Summary

The stay-private supercycle is “the greatest gift to venture in our lifetimes” — Jason Lemkin’s frame for why Lightspeed just raised $9B across six funds (only ~$2B of it venture/early by Harry’s backward math). The killer comparison: Tesla had to IPO at $1.7B; “the sister company will IPO at 1.7 trillion… it’s 1000x more,” and anyone who chose after 2010 could have a ~70% compound IRR for 15 years — returns the public investor never saw. “It’s playing the growth super cycle bet today. That’s the winning play.”
You can’t value SpaceX at $1.5T — you buy the “EOV,” the Elon Option Value. Rory O’Driscoll’s coinage: $15-16B revenue in 2025, Starlink-driven, ~78x 2026 revenue — “you can’t run the numbers on SpaceX and come up with the 1.5 trillion. You just can’t.” The premium is a call option on Elon finding another trillion-dollar market (Jason’s candidate: space-based data centers — “he’s the only one that can do it”), and the key-man risk is stated raw: “if he were ever to die, the stock gap would be something horrific.”
OpenAI’s user growth is “down to single digits on every measure” at 1.2B users — “we’ve run out of humans on planet Earth” — so the question is whether ChatGPT Robinhoods (cross-sell 800M) or Metas (go to 4B). Rory’s warning is the tradeable one: “there is nothing as terrifying as a high-growth bet that slows down” while private — you flip from growth valuation to cash-flow valuation, and the class of ‘21 bagholders show how that ends. No alarm yet — but monetization is still finding ways to grow — and a taper would be “a world of pain.”
Oracle (-45% from September highs) and Broadcom (-$300B in 48 hours) are the market interrogating who deserves a gross-margin pass. Rory calls Oracle’s RPO pop “a sugar high, a total sugar high” on a capital-intensive build for a customer who “may or may not be able to afford to pay” — no rebound. Jason’s counter: Oracle and CoreWeave (-60% from July) are the IP-less high-octane marginal bets, so they swing hardest, but “we’re still anti-gravity here” — jitters, not thesis break.
Apollo’s zero-return decade call is entry-price math, not a crash prediction: entry P/E barely predicts one-year returns but strongly predicts ten — and Cisco only just reclaimed its 1999 price, 25 years later. The venture read-through: private valuations look cheap “relative to a public market that hits all-time high”; the likely path is a 30% drop somewhere in the next 2-3 years, then a slow crawl — which would leave lofty private marks high and dry.
AI convergence means incumbents “don’t get killed — you get maimed”: logos renew, seats shrink, NRR drifts, and the next YC class defers buying. Cursor’s designer tool threatens Figma exactly this way; Klaviyo hired a co-CEO because “there will be no such thing as the marketing software that got Klaviyo public.” Coding is 55% of enterprise AI end-user spend — but total end-user spend is ~$16B against $400B of capex, and it “has to 3-4x a bunch more times” or the builders have a sad day.
Would-you-rathers with real disagreement: Figma at $17B vs Cursor at $29B — Rory takes Cursor instantly; Jason takes Figma because “there was a lot less stability in these so-called leaders than we thought.” OpenAI $500B / Anthropic $360B / Google $2T — Jason takes Google (the most energized team he’s seen in a decade), Rory takes Anthropic: “very sensibly and boringly converging on profitability, will go public, and will be a very nice public company.”

Deep dive

Lightspeed Raises Multistage Capital

Harry’s backward math on the headline: roughly $2B for venture and early, ~$7B across other vehicles, mostly growth — “the 9 billion’s a bit misleading.” His provocation, via his own tweet: if you’re not playing the big game, do you really matter?
Rory’s LP scorecard for why they earned it: a mega multi-stage manager must show early deals going the distance — Rubrik last year, Navan this year — and concentration in the hot late-stage rounds, “by all accounts, a billion in two rounds of Anthropic… that feels pretty smart right now.” “Tick, tick, insert $9 billion.”
Jason’s seed read: at that scale “you don’t care what you pay for seed… this is why we have 20 or 30 million pre-seed rounds” — seed is an entry ticket to the A, B and C, and multi-stage firms can “swamp a seed business… and just write it off as marketing” (Rory). After Harry cites David George’s numbers — Databricks 7x’d a billion-dollar fund, Coinbase 5x’d one — Jason’s verdict: “seed’s for suckers.”
One genuine dispute: Harry is “consistently surprised by LP appetite” for $50-100M SF seed funds; Jason calls that a myth — the appetite “faded since 2021… I don’t think it’s easy for emerging managers.”

Private Markets Drive Venture

Jason resurrects Rory’s old line, now “true in spades”: “all these leaders not IPO-ing is the greatest gift to venture in our lifetimes.” VCs keep the top-20 compounders for themselves; the public markets never get Databricks, SpaceX, Anthropic or OpenAI. “It’s playing the growth super cycle bet today. That’s the winning play.”
The Tesla/SpaceX pair makes it visceral: Tesla had to IPO at $1.7B; “the sister company will IPO at 1.7 trillion, or 1 trillion… it’s 1000x more” (Jason), and post-2010 SpaceX access meant “a 70% compound IRR for 15 years” — a product public investors were never offered (Rory).
Rory’s mechanism: the public/private choice is relative cost of capital. Elon went public with Tesla after “fairly predatory VC behavior” made private money too expensive; now the public market’s cost of capital feels higher, so everyone stays.
Is late-stage as competitive as 2021 — with Dragoneer raising a $4.3B fund the same week, $13B+ across the two? Rory: ‘21 was treacherous because growth “attenuated and indeed went backwards”; today OpenAI and Anthropic “can soak up 60 billion of your late stage dollars and just keep on moving,” so maybe less treacherous. “The wall of money keeps on climbing up.”

OpenAI Expands Its Empire

Rory dismisses it as the week’s least interesting OpenAI item: a cross-licensing deal that’s “very roundtrippy” — Disney gets ~$1B in equity for letting OpenAI play with its characters — while Disney simultaneously sends Google a cease-and-desist for unlicensed use.
Jason sees a template instead: a three-year deal like the ESPN/YouTube playbook — first mover gets the best terms, “everyone else is gonna pay worse, and we’re gonna ratchet up the terms… It may be the revenge of IP” after an era of “just ripping everybody’s content off.” He quotes Iger approvingly, “even though I’m not sure what it means”: “creativity is the new productivity.”
The open question Rory keeps: in three years, is old-media IP actually worth licensing versus UGC-type content — does Mickey Mouse move 800M users to 880, or not at all?

ChatGPT: #1 App in a World That’s Run Out of Humans

ChatGPT was 2025’s most downloaded app on Apple — Rory checked the decade’s winners: two years of TikTok, two of Temu, Zoom in 2020. His claim: 800M users is exactly what “makes Benioff wrong” on model commoditization — distribution is the moat.
Jason’s counter-fact: growth is “down to single digits on every measure” at 1.2B users, “because we’ve run out of humans on planet Earth.” The question is whether ChatGPT Robinhoods — stays ~800M and cross-sells relentlessly, the eternal fintech movie — or does what Meta did and reaches 4 of 5 billion humans. With no ads and cheap India tiers, Jason doubts Meta-scale footprint.
Rory’s escalation is the episode’s sharpest risk framing: “there is nothing as terrifying as a high-growth bet that slows down” — you move from being valued on growth to being valued on cash flow, ideally not while private. The class of ‘21 shows the ending: “I paid 20, 30 times revenues because it was growing at 100%, and now it’s growing at 8%.” No alarm on OpenAI — it is still finding ways to monetize — but a taper would be “a world of pain.”
Jason’s aside worth logging: maybe Anthropic is structurally advantaged precisely because it lacks 800M-user headwinds — “the enterprise and the back end have just gotten going.”

OpenAI Kills the One-Year Cliff

Rory’s explanation of why the standard four-year/one-year-cliff structure broke: at OpenAI pay scales, the cliff is real money — “if I’m getting $10 million over four years, it’s $2 million if it’s slightly less than 12 months” — so hires pushed back and won. “A sign of the extraordinary sums of money you’re dealing with.”
Jason’s Captain Obvious beat: no cliff makes leaving easier — “perhaps you are creating mercenaries” — but asking a poached engineer to wait 12 months to make a dollar “can be a tough sell in an age of plenty.” They probably made so many exceptions it stopped mattering; Rory imagines the VP of HR surrendering: “Let’s just accept this is the market today.”

Oracle Unwound: Sugar High or Buyable Jitters?

The tape: down 15% Friday, down ~45% from September highs, $12B quarterly capex against $8.2B expected, the bulk going to OpenAI-dedicated data centers.
Rory’s I-told-you-so (he flags he earns it by admitting mistakes): the 30% RPO pop “was absurd, and it’s just been unwound” — “a sugar high, a total sugar high on a huge contract with someone who may or may not be able to afford to pay.” Capital-intensive data-center building for one or two customers is a worse business than Oracle’s free-cash-flow core. His regret: “I shoulda bought those puts.” Rebound? “No.”
Jason takes the other side, pairing Oracle with CoreWeave (-60% from its July high): they’re the weak, IP-less marginal players, so they should swing hardest on jitters — “there’s no reason I don’t think they will rebound… we’re still anti-gravity here.”
Rory’s synthesis makes it tradeable: these are “the high-octane bet on AI” — buy Google for 20%, buy CoreWeave right for a double. The market is an interrogator sorting who has a plan: “Gemini, Google, you keep going… OpenAI, we’re good for it… CoreWeave and Oracle, it’s not clear to me why you’re doing it.” If the capex cycle has two more years of legs, they rebound; if the marginal investment rate rolls over, be scared.

Broadcom Loses $300B — and That’s Not a Crash

Why Anthropic ordered $21B of chips from Broadcom in the first place: “they don’t wanna pay 75% gross margins to Nvidia… if I’m gonna pay full retail, I might as well go buy the designer brand” (Rory). Broadcom is a made-to-order business, not a branded Blackwell — “this is meant to be coals here, dude” — so margin compression is the design, not a surprise.
Perspective on the “crash”: still a $1.6T market cap at a high-teens sales multiple — the first trillion-dollar company was Apple in 2018. “Let me tell you what a real crash is like… it’s not like everything went cheap, Harry. It’s just slightly less expensive.”
Jason’s honest puzzle: “I can’t keep up with who gets the gross margin pass” — Oracle got one until it didn’t, CoreWeave “apparently an entire pass,” OpenAI yes, Meta no. “We’re gonna find out some don’t deserve a gross margin pass.”

Entry Price Matters Most at Ten Years

Rory’s precise reading of the Apollo chart everyone misquoted: entry P/E has little or no correlation with one-year returns, strongest correlation at ten — “when you buy at a high price, it can still go higher… but the probability of making money over 10 years is very correlated to your entry price.” Greenspan called irrational exuberance in ‘96; “shit kept going up for three more years.”
The specimen stat: Cisco just got back to its 1999 price this week — 25 years to break even on the darling bought at the top. And “that nice Mr. Buffett has piled up three hundred billion in cash, ‘cause he reads these data too.”
The venture read-through: private valuations that look attractive are comped “off a public market that hits all-time high.” The likely path isn’t a smooth zero — “things drop 30% at one point in the next two or three years, and then you crawl back slowly” — which would leave some private marks “high and dry.” Growth might save you; it might not.

Cursor Comes for Design — Convergence Eats Every Category

Cursor (Harry initially misattributed the launch to Anthropic) ships a designer UI — tweak web apps, drag-and-drop CSS. Jason’s frame for ‘26-‘27: “massive convergence of categories.” In e-commerce it already happened — marketing, sales and support “have already converged to one agent,” which is why Klaviyo ($1.3B revenue, growing 30%) just hired the ex-co-CEO of Workday: “in 2026, 2027, there will be no such thing as the marketing software that got Klaviyo public.”
The pain point as told, from a builder: “I can find a vibe-coded site… 30% of the last YC class looked vibe-coded to me. I could see the Claude artifacts all over their homepage.” The surprise is that design/code convergence took a full 12 months.
The deeper insight both agree on: “we all want to talk to the same agent” — designers, product, engineers, DevOps. Rory’s generalization: siloed software existed because humans were siloed; if you sell AI that automates an outcome rather than a department’s work, one agent handles prospecting, selling and support. “That’s a really powerful 2026 theme.”
Who wins? Jason refuses the VC-kingmaker fantasy: “it’s gonna be who wants it the most… everyone can copy each other in weeks now, not in months or years.” Momentum says Cursor — Figma took a decade to $1B, Cursor about a year — “even though they’re not designers, they don’t have the base.”

The Maiming Thesis: “You Don’t Get Killed — You Get Maimed”

Why Cursor never bothered building a Replit/Lovable competitor: distraction. Those two together will hit ~$500M in a year; Cursor got there in nine months with what Jason says is probably ~160% effective NRR, selling into enterprise software budgets rather than high-churn prosumer.
Jason’s signature concept: incumbents don’t die, they get maimed — “the old customers don’t leave… but NRR drifts down, and new customers, the kids from YC, defer that purchase because they’re doing enough in Cursor.” Board-meeting version: “We had 15,000 customers last quarter, now we have 15,200. Hooray.” “It’s the one that never quite heals.” Exhibits: Atlassian slowing, GitLab arguably, Mongo fought back but “why isn’t Mongo growing 50%?” given the app explosion.
Rory’s mechanism: CIO priority surveys — AI rises, your item slides from #3 to #6 and doesn’t get funded. That’s the SaaS slowdown. The prescription: co-attach to where AI budget is — sell infrastructure to Anthropic and the JPMorgan AI initiative, not to the people not doing AI.
The market-sizing anchor, from the Menlo data Jason blogged: coding is ~55% of all enterprise AI end-user spend — “the epicenter of the enterprise AI revolution.” Rory’s bigger nut: end users spend ~$15-16B on AI apps while the builders spend $400B making AI — spend “has got to 3 or 4x a bunch more times… otherwise the people investing in CapEx are going to have a sad day.”

UiPath and the Incumbent’s Sisyphean Grind

Jason’s case that Daniel Dines has time: $1.8B ARR, 98% GRR, 107% NRR (down from 140% at IPO), back to 16% growth, stock up 27% this year. The job is making a $2B base buy agentic products — Databricks runs 150% NRR at $5B ARR — “and you’re back to 30% growth… back in the game.” But not infinite time: “get rid of the CEO you brought in so you could relax. Go into Sergey Brin mode.”
Harry offers the Alex Rampell quote — “can the incumbent acquire innovation before the startup acquires distribution?” — and Rory complicates it: acquiring innovation is the easy half. The hard part is two-to-three years of lifting growth from 9% to 11% while “the stock doesn’t give a shit” — “you’ve got to push that Sisyphean rock up the hill for four or five years.” His admiration list: Benioff, Dynes, Aaron Levie — “we’re not gonna roll over and die.”
The line of the episode, on why $2B incumbents can’t declare victory on a product launch: “the market will say, ‘No, you’re not cool. You’re a $2 billion boring old company.’ You gotta make yourself cool, ‘cause 30% growth, that’s what’s cool.” Rory’s reluctant concession to the cliché: “we are refounding the company in the age of AI” is the actual job.

Boom Bets on Hard Tech

Given a choice between covering Harness’s $240M round and Boom, Rory doesn’t hesitate: “anyone who chooses anything other than the supersonic plane has no soul.” Boom — designing both plane and engine from scratch, prototype out with under 100 headcount — gets an order from Crusoe and is raising $300M to sell turbines to data centers. Not crazy: GE and Rolls-Royce already do similar generator work; easier to “plop it on the ground” than propel a plane.
Jason supplies the whiplash history: an “arguably fake” $1B round from airlines with no revenue, a December 2024 crash to ~$500M with a “massive cram down,” now back to $1.5B on the AI lift. And the caveat that matters: “they’ve sold zero of either. No planes and no jet turbines.”
Rory refuses to lump it with SaaS re-acceleration — this is ultra-high-risk hard engineering — but that’s the segue: “three failed launches in 2007 and 2008, that’s probably what SpaceX looked like, too. And if it works, you’re a genius.”

SpaceX at $1.5T: Pricing the Elon Option Value

The chronology stung: last Tuesday they called the $800B secondary rich; Wednesday the $1.5T IPO plan leaked — “you can be wrong by 800 billion in a day.” Fundamentals: $15-16B revenue in 2025, Starlink the growth driver, growth down a touch year-over-year, maybe early-mid-20s next year — call it ~78x 2026 revenue.
Rory’s coined framework: the EOV, the Elon Option Value. Run normal multiples on Tesla and you get maybe $300B against a $1T+ price; the gap is EOV — and Elon “earned it” by pulling Starlink out of the hat, turning a rocket company into a communications company. “You can’t run the numbers on SpaceX and come up with the 1.5 trillion. You just can’t… once you’re lucky twice, you’re good. Three times, you’re fricking amazing.” The flip side, stated raw: “if he were ever to die, oh my God, the stock gap would be something horrific.”
Jason’s candidate for the next rabbit: space-based data centers, floated ahead of the IPO — “if all that Oracle and CoreWeave and Nebius revenue goes up in the sky… he’s the only one that can do it. And I don’t think he’s joking.” Plus the Bezos-grade superpower: “all of Elon’s dates are wrong… and everyone gives him another four to five years.”
Can it clear? A typical 8% IPO dilution means raising ~$120B — “is there 120 billion of risk capital that says what I really need in 2026 is some 70x run-rate revenue space investment?” Rory suspects a walk-back “to the merely outrageous.” Jason’s banker fix: Google anchors with $10B (already ~10% holder — TPUs in space), Fidelity $2B, maybe NVIDIA $5B, “and all of a sudden you start to panic that you’re not gonna get your shares.” Rory: “It’s all narrative all the time” — and Jason just demonstrated how fast narrative flips. Coda: Thiel fired Elon at PayPal but accelerated all his stock, then funded a dying SpaceX “in an hour” — Jason’s lesson, “in today’s age of extreme greed, be kinder.”

Would You Rather: The Closing Tape

Figma at $17B vs Cursor at $29B: Rory answers in one word — Cursor. Jason takes Figma: “we’re gonna look back in 24 months and see there was a lot less stability in these so-called leaders than we thought. It’s not that they’re gonna go to zero, but we are so early in AI.” Rory declines to argue: “that’s the beauty of money… in the end, you’re right or wrong on the numbers.”
OpenAI $500B / Anthropic $360B / Google $2T: Jason takes Google risk-adjusted — the Cloud team is the most energized he’s seen in a decade of working with them; “wouldn’t have been true a year ago.” Rory needles the timing: the stock’s already up 60%, “you may be the wrong side of that trade. You should’ve taken the quick pop.”
Rory lands on the pure play: at $170B he’d take Anthropic “all day, every day”; after hearing the $360B live price, he still goes Anthropic — “they’re being way more sensible than OpenAI… very sensibly and boringly converging on profitability, will go public, and will be a very nice public company,” while OpenAI “is more likely to just get caught in the middle with commitments it can’t meet.”

SAM 3: The Eyes for AI — Nikhila & Pengchuan (Meta Superintelligence), ft. Joseph Nelson (Roboflow)How Chinese AI Founders Navigate Silicon Valley: Manus and US-China Tech