Alex Rampell on Venture Scale, Founder Judgment and Durable AI Businesses

Harry Stebbings speaks with Alex Rampell, a general partner at Andreessen Horowitz who leads its applications fund. They discuss the firm's newly announced fundraising, venture economics, founder selection, AI and exits. Rampell invests in several companies he cites, and his predictions and descriptions of their performance are his own.

Rampell argues that venture is increasingly split between large generalists and small specialists. Private companies now remain private longer and can absorb more capital than in the era when many firms went public after only a few rounds, he says. That has expanded the opportunity for growth funds, including his own firm's.

The host asks whether large funds can preserve very high net multiples. Rampell reframes the question for a hypothetical limited partner: a lower multiple on a much larger allocation can return more absolute dollars than a spectacular multiple on a small check. That arithmetic does not establish that any particular large fund will achieve its target or outperform on a risk-adjusted basis.

Rampell describes personal investments in Ribbit's first fund and AngelPad and recalls extremely high multiples on small funds. His figures are recollections, not independently verified or necessarily comparable measures of realized fund-wide performance; he acknowledges that replicating a 120-fold outcome on billions of dollars is unlikely.

A specialist can win by knowing a particular market deeply, while a large generalist can offer broad resources and relationships. A midsize generalist without either advantage may struggle, Rampell says, because investors must find, assess and then persuade exceptional founders to accept their money.

Stebbings challenges the scale argument: a limited partner can also allocate elsewhere, and opportunity cost matters. Rampell agrees the future cannot be known but argues that both specialist expertise and a large firm's resources can help win competitive deals. He invests personally in his firm's funds and therefore has an interest in that view.

Rampell contrasts venture investment with an auction for a public company. Founders may choose investors for reasons other than price, so venture capital involves winning their trust. Some exceptional returns come from investments others rejected, he says, but highly sought-after companies can be competitive from the start.

A small fund can mathematically achieve a larger multiple from the same early investment, Rampell says, but he does not accept that it will necessarily outperform a larger one. He mentions one seed investment marked up approximately 200-fold; a paper mark is not the same as cash returned.

Stebbings points to ElevenLabs as a deal that many investors initially passed on. Rampell says he viewed the founding team as exceptional, and defines agency as taking responsibility and acting rather than waiting to be told what to do. He concedes that traction and price eventually make consensus more complicated.

Rampell likens an early venture investment to an out-of-the-money call option: investors buy ownership in a business whose present economics may not justify the price, hoping future results make the stake valuable. It is an analogy for the risk and potential upside, not a literal description of the security.

His preferred founders can attract three resources: employees, capital and customers. Getting talented colleagues to leave secure jobs, convincing investors to fund the work and persuading the first customers to take a chance on an unproven startup are, he argues, distinct demonstrations of agency.

He also looks for people who study the history of their sector. He recalls meeting Stripe's founders and being impressed by their knowledge of payments, while acknowledging that he personally passed on Stripe's seed round before joining Andreessen Horowitz; the firm later invested.

Other founders, he says, have studied the history of finance, grocery delivery or hospitality before building. His examples express a preference for informed founders; historical knowledge is not a guarantee of success.

Rampell uses The Count of Monte Cristo as a metaphor for a motivation deeper than making an early fortune. He likes founders driven by a desire to prove themselves or overcome a setback, because a very large exit may require them to decline attractive early offers. This is his personal selection framework, not a claim that revenge is necessary for entrepreneurship.

Stebbings asks whether deep industry knowledge can also make investors dismiss new ideas too quickly. Rampell says that when he knows a sector well, he brings a colleague with a beginner's perspective to the pitch and asks what has changed and what would happen if the idea worked.

He calls one approach selling a better product to newly created companies. Established customers may be reluctant to replace a deeply embedded system, but new businesses can choose without migration costs. Rampell cites Stripe and Mercury as examples; he says he invested in Mercury, so he has a financial interest in its success.

The strategy depends on the rate at which new customers form. It may be more plausible in payments or business software than in hospital records, where relatively few new hospitals appear, he says. The contrast is his investing judgment, not a measured comparison of market sizes.

Founders should also consider how long it takes a team to see progress. A startup selling to large enterprises may wait a long time for each contract, which can frustrate employees accustomed to fast releases. Rampell says large enterprise businesses can still be valuable, though he considers them culturally harder to build.

The discussion turns to customers switching easily among AI tools. Rampell distinguishes infrastructure providers from the applications built on them: an application may route among models, while model suppliers seek differentiation. Both layers face competition, and he does not claim to know which will ultimately capture the most value.

His larger concern is that software can be created and distributed much faster than in previous computing eras. He compares the long transitions among early spreadsheet products with his view that challengers can emerge within weeks today. His historical market-share recollections and the speed comparison are illustrative estimates, not verified measurements.

A system of record may still be difficult to replace when it holds important customer data and workflows. Rampell's formulation is that startups must gain distribution before incumbents acquire the innovation. He likes starting with an overlooked problem and building deeper customer relationships before the field becomes crowded.

Stebbings asks whether faster competition and longer private-company lifetimes create a liquidity problem. Rampell estimates that only about 5% of today's unicorns may eventually list publicly. He presents this as a bet, not a count or an established forecast.

Rampell dislikes large founder secondary sales when they leave founders financially secure while employees and investors remain illiquid. He allows that some liquidity may make sense when the opportunity is shared broadly and a founder remains committed. Stebbings adds that excessive funding can also encourage a company to pursue too many projects at once.

Rampell argues that more money or more employees do not automatically produce better outcomes. Small teams with strong people may solve problems more creatively, in his view. His government examples are analogies and political judgments, not evidence about the actual performance of a particular agency.

Stebbings asks whether Series A is the worst investment stage because prices may rise before the business makes much progress. Rampell says round labels vary: one Series A may fund compute for an unproven team, while another may already have substantial recurring revenue. He agrees that paying much more for little additional evidence can be dangerous.

Rampell says he led a Series B for Rillet roughly 60 days after its Series A and sits on its board. He would have preferred to invest earlier, but argues that missing a company he believes is a winner can also be costly. His assessment is a view from an investor in that company.

The issue is whether a founder can handle capital without losing discipline. Rampell recalls severe cuts at his former company TrialPay before it recovered and was sold to Visa. He says choosing between difficult options is often better than using a large cash balance to postpone every decision.

He also acknowledges exceptions: a founder who sells shares can remain committed through difficult times. The question is how financing changes incentives and decision-making, not whether all secondary transactions or all large rounds have the same effect.

Stebbings describes losing an AskLeO deal to Rampell's colleague and wonders whether he should have offered more. Rampell says both excessively high and excessively low prices carry risks. He recounts warning founders that a high prior valuation can make the next financing or a sale more difficult if performance does not catch up.

The warning often fails, Rampell thinks, because founders must be unusually optimistic to start companies. He urges them to consider the effects on culture and future choices: abundant cash can encourage waste, while a very high last-round price can narrow the range of acceptable acquisition offers.

Asked how the firm reflects after missing an early round, Rampell says price and desired ownership are often the sticking points. A large fund needs enough ownership to affect returns and must account for future dilution. Founders, conversely, want to limit the equity they give up.

Winning every deal might mean a firm is an exceptional partner, but Rampell says it might also mean it has not tested how much ownership it could obtain. Stebbings pushes back that a large firm able to follow on may benefit from accepting smaller initial stakes in more winners.

Rampell says investors must admit mistakes and update their views. If a company has become a clear leader, he would rather enter a later round at a smaller stake than refuse because the firm missed the earlier investment. That approach still involves the risk that a perceived winner does not sustain its lead.

The ownership debate has no simple cutoff. Stebbings illustrates the math with an initial 10% stake diluted to 5%, while Rampell cautions that estimates of future market size can be badly wrong. Their examples are hypothetical, not a valuation or return forecast for a specific company.

Rampell summarizes two kinds of investment he seeks: meaningful ownership in something that could work but remains unproven, or even a smaller stake in something demonstrably working. He stresses that the bar for the latter must be high; a slight lead at $1 million of recurring revenue is not proof of an enduring winner.

For his fund, Rampell describes three themes. The first is selling systems of record or vertical operating software to new businesses: growth may be slower, but embedded workflows could make customers more durable. He cites Rillet, a company in which he is an investor.

The second theme is software that performs work previously done by people. He cites Eve, a company serving plaintiff attorneys, as a possible way to make smaller cases economical. His example compares hypothetical labor and software costs; it does not establish that the tool can independently perform all legal work or guarantee case outcomes.

Fast growth alone is not enough, he says. A product combining existing models for outbound calls could attract many competitors, so an AI application needs a plan to become more embedded in the customer's workflow. He describes that discussion with Salient, another company in which he says he invests.

The third theme is access to valuable data that general-purpose models may lack. Rampell discusses legal data provider vLex and medical information company OpenEvidence as examples. His claims about their data access and growth are investment arguments, not independent verification or medical advice.

Stebbings describes a portfolio company that grew rapidly but found fundraising difficult compared with a much faster-growing AI company. Rampell says investors should weigh growth and retention together. He would prefer a slower-growing, durable system of record to a fast-growing product that customers can easily replace, though actual durability remains to be proven.

On selling a company, Rampell says the process differs from raising a round. He advises founders to build genuine relationships over time with people at several potential acquirers, particularly operational leaders who might benefit from the product, rather than approaching corporate development only when a sale is urgent.

He suggests treating that work as a small ongoing background process. Timing matters: an attractive partnership may later lead to an acquisition, but a company that waits until performance deteriorates may find buyers less interested. He describes how TrialPay pursued real partnerships with potential acquirers rather than relationships solely designed to engineer a sale.

Rampell gives similar advice for fundraising: the CEO should cultivate investor relationships before the company urgently needs money. He recalls spending substantial time with Greylock and Reid Hoffman before TrialPay's Series D, so they had a history on which to judge him and the business.

Asked whether AI-driven labor displacement will materially appear in the near term, he says he is unsure about the timing but expects uneven effects. He distinguishes established systems of record, seat-based software exposed to automation, and products in between. His examples of potential revenue and employment changes are scenarios, not measured outcomes.

Stebbings presses him on customer-support and legal work. Rampell agrees some routine roles could fall sharply, while suggesting other workers may be redeployed toward high-value customer relationships or new work made economical by automation. Neither outcome is guaranteed, and the discussion does not quantify the net employment effect.

In the quick-fire round, Rampell says he has become more skeptical about buying conventional companies and adding AI, citing a possible mismatch between founders and markets. He also mentions credit as a potential product for portfolio companies, while recognizing that a lender's remedies can conflict with a venture investor's relationship with founders.

His favorite investing advice returns to people who can attract labor, capital and customers and know their field. He also acknowledges that, eventually, financial reality matters: a founder's qualities cannot by themselves support an unsuccessful public offering.

His biggest miss was an early Plaid round. He recalls arguing over a relatively small difference in valuation, then investing in a later round at a far higher price. He presents the experience as a lesson in admitting an error and revisiting an assumption; he and his firm have an investment interest in Plaid.

Looking five years ahead, Rampell expects software and AI to expand into more markets and remain a source of value for venture investors. His optimism about new applications and robotics is a forecast from an investor with exposure to the sector, not a guarantee of returns.