Podcast
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
Startup Funding | The Pitch
- Canva Slashed Growth Because Frontier AI Was Too Expensive
- Canva cut 2026 growth from 30% to ~20% mainly because AI serving costs are high when using frontier models.
- Rory notes Canva is building an in-house image model to reduce costs and regain margin if executed quickly. (Time 0:05:27)
- Agent Interfaces Threaten Prosumer Creative Apps
- Prosumer creative tools like Canva are uniquely exposed to agentified LLM interfaces because users can generate designs directly in ChatGPT.
- Jason argues agents never suggested Canva for his workflows, so UI bundling to models displaces these apps. (Time 0:07:40)
- Reduce AI Costs By Building Your Own Domain Model
- Do build or fine-tune proprietary models if third‑party frontier models make your unit economics untenable.
- Rory highlights Canva bought its own image model to cut per-request costs ~80–90% versus buying frontier tokens. (Time 0:08:55)
- Winners Were Those Who Embraced The AI Pivot Fast
- Some companies successfully ‘grabbed the moment’ by shifting to AI-first product offerings, not just incremental features.
- Rory and Jason cite Palantir and Replit as examples that pivoted and reaped outsized growth. (Time 0:21:12)
- Jeff Dean Left Google To Focus On Big Scientific Moonshots
- Jeff Dean left Google after 27 years to chase scientific discovery and moonshots outside corporate constraints.
- Rory suggests senior researchers prefer focused science bets over being deprioritized inside a big cloud-first company. (Time 0:34:36)
- Google Must Choose Between Market Moves And Long Term Science
- Google currently faces a tradeoff: prioritize short-term consumer/coding models that drive market value or long-term scientific research with uncertain payoffs.
- Rory warns investors demand coding/chat competitiveness over far-future science. (Time 0:40:54)
- Compete By Specializing Not By Trying To Build Every Frontier Model
- Do hire A‑tier product, UI and fine‑tuning talent rather than insisting you build a frontier LLM in-house unless that’s your core advantage.
- Rory recommends specialization: excel at fine-tuning and domain data for model-driven features. (Time 0:45:44)
- Startups Must Budget God Tier Packages For Top AI Talent
- A ‘God Tier’ of AI engineers now demands seven‑figure packages and outsized equity, forcing companies to create elite compensation bands.
- Jason describes mature startups budgeting multiple seven‑figure God‑tier hires to secure their next-generation product. (Time 0:45:45)
- Data Center NIMBYism Could Slow AI Infrastructure Rollout
- Local political backlash and NIMBYism are emerging bottlenecks for new AI data centres, raising permitting and community benefit demands.
- Jason notes some counties will welcome the jobs while others will ban builds, so deployment will be uneven. (Time 0:49:19)
- TerraFab Is Elon’s Bid To Break TSMC Supply Constraints
- Elon Musk’s TerraFab aims to vertically integrate chip fabs to bypass TSMC supply constraints and secure long-term silicon capacity.
- Rory says Musk has capital and track record to attempt unprecedented vertical builds, but it’s an all‑in bet sensitive to AI spend. (Time 0:54:48)
- Massive Founder Packages Are Tools For Control Not Routine Pay
- Outsize founder incentive deals (like Revolut’s reported package) are rare, high‑cost tools designed to retain founder control and motivate long-term vision.
- Rory warns boards should tie payouts to operational performance, not just market cap jumps. (Time 0:57:13)
- Prioritize Founder‑Led Teams In High Uncertainty Businesses
- Avoid relying on non‑founder leadership for mission-critical scaling; Jason asserts many post-founder-managed companies in his portfolio become zeros.
- He recommends LPs prioritize founder‑led teams when backing high-uncertainty growth plays. (Time 1:06:03)