20VC: Canva Slashes Growth: How Much is it Really Worth | Demis Hassabis and Jeff Dean: Talent Exodus at Google | Revolut's $50BN CEO Pay Package | Elon Musk's $55BN Terrafab

Summary

Harry Stebbings and guests debate whether Canva’s deceleration from 30% to 20% growth is a temporary margin tradeoff or a sign that AI-native workflows are disintermediating prosumer creative software. They contrast Canva with better-positioned enterprise tools like Figma and Datadog, while arguing that agents increasingly route around standalone apps and that the winning response is to bundle AI cheaply enough to keep the product indispensable. The second major theme is Google’s AI talent and capital allocation problem: Jeff Dean’s departure is framed as a symptom of a large company optimizing for cloud margins and product priorities rather than moonshot science. The conversation closes on infrastructure and governance—data-center backlash, Elon Musk’s Terafab/terrafab push to escape TSMC bottlenecks, and Revolut’s extreme founder pay package—before widening into late-stage dilution, liquidity traps, and why founder control is increasingly seen as decisive.

Notes

• Canva disclosed $3 billion in GAAP revenue last year and the company is still growing, but the key question is whether 20% growth after a 30% run-rate marks a healthy reset or an AI-driven platform shift.
• AI feature rollout is not free: Rory argues Canva is paying real inference costs and may be subsidizing users with expensive AI while trying to protect growth.
• Jason’s core warning is substitution by agents: his own ad-generation system never even considered Canva, which he treats as evidence that AI workflows are already routing around the product.
• The hosts frame Canva as part of the broader no-code wave, but argue that the era of tools for things humans can now ask ChatGPT to do is fading.
• Rory makes a hard enterprise-versus-prosumer distinction: Figma can hide inside workflow and bureaucracy, while Canva is exposed because users can generate the same thing directly in ChatGPT.
• The strategic challenge for Canva is not just cheap models; it is making AI creation as easy and accessible for its users as ChatGPT is within roughly a 12-month window.
• Rory says the AI transition will punish hesitation: some companies are already in the crevice between before and after, and 2026-2027 will be where delayed decisions get billed.
• Jason argues the app-layer winners are the ones that moved into AI early enough to benefit from the wave, while Canva is being dragged by gravity after adding AI too slowly.
• Palantir is held up as the rare app-layer company that successfully reoriented around AI using outcome-based pricing and a field-deployment model.
• Rory agrees Palantir won by committing fully to the enterprise AI version of its product and pairing AI literacy with deployed personnel.
• Jason gives a live valuation estimate for Canva of about $12 billion, based on roughly $4 billion ARR and 20% growth in current public-market conditions.
• Rory pushes back that companies with similar growth and cash flow can trade much higher if there is no existential question, but says existential risk can compress multiples sharply.
• On liquidity, Rory says private-company holders are locked in until the end of the journey; if the market window closes, secondary sales become extremely hard.
• Jason warns that everyone is already pushing the meme that ChatGPT kills Canva, but he thinks this kind of simple substitution story is too crude for most software markets.
• Rory insists valuation should be grounded in actual growth and forward revenue, not stale private marks like the old $42 billion headline number.
• Jason argues venture math is harsh: smaller funds need concentrated winners, so selling too early can destroy the ability to produce an outlier fund.
• Rory reinforces the power-law argument: one percent of companies generate roughly ninety percent of public-market capital gains, so trimming winners often sacrifices most of the compounding.
• Jeff Dean’s exit from Google is framed by Jason as a signal that this is the moment to leave a high-paid, comfortable role and build in AI.
• Harry provides context that Jeff Dean left after 27 years and that Demis Hassabis and other Google AI talent departures were part of the same story.
• Rory says Google Cloud compute immediately turns into margins, while science-heavy AI and drug-discovery bets are long-duration moonshots with lower probability.
• Jason’s personal example from Adobe is that the number-three business unit felt invisible inside a large public company, reinforcing why top AI teams may want to leave.
• Jason says teams that are not the top priority can now raise directly, so leaving a large company becomes rational if a crew can finance its own plan.
• Rory says venture has never had higher appetite for science-oriented moonshot AI bets, especially for scientists who want to pursue discovery without corporate bureaucracy.
• Jason says an Anthropic IPO would make the AI market feel present-tense and relegate older software wealth creation stories to the background.
• Rory says Google’s AI performance is decent but not top-tier, and the company still lacks a meaningful coding-model breakthrough.
• The internal corporate imperative, Rory argues, is to prioritize chat and coding models over long-shot science projects because Google’s stock cannot afford to miss the near-term AI platform race.
• Jason says elite AI researchers want only frontier work and no longer accept dull or obvious assignments.
• Rory praises Anthropic for combining mission language with rational financial behavior, including a desire to go public early.
• Rory notes that if an AI company is directly building a frontier model, talent quality becomes existential; B- or C-tier engineers are not good enough.
• Jason describes AI compensation as split into three bands: regular employees, AI specialists, and a tiny super-elite of people who can command seven-figure packages and outsized equity.
• Rory says a $1 million Anthropic stock grant from 2023 becoming $51 million is a hiring signal that ripples across the whole ecosystem.
• Jason says frontier talent increasingly expects ‘God mode’ packages, because the jobs themselves can be difficult and unglamorous.
• The Terrafab project is presented as an attempt to create jobs and scale vertically integrated chip manufacturing outside TSMC’s queue.
• Rory says data-center deployment is constrained by power, compute, and political willingness, not just by technical feasibility.
• Harry describes Terrafab as a $16.8 billion effort to sidestep TSMC and build in-house fab capacity.
• Rory says Terrafab is vertically integrated at every level, but that the project is most vulnerable if AI capex slows.
• Jason says the fab plan is driven by long-standing supply-chain limits and the inability to get RAM, chips, or TSMC capacity at tolerable prices.
• Rory mentions Intel inside the Terrafab consortium and says Intel’s reported equity raise would be its first since the 1979 IPO if true.
• Revolut’s incentive plan is described as ratcheting with company valuation, potentially taking the CEO to roughly 39-40% ownership at a $500 billion valuation.
• Rory argues these giant packages are unusual and should be judged by the incentives they create rather than treated as a new market norm.
• He adds that stock-price-only plans often fail because executives can perform well while the market is down and awards never vest.
• Rory says Elon’s own package is easier to defend because it ties pay to explicit delivery milestones across Mars, Optimus, and cars.
• Rory estimates the Revolut structure could give the CEO about 16% of incremental value created in the $200 billion to $500 billion tranche.
• Jason’s bottom line is that founder-led businesses deserve the control premium in this market, while non-founder-run companies are getting zeroed out.
• Jason says late-stage venture outcomes are seeing severe dilution and compressed returns, forcing investors to assume much higher effective entry prices.
• Rory says if he were approving the Revolut package, he would only do it with non-market operational metrics tied to building Europe’s biggest bank.
• Rory compares Whatnot to QVC and the Home Shopping Network, arguing that live shopping has a proven consumer behavior and a plausible digital analog.
• Whatnot’s unit economics are described as GMV of about $8 billion last year, expected to reach $16 billion this year, with a 12% take rate.
• Atlassian is used as an example of a company that eventually delivered and got rewarded by the market after a period where the hosts were skeptical.
• Jason says Atlassian and Canva both made monetization moves that look like stress: removing free Loom seats and pushing features into higher-priced Canva tiers.
• Jason argues prosumer tools like Canva are directly threatened by ChatGPT, while SMB software such as HubSpot faces pressure from strong low-end AI-native competitors.
• Jason says talent-recruiting strategies based on acquihires are now crowded out because hot startups are already vacuuming up teams from around the ecosystem.