Sam Lessin, Jason Lemkin and Rory O’Driscoll Debate DPI and the Future of Venture

Harry Stebbings hosts Sam Lessin, Jason Lemkin and Rory O’Driscoll for a discussion of venture-fund performance, AI infrastructure spending and the market for startups.

Stebbings opens with a debate over whether distributions to paid-in capital, or DPI, are the only meaningful measure of fund performance. DPI measures capital distributed relative to paid-in capital; it differs from a paper valuation of remaining investments.

Lessin argues that the job of one kind of venture investor is to return money, making realized distributions central. He contrasts this with building an asset-gathering business that earns fees on larger funds, while disclosing his perspective as a seed investor.

O’Driscoll agrees that actual distributions matter most for an old fund but rejects the idea that total value to paid-in capital, or TVPI, has no use. TVPI includes both distributions and estimated residual value; early in a fund's life, the latter is uncertain.

Lessin says private marks can be driven by financing prices and incentives rather than realizable cash. He prefers examining whether a young portfolio contains credible potential fund-returning companies.

O’Driscoll argues that higher interim TVPI could contain some statistical signal even if it is imperfect. Lessin says valuation practices vary and proposes that the question should be tested; neither presents a completed study.

The guests agree that paper marks need interpretation and that eventual money returned matters. They differ on how much an interim valuation can tell a limited partner before exits occur.

Stebbings asks whether medium-sized venture funds are being squeezed. Lessin says his smaller early-stage fund and large asset managers play different games and questions whether a billion-dollar fund can consistently earn high multiples.

O’Driscoll, whose firm manages a fund in the disputed range, disagrees that size alone determines viability. He says fund size, check size, stage and portfolio construction must fit together.

Lemkin asks whether founders seeking larger Series A rounds will have fewer choices if middle-sized funds disappear. The guests debate the influence of very large funds rather than asserting that every mid-sized firm will fail.

Lessin argues that some capital is misallocated and that niche opportunities can still exist. O’Driscoll says a large fund writing many small checks must also consider how many companies its partners can support.

The panel considers why some companies struggle to raise while large funds have abundant capital. Lessin's answer is that only a few companies will matter enormously; O’Driscoll later challenges the implication that other investments cannot make money.

Lessin values the option to sell early holdings to other private investors when the investment thesis changes. He says a fund needs to preserve exposure to its most important potential winners while evaluating the price and reason for a sale.

O’Driscoll argues that the largest winner can dominate an era without making all other successful investments worthless. He cites a recent IPO as an example of a meaningful return below the very largest outcomes.

The investors discuss Chime's prior private valuation and possible early shareholder sales. Lessin explicitly says he is inferring from public filing information; the panel does not verify individual investors' transactions or returns.

O’Driscoll says a Series B investor can make a good return even if a later investor bought at a much higher valuation. The figures discussed are estimates, not confirmed distributions from Chime.

The group returns to DPI and the time value of money. Receiving a smaller amount earlier can be valuable, while selling also gives up potential future upside.

Lessin argues that pricing private companies is difficult and distinguishes seed investing from large late-stage bets. O’Driscoll says early growth investors have some operating data, but late-stage prices also demand discipline.

O’Driscoll says a small number of later-stage investments can still produce exceptional multiples, mentioning Anthropic and OpenAI as examples in his opinion. He warns against treating those exceptional cases as the average deal.

Stebbings proposes that Revolut could become much more valuable; this is his forecast, not a current valuation. Lessin questions whether venture managers should hold public shares to express a public-market view or distribute them to investors.

Lemkin discusses market share and customer-acquisition costs as possible constraints on fintech growth. The percentages are his estimates, and the panel notes that word of mouth can change acquisition economics.

Lessin questions whether a company can be a good business without being historically important. O’Driscoll and Lemkin push back: investors and customers can benefit from companies that do not become the largest of their generation.

The panel debates what it means for a company to matter, moving between value to customers, returns to investors and lasting historical significance. These are different judgments, not a single measurable standard.

Stebbings asks how founders should react to investors' definitions of relevance. The guests acknowledge that not every good business needs the same scale or venture-fund economics.

They discuss the trade-off between pursuing a very large outcome and building a durable company. Their examples reflect their own investment perspectives and incentives.

The conversation turns to Mary Meeker's AI report and the speed of adoption. The investors compare AI products with earlier consumer platforms, while cautioning that rapid usage does not automatically establish durable revenue.

O’Driscoll raises the scale of AI capital expenditure and asks when infrastructure providers will earn a return. He observes that hyperscalers have strong cash flows but are committing significant capital; his figures and payback scenarios are estimates.

The panel distinguishes infrastructure revenue from the downstream application revenue needed to support sustained spending. Future demand and the economics of deployment remain uncertain.

The guests discuss whether lower unit costs will expand use enough to absorb the new capacity. They treat that as a hypothesis about supply, demand and product development rather than a proven outcome.

O’Driscoll considers how public investors might shift from rewarding investment to demanding returns. Lemkin argues the market can change its view quickly; both describe possible market behavior, not a timetable.

The group discusses competition among model providers and the implications of additional models from China. More competitors could affect prices, but the extent and timing of commoditization are uncertain.

Lemkin cites a reported steep decline in token prices and warns against assuming current inference costs will remain fixed. The percentage is a reported comparison, and total customer costs also depend on usage and model choice.

The guests argue that companies should experiment with AI instead of waiting indefinitely. Their suggested pace is an operating view; lower model prices do not ensure every product will work or become profitable.

Stebbings relays a Windsurf founder's argument that startups face existential pressure to ship. Lemkin says many business-to-business companies should feel greater urgency to adapt to AI.

The investors debate how younger users may interact with software through voice and agents. They predict interface changes but do not establish that traditional interfaces have already disappeared.

O’Driscoll imagines an agent-facing system of work sitting above existing systems of record. If users interact mostly with agents, he argues, the layer controlling the workflow could capture more value over time.

Lemkin discusses Model Context Protocol integrations as a possible way for an assistant to act across applications. He presents the effect on established SaaS interfaces as a threat and a prediction, while acknowledging current integration friction.

Lemkin discloses an investment in Mangomint and describes the founder's concern that an agent could obscure the underlying booking application. This is a portfolio company example and a hypothetical competitive risk.

O’Driscoll says incumbents may continue to grow on existing customers even as new products win incremental budgets. Venture investors seeking the next source of growth may therefore focus on different revenue from an incumbent's existing base.

Stebbings asks about a set of contemplated IPOs and acquisitions. The guests discuss Circle's reserve-based business model and possible economics using figures and assumptions from the period, not current financial results.

They distinguish gross income on reserves from amounts shared with distribution partners and operating costs. O’Driscoll's valuation discussion is his estimate, not a recommendation or a guaranteed market price.

The discussion covers how public investors assess growth and margins. The guests caution that a company's listing price, future trading and eventual returns are separate matters.

The panel turns to reported high valuations for Y Combinator startups, especially AI companies. Lemkin argues that higher entry prices can reduce a seed investor's ownership for the same amount invested.

O’Driscoll says YC provides a valuable service to founders and that investors must either pick carefully at the resulting prices or find other opportunities. He does not claim every company in a batch will become a winner.

Lemkin says a small ownership stake may make sense for an angel or very small fund but can be difficult for a larger fund seeking a fund-returning position. Outcomes depend on dilution, entry price and eventual exit value.

O’Driscoll says it may take years to learn whether a cohort's high prices were justified. A single successful company does not establish the average return for all investments at those prices.

The guests discuss founders receiving more cash for a given dilution and the possibility of later financing at a lower price. Their assessment of SAFE terms is a general discussion; actual rights depend on the documents.

The group says an active IPO market and private-equity fundraising can create liquidity, but price still matters. The pace of transactions and acceptable valuations can change.

In a prediction-market round, the guests speculate on a potential OpenAI–Jony Ive device, including whether it would have a screen and how well it might sell. No product specifications or sales outcome are established.

They debate whether Meta might release a closed AI model, offering different predictions about competitive and security incentives. These are views at the time, not confirmed product plans.

The final wager concerns whether Elon Musk will leave the Tesla CEO role before 2027. The guests disagree about the odds and explicitly discuss uncertain circumstances rather than any known decision.

They consider the challenge of managing several companies and discuss possible leadership choices as speculation. Stebbings closes by thanking the guests.