Summary

Seed is crowded, not dead. Frankel’s analysis puts the median of the top 500 companies created over 25 years at $2.6 billion; owning 5% of one can return the fund, while even a $500 million outcome remains exceptional. The edge is patience: wait until a founder makes you think, “I have to be there,” often somewhere “off-piste.”
Triple, triple, double, double can still be a venture path if the company is genuinely compounding. A fund advertised as 10 years may take 18, and Frankel argues investors miss value by demanding a $1.5 million ARR company quickly reach $10 million or $15 million; retention, account expansion, DAUs, and execution may expose traction before headline revenue does.
This is the wave of their lives—and it will produce enormous roadkill. Frankel sees OpenAI, Anthropic, and SpaceX as highly likely to become era-defining platforms, but says “like 95% are not gonna be there” and another dot-com-style crash is definite: “It is not a question of if; it is a question of when. Nobody knows.”
Mega-platform seed money can function as a call option, not necessarily committed partnership support. A junior investor may leave, the company may miss the fund’s threshold, and its follow-on mandate can disappear; for the 95% outside the breakout cohort, a patient seed firm can function as a cheap “insurance policy.”
Price and ownership still matter mathematically. Frankel says uncapped seed notes “suck,” yet he has never rejected an extraordinary founder merely because only 1–2% was available. He calls pro rata “almost like the original sin”: useful when others receive it, but fundamentally a call option against the entrepreneur.
Small-fund discipline sacrifices quantum for multiples and DPI. Frankel acknowledges that following every branded round in Uber, Coupang, Shield AI, or Suno probably would have increased absolute gains, but doubts it would have improved fund multiples. Stebbings counters that “a billion-dollar valuation is the new Series A”; Frankel says that is a top-200-or-300 momentum game requiring the ability to “run for the exits.”
Secondary markets now make liquidity an active portfolio decision. In top private names, positions may trade at the latest round price or even a premium; selling 20% of a winner can return 25% of a young fund while preserving 80% upside. Stebbings’ sharper framing is cash velocity: certainty today can beat waiting six-and-a-half years for a possible double.
Today’s AI platforms will also be disrupted. Frankel calls Google a net winner, Microsoft’s AI “second rate” versus the top tier, and disruption of OpenAI and Anthropic “unequivocal,” with an “excellent chance” it comes from China. Longer term, he believes photonic chips could disrupt Nvidia—or be acquired by it—while AI enables sub-10-person companies, productivity gains, and human-plus-AI services rather than mass unemployment.

Deep dive

1. Seed still works because modest ownership can return the fund

Stebbings’ challenge is structural: $50 million–$100 million funds are too large to collaborate through $100,000–$250,000 checks, yet too small to lead today’s $8 million–$10 million seed rounds. Frankel agrees seed is brutally crowded, but rejects the conclusion that the strategy is broken.
Founder Collective’s analysis found the top-500 median at $2.6 billion across companies created during the past 25 years, with fewer than 100 sustainably above $10 billion. At 5% ownership, one median outcome returns the fund; even $500 million “is incredible still.”
The seed edge is behavioral, not industrial scale: “You can wait and wait and wait and wait,” then meet a founder who triggers, “I have to be there.” Frankel describes finding another Uber, Suno, or Shield AI as “a drug,” and says many of the best opportunities remain “off-piste.”
Founder Collective can still invest $500,000 to $3 million, and continues to find $3 million–$4 million rounds. Frankel also sees little evidence that the hottest AI companies raising enormous sums are capital-efficient: “The jury’s out on whether that’s gonna work still.”

2. Founders are plentiful; entrepreneurs and founding alchemy are rare

Frankel worries startup formation has become “du jour”: there are more founders but fewer entrepreneurs. The distinction appears when conditions deteriorate—fortitude, the ability to energize people, and the willingness to climb an increasingly steep managerial learning curve become decisive.
A CEO must evolve into a recruiter and organizational builder. Suno’s Mikey Shulman told Frankel that 30%–40% of his time goes to recruiting; decades earlier, Jeff Bezos said 50% of his time was “bums on seats.” Frankel’s conclusion: “That’s the entrepreneur’s journey.”
Stebbings argues investors wrongly reject companies because a weaker co-founder may disappear within three years. Frankel concedes the logic but still seeks CEO–CTO magic—“the CTO to be a bit of a magician and the CEO to be a good salesperson”—plus alignment, trust, and complementary differences.
Genuine founding “alchemy” may have appeared only four or five times in Frankel’s career. His preferred founder psychographic remains youth-like energy, intensity, focus, and intelligence; it can persist for decades, but the startup journey requires “so much energy” that experience alone cannot substitute for it.

3. Triple, triple, double, double can still be a venture trajectory

Stebbings says he rejected a company projected to grow ARR from $1.5 million to $5 million, then $15 million and $30 million, because perhaps reaching $70 million after four or five years no longer felt fast enough. Frankel’s answer: “These 10-year funds are taking 18 years.”
SeatGeek, invested in during 2010, became a top-three global ticketing business while Founder Collective retained every share. Frankel’s lesson is that promising companies sometimes take twice as long and cost twice as much; demanding an immediate jump from $1.5 million to $10 million or $15 million creates neglected opportunities.
Revenue is not the only traction. A customer spending 4× more than a year earlier, strong retention, or rising DAUs may reveal execution that outsiders miss. Frankel sees potential in abandoned “seed-plus” rounds where large funds have moved on; Stebbings recalls Bullpen pricing precisely that risk and extracting aggressive ownership.

4. The AI boom will mint giants and leave 95% as roadkill

Frankel’s historical framing is blunt: “The bubbles get bigger. This is the wave of our lives.” Internet, SaaS, mobile, and AI are not comparable in scale, but repeated claims that “this is different” do not repeal the base rate for enduring outcomes.
OpenAI, Anthropic, and SpaceX are, in his view, highly likely to become the Metas and Googles of this era. Yet venture still resembles Hollywood: “Like 95% are not gonna be there,” even though the survivors may permanently change technology’s trajectory.
Stebbings argues that the mega-platform seed model offers more capital at a higher price, often with a junior investor who stays out of the way—exactly the product he says founders want. Frankel’s rebuttal is champion risk: that investor may leave, lose mandate, or be unable to persuade the partnership to invest another $5 million–$10 million.
Once growth misses the expected $1 million–$15 million ARR ladder, a mega-fund may concentrate on winners already worth $2 billion–$3 billion. Frankel frames the mega-platform model as taking a “call option”; patient seed firms become an “insurance policy” that can validate and help finance an otherwise orphaned company.

5. Founder Collective chooses value and DPI over maximum asset gathering

Frankel admits the temptation to raise more is real: “It’s hard to be contrarian” when capital is abundant. The restraint comes from alignment—the GP has been the largest LP in recent funds—and an explicit preference for being “greedy for returns, not management fees.”
Stebbings presses the opportunity cost: with Coupang, Uber, The Trade Desk, Shield AI, and Suno, why not add a $300 million–$500 million vehicle? Frankel’s practical answer is that the partnership loves early-stage work; the emotional return comes from being first and backing the founder before consensus arrives.
Stebbings’ counterexample is Wix at roughly $2.1 billion of value on $2.1 billion of revenue: markets can stay irrational, and investors cannot force neglected sectors to rerate. His prescription is to “swim in the swim lane that’s swimming in your favor.”
Frankel concedes large funds can work—he cites Thrive and a16z—but says post-2020 DPI remains unresolved, outside rare exposure such as SpaceX and OpenAI. He accepts that his value orientation may look economically irrational: “I was in that company, I was first, I wrote the biggest check” remains the thrill.

6. Low ownership is acceptable; indiscriminate follow-ons are not

Stebbings recounts passing on Deel, ElevenLabs, Granola, Starcloud, and Fractile because only 1%–2% ownership was available, costing hundreds of millions in potential returns. Frankel says he has “never” rejected a deal for that reason: “When you meet the right people and you’re all in, you get what you get.”
With Suno, Founder Collective invested every cent the founders would accept before dilution became unacceptable to them. Frankel wanted more and asked for all available allocation when Matrix later led, but limited ownership never changed his conviction in Mikey Shulman.
Frankel calls pro rata “the original sin” because it is a call option against founders, though he will not accept exclusion when peers receive it. He also questions rounds granting rights only to the lead, preferring equal treatment, while Founder Collective has never led a subsequent financing.
Preemptions can arrive before the seed money is wired, so Frankel relies on post-money thresholds that rise with the market but still define when the opportunity no longer belongs to his strategy. Frameworks permit fast decisions; they also create misses when valuation becomes a lazy shorthand for “no.”

7. A billion-dollar Series A is a momentum trade, not seed investing

Stebbings argues “a billion-dollar valuation is the new Series A”: instead of entering at $50 million and hoping for $1 billion, investors enter at $1 billion and target $20 billion. He points to Mercor at 20, Cognition at 26, and Cursor at 60 as evidence of expanded outcomes and liquidity.
Frankel says that framing describes perhaps the top 200 or 300 companies, not venture broadly. At valuations far ahead of operating reality, investors must know “how and when to get out quickly”; the underwriting becomes momentum and exit timing rather than finding neglected value.
He nevertheless concedes some rigidity was anachronistic. Following Uber, Coupang, Shield AI, and Suno likely would have produced higher absolute gains, but requiring follow-ons across the whole portfolio might not have improved fund multiples: “We’ve captured 80% of the value” does not automatically justify buying the final 20%.
Frankel’s tension is explicit: Stebbings asks how to move the greatest quantum of cash; Founder Collective asks where a smaller fund can multiply ownership rather than accept a prospective 5× or 10×. “Of course, the environment makes you look quite silly in retrospect. The question is how long does this environment go on for?”

8. LP objectives determine whether mega-funds actually succeed

Frankel largely rejects the claim that outcome expansion guarantees mega-platform returns without first asking, “Who are they working for?” Sovereign wealth funds and public investment corporations prioritize IRR, not how many times a manager returns the fund; large vehicles can serve that mandate without matching seed-fund multiples.
Many longstanding LPs now require minimum $50 million checks, making Founder Collective too small. Fund-of-funds investors still need marked-up TVPI to sell their own product and may sell entire billion-dollar funds—or vertical slices—to generate liquidity for their next vehicle.
Fund II’s results were less concentrated than expected across Verkada, Shield AI, Whoop, and PillPack; Fund I retained meaningful exposure beyond The Trade Desk, Uber, and Coupang through Airtable, Simply, and SeatGeek. Frankel’s portfolio construction assumes every investment could become “ginormous,” not merely a planned 10×.

9. Applied AI paid before it had a fashionable label

Frankel retrospectively describes Fund II as applied AI: Shield AI was already called Shield AI in 2016, while Verkada and Whoop placed AI around commoditized hardware such as cameras and drones. None was bought because “physical AI” had become the consensus theme.
His job is to enter a theme five or ten years early, before it attracts momentum capital. The next category is unknowable, but the pattern is consistent: “Those weren’t the expensive ones. They never are.”
In public SaaS, Frankel suspects the selloff may be throwing out the baby with the bathwater. Veeva, which he places near a $30 billion market cap after losing at least half its value, remains difficult to replace when mission-critical biotech work runs through it.
The dividing line is the last 5% of embeddedness: Olo processing vast order flows or Veeva carrying critical research is harder to displace than lightweight software recreated with Claude. The contrarian in Frankel would say to buy a basket of battered top SaaS stocks; Stebbings replies that his momentum bet on Palantir performed better, underscoring cash’s opportunity cost.

10. Great underwriting starts with “I love it because”

Founder Collective uses “I love it because…” at team meetings and requires a belief in a 10× outcome, not merely Jason Lemkin’s suggested 3×. The best answer is founder obsession: every hard question yields a better, candid answer, with no evasion about competition or bad news.
Valuation comes last, after opportunity, market, and founders, with differentiated insight also able to supply the investment edge. Frankel rarely believes the price is perfect and agrees with Stebbings that the best deals leave both sides uncomfortable.
Frankel provocatively says he likes funding “nepo babies,” meaning founders raised inside a vertical, not trust-fund heirs. PillPack’s TJ Parker worked in his father’s pharmacy as a teenager; Suno’s team lived audio at Kensho; each possessed accumulated domain edge that was difficult to manufacture.
Rebar’s Evan worked in his uncle’s HVAC company, searched for an AI product for the quoting workflow, found none, and built it. Frankel highlights 100,000-plus US mechanical engineers, often earning at least $100,000, spending time on blueprint-driven quotes: lived exposure revealed both workflow pain and market size.

11. Secondary liquidity turns DPI into an active choice

Frankel has never seen secondary markets this liquid. Positions in the top 100 private names can be priced reasonably efficiently; a 25% discount to a $10 billion round implies roughly $7.5 billion, while top-50 names may clear at the latest price or a premium.
Momentum sometimes means a round closes in December while the board already discusses March. For a 2024 fund, selling 20% of a top holding to return 25% of the fund can be rational: “You’re still long. You still own 80%.”
Stebbings emphasizes cash velocity: a possible double after five years, an IPO, and an 18-month lockup may be inferior to taking 50% now. Frankel accepts there is no precise science; the ideal outcome is to sell 20%, discover the sale was early, and continue benefiting through the remaining stake.
Frankel says Founder Collective probably sold some Uber too early as it approached a $10 billion valuation, though it remained net long at IPO. Dilution also follows time: Suno’s rapid repricing preserved ownership, while hardware-heavy Whoop required more years and capital.

12. Today’s platforms, labor model, and compute stack will all turn over

OpenAI and Anthropic demonstrate that apparently impregnable platforms can be challenged. Frankel says Google is a net winner, with contextual search and native AI advantages, but calls Microsoft’s AI effort “crappy” and “second rate compared to the top three or four.”
Frankel does not expect mass unemployment; he expects sub-10-person companies, major productivity gains, and a sharper divide between people who can use AI and those who cannot. Stebbings’ concern—“it’s much easier to train than it is retrain”—leaves younger, tool-native workers advantaged over experienced employees with less mental plasticity.
Frankel’s rebuttal is vertical knowledge and human trust. AI may write a low-stakes contract, but in $100 million litigation a client still wants an experienced lawyer across the table; services become a human interface over automated grunt work, while cheaper delivery expands previously unaffordable legal, insurance, and administrative markets.
The next disruption may come from China or photonic computing. Frankel calls OpenAI and Anthropic’s eventual displacement “unequivocal,” sees an excellent chance Chinese open models drive it, and predicts optical, energy-efficient chips could disrupt Nvidia—or become acquisition targets. He argues the US needs more long-term R&D investment.
Despite the boom, Frankel says another dot-com-style crash is certain, with only its timing unknowable. He also changed his mind about consumer AI’s slow impact beyond voice and Suno; at Suno’s $5 billion level, investors are underwriting a Spotify-like consumption product, not merely creation technology.
Suno CTO Martin Camacho said he would adopt a superior external model “without thinking twice”, because users care about experience, not model provenance. Frankel likewise refuses retrospective clairvoyance: anyone claiming to have predicted Uber- or Suno-like speed “is just full of shit.”
His longer horizon remains optimistic: autonomous driving may move “slow, slow, slow” and then arrive overnight, leaving today’s buyers with the last manually driven cars within five to ten years. AI-enabled discovery could also make chemotherapy look “prehistoric,” bringing potentially major advances in health and cancer treatment—“not fast enough,” but profoundly consequential.

From Distillation to RSI: What's Next in the Model Race, with 孟繁青 Google’s AI Brain Drain, SpaceX's Huge Quarter, Airtable’s 90% Collapse, US Data Fuels China AI