20VC: The AI Boom Will Create Enormous Roadkill: Who Wins & Loses | Why Founders Should Never Take Multi-Stage Money at Seed | Why Triple, Triple, Double, Double is Good Enough

Summary

David Frankel makes a broad case that the current AI wave is real but structurally over-exuberant, with bubbles getting bigger, lots of “roadkill,” and a plausible dot-com-style crash somewhere ahead. He argues seed has become crowded and commoditized, pro rata can become founder-unfriendly, and large platform and multi-stage investors increasingly behave like buyers of call options rather than classic seed partners. At the same time, he says venture math is still governed by ownership times outcome size, with secondary liquidity, fund sizing, and DPI all becoming more important in a market where top names can reprice quickly. He is bullish on applied AI and physical AI built on commoditized hardware, skeptical of broad SaaS-apocalypse narratives, and highly focused on founder-team chemistry, recruiting burden, and vertical edge. The discussion also widens into AI’s second-order effects: future disruption of incumbents, possible Chinese competition, photonic computing as a potential Nvidia challenger, and major productivity gains rather than mass unemployment.

Notes

• AI is presented as a massive wave with inevitable “roadkill,” even if the timing of a crash is unknown.
• Frankel explicitly calls another dot-com-style crash likely, but says nobody knows when.
• He frames mega-platform seed participation as call-option behavior, not true early partnership.
• He says pro rata can function like an “original sin” against founders.
• The venture math still comes back to ownership multiplied by outcome size.
• He emphasizes that a 5% stake in a $2.6B outcome can return a fund.
• He says the current hot AI financings do not yet look capital efficient.
• The jury is still out on whether the large-check AI funding model works.
• Large seed rounds are sometimes being treated like insurance rather than classic venture.
• He says smart founders increasingly value a small check as downside protection.
• Seed is crowded and increasingly commoditized.
• Brand, distribution, and portfolio reputation mainly help get a manager to the table.
• He thinks many hot AI companies are still not demonstrably capital efficient.
• The top 500 companies from the last 25 years had a median value of $2.6B, per Frankel.
• He argues many founders underestimate the CEO job, especially recruiting and headcount-building.
• CEO time is heavily consumed by recruiting, not just product.
• He looks for a specific early-team alchemy: salesperson CEO plus magician CTO.
• Interpersonal fit and founder-team composition matter a lot at seed.
• He says 10-year funds can effectively run 18 years.
• His model assumes companies often take twice as long and cost twice as much as planned.
• He distinguishes his approach from momentum investing.
• His stated style is value-oriented seed investing, not fast-turn momentum.
• LPs are increasingly sovereign wealth funds and public investment corporations.
• He says LPs care more about IRR than headline fund multiple.
• He says TVPI can look strong while DPI remains weak.
• His fund’s winners are built on commoditized hardware plus AI.
• He sees the best opportunities as AI layered on already cheap, commoditized platforms.
• Return concentration has been less extreme than people expect.
• He says multiple winners can matter more than a single outlier in a fund.
• He worries that leaders like Airtable and Snyk can be cannibalized.
• Embedded systems are harder to replace than light-weight SaaS tools.
• He believes the “SaaS apocalypse” thesis is overstated for deeply embedded products.
• Every investment is underwritten as a potential Uber or Suno-scale winner.
• He says he does not underwrite deals to mere 10x outcomes.
• His internal bar is that every deal must clear a 10x possibility.
• He uses a strict 10x test to decide whether to invest.
• He likes founders with unusual vertical edge, even “nepo babies.”
• Vertical expertise can be a real moat in commoditized markets.
• He cites TJ Parker as an example of an operator with deep domain edge.
• Suno’s team is framed as being built from prior domain experience at Kensho.
• Secondary markets are unusually liquid for top private names.
• Less IPO and M&A activity is helping make secondaries more important.
• He says secondary pricing can be estimated reasonably efficiently.
• A 25% secondary discount may imply a valuation of 7.5 instead of 10.
• He has seen late board discussions about March rounds only months after December rounds.
• Momentum in secondary markets can foreshadow very fast follow-on rounds.
• He admits he probably sold Uber too early.
• Faster companies like Suno can create much less dilution than hardware journeys like Whoop.
• Speed of execution directly affects dilution and ownership retention.
• At very high valuations, he expects some secondary selling.
• Once a company reaches rarefied scale, the firm may trim exposure via secondaries.
• Many LPs now require minimum check sizes of $50M.
• He views the whole environment as inflated in scale.
• Some fund-of-funds are even selling their whole positions via secondaries.
• Fund-of-funds can be forced sellers even when a manager thinks they should hold.
• He notes some founders step back into exec chairman roles after bringing in a CEO.
• Founder behavior ranges from rational persistence to obsession-driven overstay.
• He says kindness and human presence matter more than most other things.
• He argues phones have become remote controls for our lives.
• AI is already displacing search entry points and top-of-funnel behavior.
• He thinks even dominant AI winners will themselves be displaced over time.
• He says Microsoft has done a poor job on AI so far.
• Google may be in pole position because it already has contextual search infrastructure.
• He expects the biggest AI winners to throw off capital into venture and adjacent investing.
• Top-tier AI returns should spill into the broader venture ecosystem.
• Photonic computing is his candidate for a future Nvidia disruptor.
• He sees a plausible acquisition path for photonics into Nvidia.
• He wants much more U.S. R&D spending, especially after seeing China’s spending on energy research.
• China is spending heavily on energy-efficiency and energy research, in his view.
• He says Chinese biotech investors are traveling to China because the regulatory environment is more permissive.
• He sees China as a source of faster grassroots R&D and licensing activity.
• The seed-stage risk premium was once overstated, but that arbitrage has largely closed.
• If later-stage companies look orphaned despite real revenue growth, he would consider raising a larger fund.
• He says AI has delivered less disruption than hype suggested, except in models and some voice interfaces.
• Consumer AI still feels underdeveloped to him.
• Suno is framed as a Spotify disruptor with the user experience front and center.
• He says Suno’s CTO would swap in a better model without hesitation if it improved the product.
• He was surprised by Suno’s speed of scale relative to older SaaS benchmarks.
• On Uber, he says no one truly knew the outcome in advance.
• He says controversial bets often came from geography or product category, like Coupang in Korea and Shield in defense drones.
• He compares prediction markets to gambling businesses.
• He uses TVPI versus DPI as an analogy for what is real versus what merely looks real.
• Frameworks helped him avoid bad deals, but also caused misses like Klaviyo and Suno.
• Using valuation as shorthand can create false negatives.
• He returns to a simple philosophy: kindness, presence, and attention.
• He is explicit about no phones at dinner and no phone by the bed.
• He thinks autonomy is arriving slowly, then suddenly.
• He predicts kids may not need to drive in five to ten years.
• He sees AI plus compute as a force in healthcare discovery and treatment.