case. Rory O’Driscoll’s read: “underwhelming is great” — it deflated the “we’re underway to AGI” noise and moved the industry into the Windows 95/iPhone phase of “10 or 15 years of incrementally making it better until you plateau.” He’s more confident in OpenAI at $500B post-launch: the grandiosity was needed “when you have to raise 30 billion on nothing,” and now both OpenAI and Anthropic can be “half a trillion to trillion dollar companies” without any AGI at all.
The real GPT-5 story is a pricing attack on Anthropic’s coding revenue. The product is “not as good as the high-end Claude code… but a damn sight cheaper” — 8-10X cheaper on the most expensive token load — and Cursor is already pushing a GPT-5-based product. Rory’s channel logic: “my gross margins are set by my worst competitor,” so for Cursor “this is the best damn thing that ever happened.” Anthropic is laughing at the launch, but “no one ever said, ‘I’m really delighted my bigger competitor entered my market with a product five X cheaper than us.’”
The co-attach trade is where the AI money is actually showing up. OpenAI reportedly pays Datadog $240M a year; “almost the entire GDP growth is AI CapEx,” so anything on the bill of materials — chips, power, observability, software tools — gets a pop even if it isn’t an AI company. Meanwhile ignore the street’s reaction to earnings entirely: Datadog printed its best net-new ARR quarter ever and fell 10%; “it’s like dealing with a deranged madman trying to estimate what the Street will do.”
Palantir’s re-acceleration is unprecedented — and both panelists still take the under on $2T by 2030. Growth went from 12% at ~$2B revenue in 2023 to ~45% at $4B, with US commercial bookings of $843M up 222%; only ~1 in 10 companies re-accelerate two years running. But at ~120x revenues, five years of 40-50% growth just gets you valued like Google today — “I would be terrified of buying it at that price.”
Perplexity’s $34.5B Chrome bid values an asset worthless in itself. Google will “litigate this to the end of human time,” and Chrome’s value depends heavily on the buyer — but plug ~1B users into ChatGPT-style ~2% paid conversion and “suddenly Chrome matters… ‘cause you can plug it into a machine now that can collect checks.” Jason Lemkin’s cynical layer: the bid is also marketing — in AI you must be in the top-two conversation, and the quiet ones (likely Cohere, Mistral) “are also the ones which people have deemed the losers.”
“You don’t need half your company” — the efficiency reckoning is here. Shopify grew revenue 91% while cutting headcount from 11,600 to 8,100 ($1.3M revenue/employee); Karp says Palantir at 10X the revenue will have 10% fewer employees. Jason’s warning for labor: he could once place any veteran B2B exec with one email, now odds are 10% — “go to Cisco before Cisco realizes they don’t need you.”
Venture concentration is unprecedented and not mean-reverting. Highest seed/A valuations ever on fewer rounds; in Q1 Rory’s entire enterprise-B2B sweet spot totaled ~$12B raised while OpenAI alone raised $40B — “twice our entire addressable market… done in one deal.” And Monday trading at 8.4x ARR after a 30% drawdown is the private-portfolio soul-crusher: “you sure your portfolio’s so good if it’s trading at 8.4X?”

Deep dive

1. “Underwhelming is great” — GPT-5’s thud ends the AGI trade

Rory’s contrarian frame sets the episode: the launch “took a little bit of the air out of” the techno-optimist AGI noise, “which is great because we’re now at the it’s a really great piece of software for doing business, let’s make it better stage of life.” His analogy: Windows 95 and the early iPhone — a big launch, then “10 or 15 years of incrementally making it better until you plateau.”
The implicit call, stated plainly: “I don’t buy any of this… exponential takeoff, all that AGI rubbish. I’ve always assumed it’s rubbish, and maybe I’m wrong, but at least right now, the evidence shifted a little more in the favor of perhaps not nearly as quickly as you think.”
Jason’s split verdict: his own first experience was bad (“it said we had the greatest market crash since the tulip era”), but Aaron Levie running it through Box reports redlining, document comparison and term extraction are materially better — and if it’s materially better at coding, it goes straight at Anthropic’s coding business. Harry on the launch itself: demos terrible, the headline feature was model routing — “We waited for GPT-5 and you gave me better model? This is disappointing.”

2. The pricing attack: Cursor is the biggest winner

Rory’s business read: the model is “not as good as the high-end Claude code… but a damn sight cheaper. 10X cheaper than most expensive token load based on what I saw, maybe 8X to 10X” — and Cursor is already pushing a GPT-5-based product with free demos to its user base.
The mechanism that makes it tradeable: “If I’m now Cursor… my gross margins are set by my worst competitor, Anthropic, and now I’m ecstatic ‘cause the big-ass guy on the block is now another vendor of tokens, significantly cheaper.” His conclusion: “It’s not as sexy as AGI making us all unemployed, but if you’re Cursor, this is the best damn thing that ever happened.” He adds Anthropic “maybe overplayed their hand a little bit by kind of bullying Windsurf.”

3. Is Anthropic right to laugh? Monopolist vs oligopolist

Harry reports the Anthropic team “bluntly laughing” at the launch. Rory’s pushback — worth keeping: laughing at the thud and the “mathematically wrong” graph, sure, but “no one ever said, ‘I’m really delighted my bigger competitor entered my market with a product five X cheaper than us.’” Better to be a monopolist with the best product than an oligopolist with the best product.
The gross-margin logic buyers will run: “you’re gonna have some kind of use the cheap shit where you can and use the dear stuff where you have to” — consensus being that token-for-token, Anthropic’s products do more and are more efficient.
Harry’s counter: this is a moment-in-time view — token costs are collapsing within 12-month windows, and Anthropic is thinking “fine, they might have an efficiency on us today, but if our models are better, we’ll win.” Rory concedes but reframes: “Yes, you’ll win, but it just won’t be quite as easy… We’re dealing now just business fundamentals.”

4. Did Sam ship a known thud on purpose?

Jason’s theory: Sam Altman is “a marketing mastermind like Elon Musk” — the thud was known in advance. GPT-5 didn’t achieve what Sam wanted by July 31, and with attrition and resources flowing to Meta, he made the founder’s call every shipper knows: “I’m gonna box it up and ship it… Maybe it’s 4.9, right? Or 5.000.”
Rory doesn’t buy the omniscience: “Put me down as not quite buying that.” The most popular product OpenAI ever shipped — ChatGPT, “70, 80% of their revenue dollars” — they didn’t even think was a major release at the time; they were focused on GPT-4. Directional brilliance plus ability to raise money isn’t “complete clarity on every step.”
Harry adds the financing angle: with a $500 billion fundraise aligning, better to get it out and keep pressing than keep the market waiting — “like the Replit $3 billion round, just get it done.”

5. More confident at $500B: the business no longer needs the dream

Asked directly if he’d do OpenAI at $500B, Rory says more confident post-GPT-5 — because the grandiosity was a fundraising instrument: “When you have to raise 30 billion on nothing, you’re selling dreams, the future of humanity.” Now it’s “I gotta make a business converge roughly on profitability over the next three to five years, so just go run the business.”
His evolved view: without any AGI, both OpenAI and Anthropic “can be plus or minus half a trillion to trillion dollar companies” — everyone paying $20 a month, with the high end mattering more for Anthropic. From $500B “you probably can see a 2, 3X just on that. Probably some noise along the way.”
Even the sneered-at single model selector is the tell: “It’s the kind of thing you do when you’re done with imagining being the next Robert Oppenheimer and now you’re focused on what the product manager tells you will get 5% extra conversion.”

6. Keep the main thing the main thing — Nvidia as Bitcoin

Harry’s analogy: AI is like crypto — don’t agonize over which underlying asset wins (“Do we think OpenAI or Anthropic or Groq are gonna win? I don’t freaking know”), just own the core: Nvidia.
Rory’s self-diagnosis of the rationalist trap: “the crazy stories are bullshit, therefore the thing itself is bullshit, and that’s wrong.” Every 2013 Bitcoin use case except stablecoin payments was bullshit — and Bitcoin worked anyway. Applied here: OpenAI’s market share “looks pretty hard to change,” so a paid tier times a lot of people plus an eventual ad tier means “plus or minus, you’re Google with a subscription business. There you are. It’s $1 to $2 trillion. Thanks for coming.”

7. Perplexity’s $34.5B Chrome bid: a worthless asset bolted to a cash machine

Rory’s anatomy of the deal: Google doesn’t want to sell “in a million years” — the DOJ is forcing it, and Google will “litigate this to the end of human time.” Chrome’s value depends heavily on who buys it: “you don’t get paid for owning a browser,” and ironically the highest monetizer of a divested Chrome would be Google itself, which already pays Apple ~$20B a year for Safari search placement.
Why it works for Perplexity: jam yourself in as the default AI engine for every Chrome user. The math via ChatGPT’s funnel — ~700M free users, 20-30M paying, a bit under 2% conversion — means “suddenly Chrome matters, not ‘cause it got better… but ‘cause you can plug it into a machine now that can collect checks.”
Jason’s marketing layer: in AI you must be one of the top two names in the conversation — hence hackathons, Sam everywhere, Jensen everywhere — and Rory’s additional hypothetical is that even if Perplexity had $38B, “Satya would raise his hand and say, ‘I got 39.’” Harry’s corollary: the quiet ones (likely Cohere and Mistral) do no press “and they are also the ones which people have deemed the losers.”
Rory’s DOJ rant, verbatim and worth the price of admission: “The Department of Justice gets around to killing American companies and technology just when they’re at risk… Leave poor Google alone. They’re just a simple, humble trillion-dollar company trying to survive. So dumb as rocks.”

8. AI for normies: Google doesn’t roll over

Harry’s Threads argument: things his circle sneers at can be massive successes with normal people — and “Gemini is great, actually,” getting plugged into the core distribution channel, so “there’s a real chance they don’t lose any market share” among mainstream users.
Rory agrees on substance: for “a good slug of the population,” an AI answer on Google meets the need; search revenues will keep going up. The compression of the whole debate: “It’s always better to be a monopolist than an oligopolist, but it’s still pretty okay to be an oligopoly.”

9. N8N at $3B: co-attach to AI, and the Accel arms race

The deal: $3B on reportedly $40M ARR ending the year at $80M, Accel led, prior round around $300M, a 2019 company. Rory’s first word: “With envy” — he had talked to them in ‘22/‘23, when workflow automation was an “indigestible mass” of RPA, process mining and low-code. Co-attaching to AI changed the value prop from “we’re gonna automate a little bit of shit” to literally doing the work.
His founder filter for who wins a crowded category: the one who “puts everyone in the room and says, ‘No one’s leaving till we’re shipping an LLM-enabled version of this, and we’re gonna get it in front of 20 customers by Friday.’”
Harry’s market-structure read: Index is making everyone feel like shit in Europe — Accel has to win Lovable and N8N “at any price” or the gap widens. Jason’s precedent: Accel was called washed up when they bid Facebook at $500M pre — “if you pick right, no price is wrong.”
Rory’s caution on relevance-driven deals, via Catholic school: “they’d say you start on little things and then it will just get worse… Once you start doing deals not for return, does it just get worse and worse forever?”

10. Growth math, brand checks, and the case for early DPI

Jason’s growth-investing rule: you need a believable 5X path with dilution — a $3B entry has to support a $20B outcome — and overpaying to stay in the game is defensible “if you genuinely see your way to 5X,” even if it’s no longer a fund-returner. Rory adds the power-law footnote: growth investors underwrite 3-5X and “one in every 10” goes beyond that; an early Facebook investor ultimately made 100X.
Harry’s confession: his fund’s ~$2-3M “YOLO segment” of brand-driven, high-priced positions is at 7X — and it buys founder relationships (“Would likely Aravind and I chat a lot if I hadn’t put a tiny check in? I don’t think so”). Jason’s rejoinder: great, “but it returns 5% of the fund is the problem” — though early DPI has real fundraising value: his 2017 fund returned ~20% early, and “you gotta put some points on the board.”
Harry’s likely Horsley Bridge lesson, the episode’s cleanest venture aphorism: venture is “a very challenging asset class unless you take advantage of very small windows of hyperliquidity.” Rory: “There’s one year in seven where you make most of your money — you don’t wanna not have assets to sell when that year comes around.”

11. Datadog pukes on a record quarter — ignore the deranged madman

The setup: best net-new ARR quarter in company history ($260M in the quarter), initial after-market pop — then down 10% the next day. Rory’s board-game evidence: CEOs with perfect internal information guess the market’s reaction wrong “at least half the time… perfect information doesn’t tell you shit.” His conclusion, the episode’s opening line: “It’s like dealing with a deranged madman trying to estimate what the Street will do. I spend no time on it.”
Jason’s structural point: OpenAI reportedly pays Datadog $240M a year — a second-order AI beneficiary, like RevenueCat (powers 40% of mobile subscriptions, usage already doubled this year without being an AI company). The fragility is real: concentration risk, and OpenAI has already said it will renegotiate the Datadog deal down.
Rory zooms out: “almost the entire GDP growth is AI CapEx” — write the bill of materials for building an LLM (chips, data center, power, coders, observability) and co-attach to it. “The productivity’s not showing up, the revenue’s not showing up at the app level, but the CapEx is showing up in the GDP numbers.” A concentrated AI customer beats not having it.

12. Palantir: re-acceleration without precedent, at a terrifying price

Jason’s numbers: growth went from 12% at ~$2B revenue in 2023 to almost 45% at $4B ARR, record $5M+ contracts, and $843M of US commercial bookings up 222%. “I don’t think in enterprise software we’ve ever seen that level of re-acceleration ever.” Rory’s base rates make it starker: over 20 years, ~1 in 3 companies re-accelerate for one year, ~1 in 9-10 for two — Palantir has done it at scale for two-plus.
The valuation problem: at ~120x revenues, five years of 40-50% growth gets you valued roughly like Google is today — “part of me says, oh my God, five years where you gotta pull it off… I would be terrified of buying it at that price,” with the Scott McNealy 10x-revenues quote invoked as the standing warning. But three trends run in their favor: AI for large corporates, the defense boom, and the administration.
Rory’s moat mechanics: Fortune 100 CEOs doing big AI initiatives need big vendors, and the alternatives are “old and stodgy like IBM and Accenture” or a risky $100K SaaS startup. Palantir can “look the CEO of a Fortune 100 company in the eye and say, ‘We’ve done 10 of these. You give us the $10 million, we’ll get this puppy done.’” And on the services critique: “you can squint and say a lot of it’s services, but who the hell cares? The margins are 50%.”
Karp’s forward claim Jason flags as the future of B2B: a Rule of 94 company today, and at 10X the revenue (~$40B) Palantir “will have 10% less employees than today” — from a founder “in his farm in Vermont in sandals crushing the numbers.”

13. “You don’t need half your company” — the ruthlessness reckoning

The Shopify proof: peak 11,600 employees in 2022, now 8,100 with revenue up 91% at $11B — $1.3M revenue per employee. Jason’s thesis, delivered without hedge: “Tobi’s ruthless. Zuck’s ruthless. Karp’s ruthless. And if you think you’re gonna win in B2B, if you’re not ruthless, you’re gonna lose.” The coddling era of 2020-22 (“take three jobs, work two days from home”) is over.
His most vivid evidence, as told: every company that installs Momentum.io — a tool mashing up Gong, Granola and everything the sales team does into real-time insight — someone on the sales team quits the first day. “Every single time… He quit that afternoon because the gig was up.”
The fear debate: Rory draws tiers — teachers and blue-collar workers shouldn’t “live in terror,” but anyone extraordinarily paid at an entrepreneurial company should be scared; a $100K job automated into a $95K job is “life. That’s American capitalism. Get over it.” Harry’s worry is the 23-30-year-old non-specialist SDRs and marketers “about to get a train hit them in the face” — Rory: “to a rounding error they’ll be fine.”
Jason goes darker, and wider: until ~18 months ago he could place any seasoned B2B executive with one email; today the odds are ~10% — “the fire’s not there… they don’t know the AI tools. I can’t get them a job anymore. I say go to Cisco before Cisco realizes they don’t need you.”

14. Monday at 8.4x, unprecedented concentration, and the quick-fire

Monday is the quarter’s warning shot: a good print, down 30%, “priced to perfection but growing to perfection” and slightly soft on growth. Post-correction it trades at 8.4x ARR — Jason’s gut-punch to private investors: “you sure your portfolio’s so good if it’s trading at 8.4X?” Rory’s history: for 20 years the median SaaS multiple was ~6.3x on ~30% growth — “when the growth dies, all these businesses become very uncompelling… you have to be just growth bigots all the way.”
The likely Carta data: highest seed and A valuations ever on fewer seed rounds, and concentration everywhere. Rory’s internal stat: in Q1 his entire enterprise-B2B sweet spot totaled ~$12B raised — OpenAI alone raised $40B in the same period, “twice our entire addressable market for us and 100 other A and B firms done in one deal.” It’s not going back: LLMs and defense make capital-intensive industries venture-accessible, companies stay private longer, and mega-funds pitch LPs “this is the only way to access that risk.”
Kalshi quick-fire: Palantir over/under $2T market cap by 2030 (from ~$450B) — both take the under; Jason feels “gravity” comparing to Salesforce at $222B, which Palantir already doubles. Stripe IPO before June ‘27 — Rory declines to call it, noting “aren’t they so post-public?… strongly cashflow positive at huge scale, and there’s liquidity.” Will xAI sue Apple ($100 pays $209 on yes) — both lean yes: the ChatGPT-versus-xAI fight is “existential… both personal and business,” and Apple gets pulled in “like getting named in the lawsuit.” Both name Alex Karp as the dream guest.

No Priors Ep. 127 | With SemiAnalysis Founder and CEO Dylan Patel Chinese Founder Michael on Building FinalRound to $10M in the US