Canva Slahes Growth | Talent Exodus at Google | Revolut's $50B CEO Package | Musk's $55B Terrafab

Summary

• Canva’s reported 2026 growth cut — from 30% entering the year to likely 20% by year-end, partly attributed by CEO Melanie Perkins to AI serving costs — reframes the prosumer-SaaS complex, because the real question isn’t margin compression but “are you thirty on the way to twenty on the way to ten?” Jason Lemkin pegs fair value at ~$12B for ~$4B ARR growing 20% and not decelerating; Rory O’Driscoll notes comparable infrastructure names at mid-20s growth trade at 15–17x NTM because they carry no existential question, so Canva could be $12B or less while exposed, and “$12B and up” only if it disproves the ChatGPT-cannibalization narrative by growing.
• The scariest datapoint isn’t user churn — it’s agentic distribution quietly disappearing. Lemkin’s SaaStr churned both Canva and Notion (“Canva and Notion did nothing wrong at all. Nothing, and we didn’t need them” in the agentic era), and the ad-creative network SaaStr built on its own agent “never once occurred” to use Canva. Layer on Amjad’s line about Airtable — “the era of no-code is over” — and the prosumer tier looks structurally exposed, while Gartner-type data Jason believes puts successful enterprise agentic deployment below 10%.
• The Jeff Dean exit (with three others) and Demis Hassabis stepping back to chairman is a compute-allocation story, not palace intrigue. O’Driscoll’s mechanism: compute given to Google Cloud “turns into 30% operating margins in a day,” compute to Gemini might work, while compute to science is a 5–7-year moonshot that might be indulged “at the 10% level” — so the scientists left. His grade: “Google’s efforts so far are B+, A-. They’re not A+,” with “no impact whatsoever in coding, which is the mother lode that’s feeding the Anthropic beast.”
• Rory urges Anthropic to sprint to an IPO; Jason says an IPO within roughly 60 days would make the AI bets tangible. O’Driscoll: “You’ve pulled ahead of ChatGPT comfortably… to a point where it’s embarrassing. It’s never going to be better… You should go now. You should go fast.” Lemkin says an IPO would make Canva and friends “a distant memory of a bygone era of software.”
• Revolut’s reported CEO package — ratcheting toward ~40% ownership at a $500B valuation — would give roughly 16% of the $200B→$500B value creation, which O’Driscoll calls “abnormally high,” while Lemkin reads it primarily as control and incentive, not merely money. Rory would probably approve this particular package, but only against operational metrics (“If you’re going to give someone 50 fucking billion dollars, you owe it to yourself to spend more time thinking about what you’re getting”); Jason’s categorical counter: “Any investment I’ve made that is not run by a founder is a zero. It’s gonna be a zero in this age” — the baby Elons will get these packages regardless.
• Musk’s Terrafab — $16.8B, with a reported first installment — is vertical integration forced by a decade of supply constraint: “I can’t even get TSMC on the phone ‘cause Jensen’s out there all the time.” O’Driscoll flags it as the position most exposed to any AI-spend slowdown (“the all-in bet is the one that slows down the most the fastest”), with Intel — part of the consortium — reportedly completing its first equity raise since it went public around 1979.
• Data-center NIMBYism, including Ro Khanna’s proposed “data center bill of rights,” may self-resolve through 50-state regulatory competition — but tech earned the hostility. O’Driscoll says locals mostly ask, “I don’t know what I’m getting here,” so developers should guarantee no electricity-cost increase plus a $5K–$10K local dividend; and “if you spend three years saying AI’s gonna kill you all, you shouldn’t be surprised we hate it.”
• Post-Airtable (~$6B blended exit) and post-Canva, Lemkin says it’s “time to be a little extra skeptical of marks,” while the trimming debate stays unresolved. O’Driscoll says statistics would suggest you probably should have sold roughly 70% of the time, “but the holders compound forever and the ones you sell don’t compound at all” — citing Hendrik Bessembinder’s research that less than 1% of companies generate 90% of public-market capital gains. Meanwhile Whatnot ($545M at $20B, GMV doubling toward $16B at a 12% take) shows “there’s gold in the things that aren’t going to be destroyed by AI.”

Deep dive

1. Canva cuts growth by a third — and the real question is 30 on the way to 20 on the way to 10

• Rory’s fact base: Canva did $3B in GAAP revenue last year, entered this year growing 30%, and Melanie Perkins disclosed midyear that they would likely exit the year growing ~20% — attributing part of the slowdown to AI serving costs, since subsidizing users on frontier models was, in Rory’s paraphrase, “costing us a shit ton.” He flags the implied elasticity claim — growth slowed, but might not have slowed as much if Canva had spent more — as something “I’m not sure I fully buy.”
• The competitive map he draws: Adobe at $23B growing 12% on 3–4x revenue; Figma, also public, at $1.4B growing 40%, the fastest; Canva still private at ~$3.6B growing 20%. The meta question isn’t second-order gross-margin compression: “The real question for all these companies, are you thirty on the way to twenty on the way to ten?”
• The mooted fix — an in-house image model 80–90% cheaper than frontier models — draws Jason’s dry rejoinder: “Well, why didn’t they do that last quarter?” And even if it lands in six months, Rory says the question remains whether users would have stayed.

2. Jason’s tell: agents never even suggest Canva

• Jason found the news “kind of depressing” because Canva had seemed to defy AI gravity even as SaaStr churned both Canva and Notion — “not because they’re not great apps; we just no longer had any need for them in the agentic era.” Thirty to twenty in one year: “I think it’s terrifying.”
• The sharper signal than churn: SaaStr built its own ad server and creative-generation network on its agent, and “it never occurred to the agent to use Canva for this. It never once occurred to it.” As buying becomes agentic, distribution can vanish without a human ever deciding to leave.
• The exposure map both accept: prosumer is most at risk because “everyone is ChatGPT fluent,” while Gartner-style data Jason believes puts successful enterprise agentic deployment below 10%. Figma, more enterprise-oriented, at least took its medicine publicly — Dylan was clear gross margins would be impaired by agentic usage, and the stock traded down 20%; private Canva is less exposed to public-market punishment for a slight bottom-line miss, though Rory notes the risk remains.
• Jason’s taxonomy via Amjad’s Airtable quote — “no criticism, but the era of no-code is over”: Airtable was a no-code database disguised as a spreadsheet, Notion one disguised as a word processor, and Canva a no-code way to design. All were “breathtakingly disruptive” pre-AI; the no-code category is slowly winding down as capability moves into ChatGPT.

3. UI disaggregation, and whether anything survives on top of models

• Harry’s adjacent evidence: the president of Uber told him his single biggest fear is disaggregation of UI — “I want a car,” and ChatGPT routes the request to Lyft, Uber, or another provider on price. Jason: “That’s what our agents already do. They just routed us around Canva.” Harry’s conclusion is that ease may not matter; consumer real estate and bundled attention do.
• Rory still sees a possible business on top of models, citing Higgsfield as an aggregator with better UI and billing. Jason’s refinement: its original model-aggregation-for-short-videos business is cash-flow positive but “not an exciting business”; the ~$700M in revenue to date, heading toward $1B, comes from complex video creation — “a harness that allows you to do something that’s very complicated with the models.” He says a lot of it could have been Canva’s, just as a big chunk of Replit and Lovable could have been Figma’s.

4. Should Canva have gone public? Translate the question first

• Rory’s translation: when people ask this, “what you’re really saying is, ‘Oh my God, if we’d accessed that $50 billion valuation in 2021, I would be so gone now’” — it is the fast money, including Blackbird and possibly early Felicis and Matrix, wishing it had exited, not necessarily the founders, for whom this is their life’s work.
• Jason’s devil’s-advocate case for staying private: the founders have already given away most of their shares, Sydney may make retention easier than the Bay Area, and “maybe it is better to hide” — run it like Basecamp/37signals and increase profit sharing — “probably the VCs wouldn’t let you get away with it.”

5. Who actually grabbed the moment

• The ledger of pre-AI companies that made the jump is short: Intercom (“Eoghan… earned every dime”), Windsurf and Cursor, which pivoted fast in part because they were tiny, and Replit, which “was frigging in the wilderness for six years until it added the models.”
• Rory’s key distinction: positioning versus reinvention. Datadog, Cloudflare, and JFrog just sell more infrastructure into “the greatest infrastructure boom in history… Time to make out like a bandit.” Jason’s Twilio image: Jeff Lawson “was holding the boogie board just right, and the wave came in, and he’s flying, and frigging ChatGPT is just tumbling poor Canva side over side.”
• Palantir is the N-of-one on the apps side: 18% to 98% growth, built on real FDEs — people who had spent a decade and a half deploying massive change in the field — plus genuine outcome-based pricing. “No one puts a $2 billion deal on the line for an outcome-based resolution.” Harry gives Palantir a positive “Grab the Moment Award”: it could have chugged along at 20% on government work but instead bet heavily in 2022 and 2023.
• The through-line, from Rory: “there’s gonna be a lot of people paying the bill in ‘26 and ‘27 for a certain amount of hesitancy in ‘23 and ‘24.”

6. What Canva is worth, and what to tell the LP holding it

• Jason’s number: ~$12B — 20% growth at roughly $4B ARR in current public markets, assuming it is not decelerating. Rory’s adjustment: infrastructure names with mid-20s growth and 20%+ operating margins trade at 15–17x NTM precisely because “there’s no existential question.” With the overhang, “it could be $12 billion or even less”; transcend it and “it’s kind of $12 billion and up.”
• The meme risk Jason flags: like the “vibe-code your own CRM” narrative — “one of the dumbest AI memes… Most of the folks that say that have never used a CRM” — once Wall Street decides ChatGPT kills Canva, perception itself becomes existential risk. Rory: “The worst thing in the world to fight is a bad idea whose time has come,” and citing Atlassian, “the only way you prove that you’re not dying is by growing.”
• Rory’s answer to the LP, verbatim: “it doesn’t matter what you think, big guy. You’re in this journey for the next 12 months. Buckle up. Because liquidity will only come at the end of the journey.” Jason adds that after Airtable and this quarter, “it’s probably time to be a little extra skeptical of marks” — and Rory notes that Notion, reportedly ~$800M and growing 70–80% according to numbers he is not sure are correct, shows how wrong the mental buckets can be.

7. Trimming versus the power law — a genuine unresolved disagreement

• Airtable’s ~$6B blended exit, a point attributed by Harry to Dave Samuel of Freestyle, prompts the sell-along-the-way case, which Harry endorses via a billionaire friend: “I never regret making millions of dollars, and I say this through my G650.”
• Jason’s pushback: for a smaller fund needing 5x–10x fund-returners, early exits break the math — “I don’t care what X says. You gotta keep doubling down” — and his own lifetime portfolio analysis broke “roughly 50/50” on sell-versus-hold.
• Rory’s counter with the data: statistics would suggest you probably should have sold roughly 70% of the time, “but the next sentence is the key. The holders compound forever and the ones you sell don’t compound at all” — citing Hendrik Bessembinder’s research that less than 1% of companies generate roughly 90% of public-market capital gains. Harry’s exhibit: Emergence selling Salesforce early — holding “would dwarf all the other outcomes.”

8. Jeff Dean’s exit is a compute-allocation story

• The facts: Dean out after 27 years, with three other prominent people; Demis stepping back to a chairman role; Google losing a couple of hundred billion dollars in market value. Rory’s aside: “it must be extraordinarily validating… to leave as a non-CEO of a two or $3 trillion market cap public company and have the stock go down by a couple of hundred billion dollars. That was what the therapists call validation at a high level.”
• Harry’s pushback — Dean could access Google’s resources by going to Sergey and Larry — draws Rory’s mechanism: compute to Google Cloud “turns into 30% operating margins in a day” because it can be sold to Anthropic; compute to Gemini might yield coding or consumer revenue; compute to drug or physics discovery is a five-to-seven-year moonshot that might be indulged “at the 10% level.” Science is structurally third place, so leaving may make sense. Rory adds, “I doubt they needed” the fundraising PowerPoint: “‘Hi, Jeff Dean, I’m raising money’ would have sufficed.”
• Jason says that if Rory’s premise is right, his experience as an SVP of Adobe’s number-three business unit is relevant: “it sucked… the number three BU is invisible.” If you’re deprioritized and can take your whole team and raise roughly $1B, “I’d check out, man.” Vinod is, in Jason’s uncertain account, leading or co-leading the round — rerunning the OpenAI playbook — and Rory notes that venture appetite for AI-solves-science neo-labs “has never been higher… really a phenomenon of the last two years.”

9. Google gets a B+; Rory says Anthropic should sprint to the window

• Rory’s scorecard: good at selling cloud compute and TPUs to Anthropic, “but they haven’t made any impact whatsoever in coding, which is the mother lode that’s feeding the Anthropic beast… Google’s efforts so far are B+, A-. They’re not A+.”
• The E-staff impasse as he stages it: the CEO asks a Nobel laureate “why aren’t we building a better coding model?” while the laureate thinks “why haven’t we cured Alzheimer’s?” — “a really boring E-staff meeting because we’re just talking past each other.” Rory says the corporate imperative would favor coding and chat competitors because the largest drug company is worth $1T while Google is worth $3T.
• On Anthropic, Rory’s admiration: they feel they’re on a public-benefit mission “while simultaneously making every single correct rational financial move over the last two years” — and he argues that the next one should be an IPO: “You’ve pulled ahead of ChatGPT comfortably… to a point where it’s embarrassing. It’s never gonna be better… You should go now. You should go fast. You should be done.” Jason says that if Anthropic really IPOs within roughly 60 days, it would make the AI bets feel more tangible and make Canva and friends “a distant memory of a bygone era of software.”

10. God-tier comp is emerging as required for some AI transitions

• Jason’s observation from CEOs he advises at $100M-and-up ARR, especially $200M-plus: compensation has split into three bands — regular employees, AI engineers above normal salary bands, and “the one to five superstars” in a god tier, with seven-figure packages, outsize cash, and equity stakes 10x what a late-stage employee would normally receive. It strains kumbaya culture, but Harry argues “you’re not gonna pull off a Palantir and Intercom without a god tier.”
• Rory’s signal math: someone who got $1M of Anthropic stock in 2023 holds $51M now — “a once-in-a-lifetime change” that nevertheless “ripples through the hiring environment across the entire ecosystem,” distorting what everyone thinks is possible.
• For startups, Rory reframes Jason’s “do I accept B-tier AI talent?”: in the PC era you had fourth-tier chip talent because you weren’t building a chip — make the model a complement and be A-tier at UI, implementation, fine-tuning, and your data domain. Harry’s recruiting test is, “Did you get an offer at Anthropic or OpenAI? And what was the offer?” Many of those jobs are watermarking or perfecting a pulsing indicator, so founders should find the “pirates and romantics at the edge” who would rather do LLMs for accounting.

11. Data-center NIMBYism: real friction, probably self-resolving

• Ro Khanna’s proposed “data center bill of rights” gets Jason’s scorn — “That’s exactly what Xi Jinping is gonna say… What a fricking joke” — while Harry calls Khanna “the hero of Silicon Valley” sarcastically and Rory jokes about the “Marxist wolf in sheep’s clothing.”
• Harry reports from people he says are close to the issue that, with 50 states and many counties with water and power, regulatory competition means “this will not be one of the great issues of our time.” Jason’s corrective on tone: “this is such an entitled podcast. Oh, poor Anthropic engineer only made 35 million. Go out to the goddamn Panhandle, no one’s making 50 grand” — data-center jobs are real jobs for poor communities.
• Rory’s practical proposal, drawing on reporting he had read: locals aren’t necessarily saying “AI is awful” but “I don’t know what I’m getting here.” Developers should guarantee no 25% electricity-cost increase and consider a $5K–$10K per-person township dividend. And the self-inflicted wound: “if you spend three years saying AI’s gonna kill you all, you shouldn’t be surprised we hate it.”

12. Terrafab: the all-in bet on never begging at TSMC’s tower

• The scale: $16.8B, with a reported first installment; roughly 3,000 jobs at 10% capacity, potentially 30,000; and an attempt to sidestep TSMC’s queue. Harry argues that the constraint itself is unprecedented: “For a decade I’m not gonna be able to get what I need on a cost-effective basis. I can’t even get TSMC on the phone ‘cause Jensen’s out there all the time.”
• Rory’s frame: boundless ambition plus a record of delivering plus unprecedented capital access, with satellite launching integrated into Starlink proving the vertical-integration playbook — but “if there’s any slowdown in the AI spend, then the all-in bet is the one that slows down the most the fastest, and this is the all-in bet.”
• The era’s tell, per Rory, who explicitly said he had not verified it: Intel, part of the Terrafab consortium, reportedly just completed its first equity raise since it went public around 1979 — a company that funded itself from cash flow through four decades now tapping markets in the CapEx boom.

13. Revolut’s package: “when they say it’s not about the money, it’s about the money”

• The reported structure ratchets the CEO toward ~30% at a $200B valuation and ~40% at $500B, though the package had not yet been put in place. Rory says this will not be the new norm — 90% of public companies aren’t founder-run and “the number of people willing to run a public company for $10 or $20 million a year turns out to be remarkably high.” Many 2021-era pure stock-price packages were later unwound; he would insist on operational metrics — build the biggest bank in Europe and “I would totally say you’re worth the $50 billion.”
• Jason reads it primarily as control and incentive, not merely comp — this is going to 40% ownership, and super-voting has limits: “you can control 99 out of 100 board seats, but if you own 6%, you may get pushed out of your company.” Rory’s compensation-committee test: offer triple-vote shares instead of stock — “he would come back an hour later and say, ‘I also want the money’” — while attributing the Dale Bumpers line uncertainly: “when they say it’s not about the money, it’s about the money.” Jason also cites a story that Nik fought a $20M broker fee on a $400M yacht.
• Rory’s participation-rate math: $200B→$500B is $300B of value creation, and the extra ~10% — about $50B — is a 16% take, “abnormally high.” Jason’s retort that it is below 20% carry earns Rory’s leveler: “the nurses in the fucking emergency room are working harder than both of us.”
• Two logged changes of mind: Rory now accepts unusual founder-control terms as “an acceptable price to pay to incent them to go public” (otherwise “everyone just does what the Collisons do”); and Zuckerberg — who controls his board absolutely — said model-release decisions should be board-level, which Rory called the first piece of control he had given up in 20 years.

14. Founder-run or zero — and dilution nobody is modeling

• Jason’s categorical, worth quoting in full: “Any investment I’ve made that is not run by a founder is a zero. It’s gonna be a zero in this age.” He doesn’t buy that “Jamie Dimon’s lieutenant with his starched shirt and his cufflinks can run Revolut”; Harry’s counter is former guest Nikesh, which Jason waves off with, “when did he join? Two years ago?”
• Rory’s limit: the founder-run thesis may be true at $1M, $10M, $100M, and perhaps $1B, but Revolut is doing roughly $5B in revenue and $1B–$2B in profit — “at some point it becomes not true.”
• Rory calls out Nik’s justification — “I deserve more because the investors, after they give capital, they do nothing else” — as “bullshit… the second half of that statement is true,” but it implies no clear limit: is every next $300B another 10% dilution? If so, “you should pay less for that stock.”
• Jason’s realpolitik: “the baby Elons are gonna get these packages, and it don’t really matter what I think.” PitchBook this week reportedly said returns on outcomes north of $500M to $1B are being compressed by unprecedented dilution — his own seed model went from assuming a 2x effective entry price to 4x. And timing matters: “You do these packages too late, it’s too late.” Revolut appears externally all green, which is when such packages can be put in place.

15. Whatnot, Atlassian’s pop, Loom’s tell, and the SMB bear case

• Whatnot’s $545M raise at $20B delights Rory as proof “there’s more to life than AI”: internet QVC, GMV ~$8B going to $16B at a 12% take — “The minute I heard that story, my response was, ‘That’ll work’… People enjoy that shit.” QVC itself is now bankrupt; eBay sits at $40B–$50B. Jason’s generalization: “There’s gold in the things that aren’t going to be destroyed by AI as well as the things that are being decimated by AI” — study shopping, restaurants, and cars.
• Rory separates the earnings movies: Datadog dipped because its biggest customer — “everybody knows it’s OpenAI and no one says it” — cut spend, compressing 18x forward revenue to 15x; Atlassian was dealing with existential concerns at roughly 3x, nailed the quarter, and moved toward 5x. The lesson for Salesforce, HubSpot, and a public Canva: fix fundamentals or stay “stuck in two and 3X land forevermore until you get Bending Spoonsed.”
• Jason’s stress signal even in the beats: Atlassian cut most free Loom seats, while Canva pushed features into higher-paid tiers — “whenever I see the base getting overly monetized or harvested… it’s a sign of stress in the organization ‘cause no founder wants to do that.” His lingering worry is that agents “really don’t need these seats,” and much of Atlassian’s revenue remains developer-seat-based — “the seat is under permanent assault.”
• On Harry’s provocation — HubSpot at $10B, when does Bending Spoons buy it? — Jason says there should be offers at $12B if it is at $10B today, but he does not believe there are many. The structural bear case on SMB software: low-end AI-native competitors are exploding “like nothing we’ve seen before.” Jason’s first venture investment, Pipedrive, “would’ve taken 40 years to get competitive with Salesforce.” Rory’s PE fix — recruit or infuse new company-building DNA from YC Demo Day — gets demolished: “everyone is just sitting there hitting refresh, hoping these smart YC companies fail so they can hoover them up in an acqui-hire.”

Verification Notes

• The transcript renders several low-end CRM competitor names unclearly; they are therefore described generically here rather than treated as verified entities.