Summary

Anthropic’s $13B raise at $183B post is cheaper than it looks — Rory’s math: revenue went $100M two years ago → $1B run rate entering this year → ~$5B now, maybe $8-9B by year end; if 10x growth decays “only” to 3x, FY26 GAAP revenue lands near $20B, so buyers are paying 8-9x forward. “I went in saying, aren’t those guys so silly paying so much… you look at the numbers, you go, maybe those guys are being quite smart.” The round was ~5x oversubscribed — every growth fund faces the imperative of owning one of the two LLM leaders or making “a big balls call that this isn’t going to work.”
Rory predicts a Scale AI writedown on Meta’s balance sheet: Meta put $14B in, the VCs promptly took the cash out, and what remains is “an empty husk” whose data-labeling product lags the frontier (dog/cat labels vs advanced bio and math). “At some point the auditors are going to say… we’d like you to take a writedown — that’s going to be the entertainment factor back end of this year.” On Cliff’s $20-30B “pack of mercenaries”: Rory says “my gut is this is more like the metaverse than WhatsApp” — though “he’s won already.”
The year-2 renewal problem is the AI sector’s hidden tell. Canva’s Cliff describes his own enterprise buying as “spray and pray” — an extra $10-50M budget, four coding tools at once (Cursor leading), pick any two LLMs — consolidating within 12-24 months: “100% there’s going to be consolidation.” Early-adopter syndrome pulls revenue forward; the real test is crossing from $100M to $1B against mainstream, middle-America distribution.
The $3-4T capex math hinges on the token tax. Rory: $4T at a 20% ROE needs $800B of annual profit — “another four Microsofts” — when pre-AI cloud was only $150-200B of total revenue; “I don’t see where the macro works,” and he keeps an eye on Nvidia puts. Cliff already pays ~10% of revenue to model providers but calls it “a marketing cost, not a long-term COGS” (image costs 4¢ → 0.02¢ inside six months, 90% of queries going on-device/self-hosted). Jason’s counter: usage can go up 100x — Gamma codes every presentation from scratch, and Salesforce is “0.1% AI penetrated.”
Private markups look inefficient: Lovable went $1.8B → $4B in 60 days (Vercel’s $9B reportedly in the works), while Replit closed at $3B just 13 days after Lovable’s $1.8B on basically the same ~$100M ARR — “I don’t think either of those deals were perfectly efficient.” Rory’s taxonomy for two-month re-rounds: genuine new information (Anthropic tripling in four months qualifies), mispricing, or pure validation cascade.
SaaS is back — “reports of the death of SaaS were greatly exaggerated.” Snowflake, likely MongoDB (+40-45%), Box, Elastic, Okta and even Zoom beat — “like Madonna coming back from the dead” — as roughly half the public B2B leaders finally catch an AI tailwind (Salesforce not yet). Jason’s bar: with a billion-dollar installed base, “it’s kind of sinful if you haven’t reaccelerated by the end of 2025 — you kind of failed as a founder.”
Canva is IPO-gearing at ~$4B revenue, ~40% and reaccelerating — only ~20% of it attributed to AI — with $1B+ cash, 8 years profitable, Fidelity anchoring an all-secondary round, and publics now paying more than privates (Figma at 17-30x vs Canva’s ~10x). Its SEO engine is migrating intact to LLMs: #1 productivity app on ChatGPT, its 5th most-referred domain, ChatGPT-sourced image uploads up from 0.02% to 5%+ in 18 months.
Pay the tax and get off the stupid train: The panel’s rule is that two positive data points over time carry near-infinite information versus a new deal’s one — Rory passed on Box in early 2010, woke up nine months later and bought in anyway. In AI the strongest signal is a founder whose product “has evolved three times” and stayed ahead: “this guy has a survivor gene. Run, don’t walk.”

Deep dive

1. Anthropic at $183B post is 8-9x forward — “maybe those guys are being quite smart”

Rory’s back-of-envelope, built only on “facts, things we know today”: revenue went $100M two years ago → $1B run rate entering this year → ~$5B now, maybe $8-9B by year end. GAAP revenue roughly averages opening and closing ARR, so ~$4-5B this year; if 10x growth decays to 3x, next year’s GAAP lands near $20B — meaning buyers at $183B pay 8-9x FY26 revenue. “The stunning thing with this growth rate, if it persists — big underline — is you’re only buying in at 8 to 9 times next year’s revenues if the growth lasts.”
His honest reversal is the best part: “I went in saying, oh, aren’t those guys so silly paying so much — and you look at the numbers, you go, maybe those guys are being quite smart.” Whether momentum persists is “the billion-dollar question.”
The round (started at $5B, moved to $10B, closed at $13B) was ~5x oversubscribed. The structural read: every growth-stage firm faces a corporate imperative — the two largest market-cap creations of the last 3-5 years are the two LLM labs, so “we’re either taking a big balls call that this isn’t going to work, or we need to get one of those.” Cliff’s deadpan from Canva: “That’s barely 10x for us. We sound very cheap.”

2. Canva’s secondary: Figma “threw a cat amongst the pigeons”

Canva priced its round before Figma’s IPO and will close the year “very close if not at $4 billion, growing close to 40% and reaccelerating.” Fidelity anchors as the largest check. It’s all secondary — $1B+ cash on the balance sheet, profitable for 8 years — and the Figma pop dried up sell-side supply, leaving an oversubscribed round short of shares: “a high-class problem, but you don’t want to disappoint people.”
The scar tissue: Canva was worth $40B in 2021 at a 50x revenue multiple, marked down to $26B in 2022. Cliff’s confession — “it took discipline to take the 50x, because we had people coming in at higher multiples and even we were like, this is batsh*t crazy.” Employees get coached with a spread of public comps, not a single Figma reference point.
Harry’s framing, from an old post — “Wall Street is the madman in the back seat”: finance swore Canva was worth 50x, then 20x, then 10x, wrong “literally by 5x.” The dirty little secret: “in the end, the thing that bails out our incompetence is your growth rate” — companies that compound for five or six years “can cover a multitude of sins.”
A genuine change of mind on capital: Cliff ran Canva “to the bones” for minimal dilution at the highest price each round — “a high-risk maneuver, and I don’t recommend this to founders anymore. Be a bit overcapitalized.” Cliff argues against the wise-owl instinct: most founders are happier with a bigger balance sheet, “there will be a shakeout in the next two years — if you’ve pissed it all away in performance marketing, shame on you.”

3. The AI reckoning arrives at year-2 renewals

Cliff’s worry about AI startups mirrors failed “Canva for X” products: early-adopter syndrome pulls forward revenue from the Product Hunt/X crowd, and some products stall at the chasm. Canva’s edge was crossing to “the mainstream middle America” — “distribution at scale is a lot harder” — and “crossing from $50-100 million to a billion dollars in revenue, that’s a big leap.”
Rory’s nagging suspicion, put to Cliff directly: year-2 ARR renewal rates for AI products will be pretty low — “AI product-market fit is getting covered over a little bit by AI enthusiasm.” Cliff confirms from the buyer’s seat: Canva’s approach was “spray and pray” — an extra $10-50M budget, four coding tools at once (Cursor “leading the pack”), pick any two of Gemini/OpenAI/Anthropic — consolidating over 12-24 months. “100% there’s going to be consolidation down from year-two renewals.”
On Canva’s own reacceleration, Cliff attributes only ~20% to AI. Ninety percent of acquisition is organic, and the discipline is refusing “insular thinking and treating your user base like a wet tea towel that you need to wring out” — the mantra is “workhorses, not gimmicks.”
The organic engine is porting to LLMs: Canva is the #1 productivity app on ChatGPT and its 5th most-referred domain, with ChatGPT-sourced image uploads up from 0.02% to over 5% in 18 months — “anything we’re losing on SEO is translating to LLM SEO,” and yes, a team deliberately optimizes for it. Harry: “this whole thing about SEO being dead is stupid — it’s dead for folks that don’t have a brand.” Harry’s caveat: product companies are fine; “if you’re a mid-tier review site and you just get scraped and summarized — well, thanks for playing.”

4. Likely Statsig at $1.1B: “perfectly engineered to check everyone’s boxes”

Harry’s gut on OpenAI buying likely Statsig — $1.1B in stock for a $75M-ARR company, flat to the round Iconiq led in May: “f*, that’s cheap.” The panel mostly disagrees that anyone’s hurt: with preference, a late-stage investor is price-indifferent, and Iconiq effectively rolled ~$100M into OpenAI stock — the only way in, since leading Anthropic’s round bans them from OpenAI directly.
The flat-to-last-round price is the tell — “not 2x, not less… perfectly engineered to check everyone’s boxes.” The CEO (VJ — likely Vijaye Raji) is described by Harry as “number three at OpenAI,” with Harry saying he will run it with likely Fidji Simo and pushed the CPO aside. “I’d rather run my own company… but for 95% of people, this might be a quick upgrade without risk.” Only the angels, who wanted to play the card longer, might grumble. Cliff, offered a hypothetical $100B of OpenAI paper for Canva: “I’m not doing it… we’re cutting our own course. We’ve had acquisition offers.”

5. Meta/Scale: a mercenary pack and a $14B writedown in waiting

Rory sees “two shoes dropping, and they both feel like exactly the shoes you expected.” First, people: “it’s hard to give people $100 million, and then give someone else a billion dollars, and then give someone else $10 million, and then have them all work together.” Second, the asset: data labeling has evolved from “this is a dog, this is a cat” to training data for advanced bio and math — and a Meta whose “ass is on the line to deliver” will buy from the best (likely Surge AI’s likely Edwin Chen, likely Mercor, Turing), “not from what is now our biggest investment just because that told me to do it.”
The third shoe is his sharpest call: Meta put $14B into Scale as if it was worth $14B, the VCs promptly took out that $14B, “leaving Scale as an empty shell… At some point the auditors are going to say, hm — do you really think the empty husk of Scale is worth 14 billion? We’d like you to take a writedown. That’s going to be the entertainment factor back end of this year, early next.”
On Zuck himself, Rory refuses the melodrama — “he can cry into his $200 billion… he’s won already.” The base rates: big bets that worked (WhatsApp), small ones that worked brilliantly (Instagram, “best acquisition of the prior decade”), big flops (the metaverse). “My gut is this is more like the latter than the former. I could be wrong.” Cliff’s cultural read: “he’s given 20, 30 billion to a pack of mercenaries… some of them are going to fall in battle, some of them are going to quit” — and Cliff notes that, despite his old anti-mercenary instincts, sometimes that’s the tool for the job.

6. Two-month markups: new information, mispricing, or validation cascade

Lovable jumps $1.8B → $4B in 60 days, with likely Vercel’s $9B reportedly in the works. Rory calls it “a bit of both” — real execution plus “excess capital desperately trying to find a home”; “this gravy train isn’t going to end immediately. It will start cooling off at some point.”
Harry defends the mark with numbers: Lovable is at $125-130M ARR against a $175M year-end plan and should finish at $185-200M — out-accelerating plan — and at a 2-2.5x next year, $450-500M “is it that nuts?” Rory’s pushback: VCs should have modeled that 47 days ago; on the public side “you miss by 1% and you get your head cut off.” But he concedes the logic cuts both ways — beat plan and a step-up is justified — and Anthropic is the clean case: revenue tripled in four months, reaccelerating at a scale “probably no one imagined” on Cursor and Claude Code.
His full taxonomy of the two-month re-round: (1) genuine new information, (2) mispricing — “either the second round is too high or the first round’s too low,” and if Cliff’s “who gives a f* whether it’s two or four if it’s going to be a $20-50B company” is right, the first round underpaid, the private version of the IPO-pop debate — or (3) validation cascade: “oh my god, Sequoia were willing to do two billion… now I want to get in at four.”
The efficiency counterexample: Lovable closes at $1.8B on July 17; 13 days later likely Replit closes at $3B on basically the same ~$100M ARR — “one’s worth three billion because it’s marked up by Andreessen, the other’s worth about two billion because Accel wants the deal. I don’t think either of those deals were perfectly efficient.”

7. “Reports of the death of SaaS were greatly exaggerated”

A big week in publics — Snowflake, Box, Elastic, Okta and Zoom were highlighted (“Zoom beat — that’s like Madonna coming back from the dead”). Jason’s spitball: half the public B2B leaders are finally getting an AI tailwind. Salesforce has the demand but hasn’t seen it; Atlassian, Dropbox, Asana aren’t there yet; the database name (likely MongoDB — “Dev is an extraordinarily good CEO”) should crush it because “every time I spin up a new vibe app, I need two or three databases.”
Jason’s bar for the incumbents: “if you have a billion-dollar installed base, you have a distribution channel… it’s kind of sinful if you haven’t reaccelerated by the end of 2025 — you kind of failed as a founder. You had 18 months.” Otherwise “we have to give up on all the public guys and bet on the Canva and Databricks and Anthropic.”
Harry deflates the euphoria: the stock that jumped 40-45% just returned to a growth rate (~24% GAAP YoY) and multiple it held two years ago. The mechanism is expectations, not fundamentals — “everyone got into the oh-my-god-SaaS-is-dead… if your expectations are low and you just do moderately well, you can have a 45% jump in a week.” Hence the Mark Twain line: “reports of the death of SaaS and software were greatly exaggerated.”

8. Why IPO at all — and why not direct list

Canva is “gearing up to be an IPO-ready company” (Kelly, who led Zoom’s IPO as CFO, just joined). Rory’s challenge — $1B cash, profitable, tender offers available: why bother? Cliff’s three reasons, the middle one biggest: capital access, employee liquidity after 13 years (“annual secondaries are pretty janky, and in some jurisdictions downright impossible”), and publicity they’d honestly rather not have.
The kicker: public capital is now cheaper — Canva gets ~10x revenue privately while “the fine folks at Figma are getting between 17 and 30.” Cliff’s explanation: crossover funds keep 80% of capacity in publics, so privates chase a smaller pool. Rory’s systemic point: it’s absurd that ordinary people must “pay 2-and-20 and enrich the middleman like us” instead of 50 bps to Fidelity — and Canva’s 240M monthly actives are exactly Figma-style retail demand.
On direct listings Cliff has done the homework and passed: “none of them have really nailed it over that short term.” Harry’s flip: “that’s the freaking point” — no one wants to say it, but if you get the direct listing totally successful, the people buying don’t make any money. Had Figma direct-listed it opens at $36-40, not $75. And the theory-vs-practice truth: “when you’re the guy on point with your life’s work on the line — do I want to be an experimental baby on the biggest day of my life, or do I just want to land this freaking plane?”

9. The $3-4T capex bet hinges on the token tax

Jensen says Blackwell and Reuben scale into a $3-4T AI infrastructure opportunity over five years. Rory’s macro: $4T at a 20% return on equity needs ~$800B of annual profit — “you’ve got to believe you’re going to create another four Microsofts, another four Facebooks… it feels deeply lofty to me. I don’t see where the macro works” — though “it’s hard to argue against a guy who built the most valuable company on the planet.” His hedge is literal: “there are Nvidia puts that I keep my eye on.”
Jason takes the other side: Benioff says the Salesforce base is “0.1% AI penetrated,” implying ~$200B of attach on a ~$50B business — “it’s hard not to see everything easily being 100x bigger than it is today. We just started… I do think Jensen and Sam Altman have a pretty good sense of it.” Cliff confirms billions of AI usages per month in Canva, accelerating — “but they’re all getting distilled and run on device a lot more.” Jason’s side-bet: the energy demand “is going to rapidly accelerate our shift to green energy, particularly nuclear.”
The load-bearing variable is the software industry’s model-provider tax. Notion’s gross margin went 90% → 80%; Cliff says Canva is already at 10% and expects it to fall — the eventual model is to route 90% of queries on-device or self-hosted, reserve frontier APIs for premium queries, and a 4¢ image becomes 0.02¢ within six months: “we view those upfront costs as a marketing cost, not a long-term enduring cost of goods.” Rory’s implication: pre-AI cloud was $150-200B of total revenue; if AI spend merely matches cloud spend, that’s ~$200B of revenue and ~$100B of profit — “you wouldn’t want to spend $4 trillion to make $100 billion of profits.” Something well above today’s usage is required.
The bull case is that token consumption can rise orders of magnitude — Gamma “is coding every presentation from scratch,” and Canva Code (already 20M active users) “could be Lovable prime if you wanted, but you’d use a hundred times the tokens.” Cliff’s answer is architectural and commercial at once: build Canva’s own foundational model so a presentation doesn’t require coding a website each time, launch a “whole slew” of AI products in October, and move to a unified credit model with hybrid seat-plus-consumption pricing — one marketer deploying tens of thousands of content pieces “can’t be $20 a seat.” On $10K/month coding-agent budgets (Farhan at Shopify’s number): not per-individual across 2,000+ engineers, “but we’re not price sensitive around that at all.”

10. Pay the stupid tax: two data points beat one

The pre-show thread Harry surfaces: he saw Riverside at seed, passed, and has refused every round since; same with Revolut — “it’s a bit embarrassing as an early-stage investor to buy Revolut off Goldman Sachs.” Jason and Rory’s rule is that “the difference in information content between two data points over time, both positive, and one data point where you have no calibration is almost infinite.” Rory’s specimen: passed on Box in early 2010, and nine months later “I literally woke up and said, what’s the dumbest thing I did all year? I didn’t do that deal. And I went down and did it.” The discipline: “that is the tax you pay for being stupid. Pay the tax and just get off the stupid train.”
Does this survive AI’s revenue-sustainability problem? Rory says it’s a separate risk that hits new deals equally — and AI adds a stronger signal: when “the product has evolved three times… and the founder’s been able to evolve it — oh my god, this guy has a survivor gene. Run, don’t walk.”
On follow-ons, Cliff’s observation from the cap table: Canva’s best-performing investors were early funds that “realized holy sh*t, we’re on to something” and raised SPVs to compound up the value chain — “if you’re on a winner, keep betting on that winner.” Rory: the outside-led up-round right after yours “always feels expensive — that’s the round you should do every dime,” citing Peter’s (surname not audible) line that it’s the strongest positive signal, consistently underestimated. Cliff’s caveat: distinguish price inflection from company inflection — “I’d rather pay up for the 800-to-a-billion where there’s real company inflection.” Rory’s closer: concentration gets you the last dollar of outperformance, but “in the end, you still have to be vaguely good at picking, unfortunately.”

Verification Notes

Raw captions do not establish which Jason identity is speaking, and several normalized entity names remain likely identifications; the transcript labels were marked [Speaker?] where necessary.

Salomon in China with 殷一 and 欧迪: Trail Running to Salomon Girl Episode 153 - September 5, 2025