Benchmark, a16z and the Math of Venture Capital in the AI Era

Harry Stebbings talks with Scale Venture Partners’ Rory O’Driscoll and SaaS investor Jason Lemkin about the economics of large and focused venture firms, AI and employment, and the consequences of rapidly changing software markets. Their numerical examples and forecasts reflect the discussion in May 2025.

The panel opens with a reported potential OpenAI acquisition of Windsurf for about $3 billion. The speakers discuss it as news circulating at the time, not an established completed transaction, and consider what a large model provider’s distribution could mean for independent coding companies.

O’Driscoll says declining an acquisition offer involves considering what happens if the buyer then purchases a competitor. He explicitly treats an alleged offer to Cursor as a rumor, and does not claim to know what its investors or founders decided.

Lemkin observes that investors in the same company may react differently to an offer depending on when they invested and what they have already sold. O’Driscoll advises founders to understand each stakeholder’s position, because incentives around an exit do not always align.

The guests compare investors’ reactions to exits and recall deals such as Instagram and Loom. A quick return of capital, a multiple and the reputational benefit of a visible transaction are distinct outcomes, and neither a mark nor an offered price necessarily equals cash distributed to every investor.

Lemkin says early investors may become risk-averse after a holding has appreciated substantially. O’Driscoll argues that letting a strong winner continue to grow can matter enormously: a further doubling of a 10x position would turn it into 20x, though that additional gain is not guaranteed.

Stebbings advances the hypothesis that large multistage firms may win the next decade because they can finance companies from seed through very large private rounds. He expects more exceptionally valuable private companies, but acknowledges that this is a view about the future.

O’Driscoll separates winning assets under management and deal access from earning sufficient returns for limited partners. Large firms may see more winners because they deploy more capital; whether all their funds can meet their return targets is a different question.

For a midsized firm such as Scale, he expects competition with large pools of capital to make investing difficult for several years. Even if those pools ultimately fail to deliver the hoped-for returns, the near-term ability to pay high prices can change the market.

Lemkin describes sometimes declining to compete for seed rounds priced beyond what his fund can support. The panel notes that such firms may use early checks to build relationships and earn opportunities to invest much larger amounts later.

O’Driscoll calls venture a bundled product: a firm able to invest across stages can treat an early round as part of a broader relationship. He compares an early check to a loss leader as an analytical analogy, not a claim that every multistage investor intentionally loses money at seed.

Stebbings asks whether investors now need to participate at pre-seed to see the best Series A and B rounds. O’Driscoll says Scale is considering the question, but that a firm should not promise a seed strategy to founders unless it is prepared to follow through.

O’Driscoll argues that founders primarily seek funding and support, while the quality of a fund’s return is ultimately a matter for its limited partners. Stebbings counters that many founders value the attention of an investor who makes fewer commitments.

Discussing an outside analysis of Series A investments, O’Driscoll cites roughly 63 Benchmark investments with a 10% hit rate and 454 Andreessen Horowitz investments with a 2% hit rate, using $5 billion company outcomes as the threshold. He attributes the figures to the analysis, not to a complete audit of either firm.

On those figures, he says Benchmark had six such outcomes and a16z ten: a higher hit rate for the focused firm, but more successes in absolute numbers for the larger-volume strategy. Hit rates and counts alone do not establish fund-level returns or prove that focus caused the difference.

The panel cautions that the deals come from an earlier market. O’Driscoll adds that a fund’s share of all major outcomes is another measure, distinct from the percentage of its own deals that become hits; a strategy requiring an unprecedented share of winners may be difficult to sustain.

Stebbings thinks a larger number of extremely valuable companies could make multistage funds work. O’Driscoll responds that the model may depend on putting substantial capital into companies that continue compounding while private, rather than simply winning more Series A deals. Both are forecasts.

Lemkin predicts substantial job displacement as AI replaces knowledge work and asks how much of the resulting value software companies might capture. His estimate of rapid displacement is a prediction drawn partly from his own operations, not a measured economy-wide outcome.

O’Driscoll agrees AI could create valuable companies and improve efficiency, but disputes a sudden wave of mass unemployment over the next one or two years. He expects adoption and aggregate economic effects to take longer.

Lemkin proposes a $100,000 bet on employment outcomes and says his team had eliminated five roles in the preceding 90 days because of AI. His broader claim that people do not want to work is an opinion based on personal anecdotes, not evidence about job seekers generally.

The disagreement is not whether individual tasks can be automated. It is how quickly organizations adopt the tools, whether new demand offsets some reductions, and how much change reaches sectors outside technology.

Stebbings argues AI can spread faster than earlier technologies because many users already have the necessary interfaces. O’Driscoll points to organizational adoption and historical productivity growth, while allowing that a larger technology sector could accelerate early uptake.

O’Driscoll suggests productivity could rise sharply in some industries while other parts of the economy change slowly. Lemkin worries that people displaced from support and other roles may not find equivalent jobs; neither can establish the eventual labor-market outcome in the conversation.

The panel discusses college, graduate employment and demand for practical skills. Their comments are judgments about a changing labor market and the value of particular educational choices, rather than advice applicable to every student.

Stebbings raises a contemporary political threat to Harvard’s tax-exempt status and asks what pressure on university endowments might mean for venture capital. O’Driscoll treats the outcome as uncertain and suggests endowments could nevertheless become more cautious about commitments.

He says smaller emerging funds often rely more on university endowments than very large funds do, so funding pressure could reinforce existing advantages of scale. He also argues that university research and graduates are important to the technology ecosystem.

The guests discuss the possibility that AI competition can hurt a software company without destroying it. A business may remain viable but suffer slower growth, price pressure or customer losses that change its prospects for a venture-scale exit.

They discuss reports of Decagon’s valuation relative to annual recurring revenue and debate whether rapid growth justifies a high multiple. The multiple is a reported financing comparison and the panel’s valuation debate, not evidence of realized investor returns.

O’Driscoll emphasizes that venture investors buy the possibility of large future upside, while acknowledging that paying a high price makes the outcome more sensitive to assumptions. The guests disagree about how durable a fast-growing AI application’s advantage will be.

Lemkin worries that new competitors can force discounts, increase churn and slow sales even when an incumbent remains embedded in customer workflows. O’Driscoll sees integration and enterprise relationships as potential sources of stickiness but agrees the market is unusually unsettled.

The panel distinguishes horizontal developer products, where switching may be relatively easy, from specialized software embedded in complex professional workflows. Stebbings cites Solve, an AI company for patent lawyers whose Series A his firm led, giving him an investment interest in the example.

Lemkin argues that even a business not killed by AI competition can be knocked off the growth trajectory it needs for an IPO. O’Driscoll predicts that leading enterprise products may eventually settle into a more stable market, but explicitly acknowledges that he could be wrong.

The guests discuss model capabilities as a possible source of continued disruption. If a model itself absorbs more application functions, a startup’s product can face new competition; the extent and timing of that shift remain uncertain.

Stebbings asks about public restaurant-software company Olo and the possibility of a sale. Lemkin says he has not developed a detailed view of Olo, but argues that a company facing competitive pressure should seriously consider a credible acquisition offer.

They also discuss reported DoorDash acquisitions and the premium offered for Deliveroo compared with its prior public valuation. Their interpretation of the strategic value and the UK market’s pricing is opinion based on the information available at the time.

Finally, the panel returns to Josh Kopelman’s venture arithmetic: each firm should ask how much of the available market it must capture for its fund model to work. O’Driscoll says historical deal shares can be a sobering check on ambitious assumptions.

The speakers distinguish a return multiple from its duration: doubling money over a shorter period can produce an annualized return comparable to a larger multiple over a longer period. They also emphasize the need to see, win and hold enough relevant opportunities.

Stebbings argues that media can keep an investor visible without requiring a firm to write checks merely to demonstrate activity, a position connected to his own media business. O’Driscoll agrees firms need a way to remain relevant and says the large-capital market is the one investors must navigate now.