David Frankel on Seed Investing, AI Winners and the Value of Patience

David Frankel, co-founder of Founder Collective, joins Harry Stebbings to discuss seed investing through the AI boom. Frankel sees an extraordinary technology wave, but expects many businesses to fail and warns that venture funds of different sizes must pursue different kinds of returns.

The two recall meeting 11 years earlier. Frankel says what stood out about the then-19-year-old Stebbings was focus and energy: identifying that intensity is one of the pleasures of backing founders, although it is not the whole job.

Stebbings argues that $50 million to $100 million seed funds are caught in the middle: too large to collaborate with tiny checks, but too small to lead the largest seed rounds. Frankel answers that larger funds face a narrowing pyramid of possible outcomes; smaller funds can make their economics work with smaller exits.

Frankel cites his firm's recent analysis of historical company outcomes, including a roughly $2.6 billion median among its top 500 cohort. His point is that owning a meaningful stake in an outcome of that size can return a small fund, even if the company never becomes a trillion-dollar giant.

Seed remains intensely competitive, Frankel says, but patience can reveal an exceptional founder or team. Founder Collective can invest $500,000, $3 million or another amount rather than underwriting an entire large round, and he believes strong entrepreneurs can still be found away from the most fashionable deals.

In some recent large rounds, Frankel says, Founder Collective has invested $500,000 or $1 million alongside a larger fund. He describes the firm as a possible 'insurance policy': if growth falls short of the big investor's expectations, a patient seed investor may still advocate for the company. He cautions that he may be extrapolating from a limited set of deals.

The firm still sees $3 million and $4 million rounds, he says, while valuations vary widely. For the hottest AI companies raising enormous sums, he believes the evidence on capital efficiency and eventual returns is not yet settled.

Stebbings asks whether a company outside the hottest AI category can obtain its next round. Frankel replies that existing businesses can use contemporary AI tools too: a domain expert may combine years of industry knowledge with a strong CTO and AI-enabled development to solve a specific problem.

On the claim that price matters less if an investor gets into the winner, Frankel says entry price changes the scale of success required. He dislikes uncapped seed notes economically, although he has accepted one for founders and a relationship he values, and acknowledges that saying no can mean missing exceptional companies.

Stebbings worries that starting a company has become a normalized career choice. Frankel distinguishes a founder title from entrepreneurship: the latter requires resilience, the ability to energize others and the capacity to learn through successive stages of growth.

In Frankel's experience, a CTO's role may change as a company grows, while a CEO must develop management and recruiting skills. He recalls Suno's Mikey Shulman saying that recruiting occupied 30–40% of his time, and a past lunch where Jeff Bezos described spending half his time filling positions.

Asked whether he would back a great CEO with a weaker co-founder, Frankel says he rarely does, even though he regrets some passes. He looks for complementary CEO and CTO skills and, especially, trust and productive chemistry between them; founders need not be alike.

Stebbings says he has increasingly favored deeply technical founders. Frankel says focus, energy and intelligence still matter most to him, while experience can supply valuable market knowledge. He stresses that the intensity he associates with younger founders can continue for decades.

Stebbings describes passing on a company projected to grow annual recurring revenue from $1.5 million to $5 million and then $15 million, because faster opportunities compete for capital. Frankel argues that venture funds often take much longer than their nominal ten-year lives and that durable businesses can take time.

As an example, Frankel says he invested in SeatGeek in 2010 and that Founder Collective still owned its shares at the time of the interview. Besides headline revenue, he looks for signs such as customer retention, increased account spending and growing usage.

Companies overlooked for missing an aggressive growth benchmark can create opportunities, Frankel suggests, including in seed extensions. He notes that investing in these later rescue rounds is not generally his firm's strategy, though he can see a potential capital-markets opportunity.

Frankel calls AI the defining wave of his career, while expecting substantial 'roadkill' alongside enduring companies. His prediction that most companies will not become lasting giants is a judgment about venture risk, not a measured forecast for a defined cohort.

The economics of a small seed fund let it succeed without owning every giant, Frankel argues. He also finds seed crowded and increasingly commoditized: reputation and a portfolio of founders willing to recommend an investor can help earn access.

Stebbings asks whether big platforms will win on price. Frankel offers a counter-case: a junior investor at a large firm may move on, leaving a slower-growing startup without a senior champion for follow-on funding. He explicitly says this is a contrarian scenario and does not always happen.

Would Founder Collective raise substantially more money? Frankel says the partners discuss it. He says the general partners are the firm's largest limited partner in recent funds and prioritize distributions and alignment; a larger vehicle could increase absolute profits but would change the work and return objectives.

Frankel enjoys early-stage investing and looking for value where others are not funding, including consumer businesses with unusually capable customer acquisition. Stebbings counters that an investor can be right about underlying value yet wait a long time for the market to recognize it.

Frankel acknowledges that some large funds have produced excellent returns and that his preference for being a founder's earliest backer may not maximize economic opportunity. He says Founder Collective is now writing some $3 million to $4 million checks, deploying its fund faster to secure ownership in exceptional companies.

Stebbings says ownership thresholds caused him to pass on companies he now wishes he had backed. Frankel prefers more ownership, all else equal, but says he has not rejected an otherwise compelling opportunity solely because the stake was small.

Frankel calls pro rata rights a potential burden for entrepreneurs because they give existing investors an option on future rounds. He favors equal treatment of investors where rights are granted and says his firm has had to earn the chance to invest more rather than lead later rounds.

When fundraising moves so quickly that another term sheet can arrive before the first investment is wired, Frankel says a valuation framework helps the firm act decisively. His partner Eric Paley helped establish discipline around what post-money valuation still fits their strategy.

Looking back at winners such as Uber, Coupang and Shield AI, Frankel thinks systematically following later rounds could have increased absolute returns. He is less certain it would have improved the fund multiple, especially if the policy had required following on across the portfolio.

Stebbings proposes that a $1 billion valuation has become the new Series A for the fastest companies. Frankel calls that a momentum strategy focused on a relatively small set of assets; his own preference is to find value early, while admitting the firm has also invested in momentum.

Frankel says the firm has often decided its initial ownership position was sufficient and moved on to the next seed deal rather than pursue a later round. In retrospect, the continuing rise in some valuations makes that choice look questionable; how long the environment lasts is unknown.

Stebbings argues that larger platforms can generate strong venture returns when outcomes expand dramatically. Frankel largely disagrees that the same fund-multiple logic applies across sizes: in his account, different limited partners may emphasize internal rate of return or the amount of capital put to work rather than the same multiple.

He distinguishes paper value, measured by TVPI, from cash distributed to limited partners, or DPI. Frankel says certain large funds have strong paper marks, while realized distributions for more recent vintages remain less clear; this is his assessment rather than a full performance audit.

Looking back at an earlier fund, Frankel points to Shield AI, Verkada and Whoop as examples of intelligence applied to hardware platforms. He thinks future investors will need to identify emerging themes before they become fashionable, although he does not claim to know what the next theme will be.

Frankel says his funds' returns have been less concentrated than one might expect, naming several companies from each of the first two funds. These are his descriptions of the portfolio, not disclosed fund-level performance figures.

Stebbings worries that AI can weaken software leaders before investors get a liquidity event. Frankel agrees AI is powerful but questions whether all software companies are equally vulnerable: mission-critical systems embedded in complex operations may be harder to replace than less deeply integrated tools.

Frankel says he hopes each new investment could become enormous, not merely a modest multiple. The team starts discussions with the prompt 'I love it because': the answer might be a founder's candor and obsession, a distinctive insight into a market, or a defensible edge.

He says the firm considers founders and opportunity before valuation, even though pricing is often uncomfortable. A founder who admits how hard the business is and engages directly with difficult questions can be more compelling than someone who claims to have no competitors.

Frankel uses 'nepo babies' to describe founders who grew up around an industry and gained unusually deep knowledge of it, not simply wealthy heirs. He cites PillPack's T.J. Parker working in his father's pharmacy and a founder who saw an opportunity in HVAC estimating through a family business.

On exits, Frankel says secondary markets for a limited group of highly sought-after private companies have become unusually liquid, sometimes at discounts and sometimes at premiums. He emphasizes that only a handful of his portfolio companies are in that rare category at any given time.

For an early fund, selling part of a winning position can return meaningful cash to limited partners while preserving substantial upside, Frankel says. Stebbings values that certainty and the speed of cash returned, even if holding might ultimately produce a higher price.

Frankel says exit timing is not an exact science: in hindsight, Founder Collective probably sold some Uber shares too early, although it remained a net holder at the IPO. Selling a portion and later seeing the price rise can still leave a fund with a successful outcome.

The conversation turns to dilution. Frankel contrasts the different timelines and fundraising histories of Whoop and Oura, both investments he says the firm is proud of. Faster valuation increases can reduce dilution in some cases, but those conditions are not universal.

Stebbings asks whether limited partners have returned to valuing paper marks and prestigious names. Frankel says some long-standing investors now have minimum check sizes too large for his fund, and that a fund of funds' own incentives can make TVPI important.

Frankel has also seen investors sell entire funds or slices of funds on the secondary market to provide liquidity. He is unsure whether venture's increasingly sophisticated financial machinery is always good for founders: it can create options for winners while making other situations more complicated.

Despite discussing financial engineering, Frankel says he spends most of his time looking for founders with determination and a clear vision. He describes himself as more interested in entrepreneurs and their work than in managing a complex financial portfolio.

Stebbings asks whether founders are less focused as angel portfolios, side funds and multiple projects proliferate. Frankel has seen some people leave or delegate too early, and says investors can be dazzled by a famous repeat founder, but he cautions that most founders he encounters still want to make their company work.

In his experience, second-time founders whose first company did reasonably well without producing an enormous outcome have sometimes performed better for the firm. He thinks they combine lessons and a network with hunger, while acknowledging that this is an observation from his own portfolio.

The two discuss whether success has made them less patient. Frankel says rising expectations can make satisfaction harder, and that he tries to focus on kindness and how he leaves people feeling rather than winning every small negotiation.

On AI platforms, Frankel believes OpenAI and Anthropic have changed the competitive landscape and may themselves eventually be disrupted. He considers Google well positioned and is critical of Microsoft's AI execution; these are his competitive opinions and predictions, not established outcomes.

Frankel expects wealth created by leading AI companies to flow into venture, angel investing and other assets. He also predicts a future boom-and-bust cycle comparable to the dot-com crash, while saying nobody knows when it might arrive.

More wealthy founders and employees may fund their own networks, he says, giving entrepreneurs alternatives to traditional venture capital. That makes it more important for seed investors to explain the distinctive value they bring.

Asked about jobs, Frankel expects productivity gains from AI and has seen very small teams accomplish a great deal. He does not expect mass unemployment, but anticipates unequal access to skills and tools and believes retraining will matter. Stebbings is more concerned about the difficulty of retraining experienced workers.

Frankel suggests that experienced workers retain domain expertise and relationships that can be valuable in sales or high-stakes professional services. His view is that AI may automate substantial background work while clients continue to want human judgment and contact in some consequential situations.

Stebbings relays a founder's speculative idea that companies might one day spend enormous sums on a single model answer to a high-value question. Frankel sees possible applications in government and defense; neither presents this as an established market price.

Discussing US policy, Frankel questions how much credit any president deserves for an economic environment built over many years. He thinks an environment supportive of AI has helped US technology, while acknowledging that insufficient safeguards could be problematic.

Frankel views the US and China as major AI powers and, as an early investor in Shield AI, believes AI has important defense applications. Stebbings expresses concern about Chinese open models; Frankel thinks future competition could come from China, but their security and competitive assessments are opinions, not verified predictions.

Frankel speculates that photonic computing could eventually challenge existing chip architectures or become an acquisition target for Nvidia. He hopes optical chips could improve energy efficiency in data centers, but offers this as a future possibility rather than a demonstrated solution to data-center energy demand.

Stebbings notes that advanced chips and energy companies can require substantial capital. Frankel worries that the US is not spending enough on energy research and broader research and development; he criticizes cuts to research funding and points to the universities around Cambridge, Massachusetts.

The speakers contrast regulatory environments in the US and China. After broad claims about China's approach, Frankel qualifies that it is not an 'anything goes' environment; he reports what biotech investors have told him about research and licensing activity rather than supplying a regulatory analysis.

What would make him raise a larger fund? Frankel says he once saw a gap between the perceived risk of seed investing and the value of exceptional early founders, but believes that gap has narrowed. He might consider more capital if later rounds contained a comparable neglected value opportunity.

A promising company whose revenue is growing steadily but not explosively could become such an opportunity, he says. At the time of the interview he does not see a broad, obvious mispricing across later stages, because capital is still finding many deals.

Stebbings questions the cash efficiency of an approximately 17-year journey to an Olo outcome of about $2 billion. Frankel accepts that the internal rate of return may not compare favorably with faster outcomes, while describing his years alongside founder Noah Glass and the eventual sale as personally rewarding.

In the quick-fire round, Frankel says consumer AI has had less visible impact on everyday life than he expected so far, apart from model interfaces and improvements in voice. He cites Suno as an example and expects more consumer applications to develop.

Frankel says investors buying Suno at higher valuations may be betting that it can evolve from a creation tool into a consumer destination competing for listening time. He recounts CTO Martin Camacho saying he would use an external model if it delivered a better product, putting the user experience first.

Neither investor claims to have predicted the speed of the latest growth. Frankel recalls a partner saying that Uber and surrounding investments were evaluated in much the same way at the time; he is skeptical of investors who claim they knew exactly how a breakout would unfold.

Some deals provoked debate because of geography or industry: Frankel cites Coupang in South Korea and Shield AI's defense work. He says the team's 'I love it because' approach sometimes leads it into unfamiliar or controversial areas.

Frankel says prediction markets resemble gambling to him and compares paper venture marks to predictions rather than realized cash. The comparison is his analogy, not a legal characterization of any particular market.

Frameworks have saved the firm from some mistakes, Frankel says, but also made it too easy to reject opportunities on valuation. He names Klaviyo as a deal he regrets passing on and says he has become somewhat more willing to make exceptions for extraordinary founders.

For relationships, Frankel recommends kindness and presence. He keeps phones away from the dinner table and bedroom, while admitting that distraction can come from within and that being attentive remains a lifelong practice for him.

Both men close with hopes for future technology. Stebbings hopes for advances against chronic illness; Frankel anticipates progress in autonomous driving and health, while acknowledging that technologies can take much longer to mature than early promises suggest. These are aspirations and forecasts, not clinical claims.

Frankel says the possibility of discovery is what excites him about the coming decade. Stebbings thanks him for the conversation and their long friendship, and Frankel returns the compliment.