20VC Roundtable on AI Competition, Enterprise Spending and Venture Capital

Harry Stebbings opens a discussion with investors Jason Lemkin and Rory O’Driscoll about Apple's lawsuit involving OpenAI and former Apple employees. The panel distinguishes the reported allegations against individual employees from allegations about encouragement by a senior executive; it does not establish a court finding of liability.

O’Driscoll says he read part of Apple's complaint and considers the alleged conduct serious. His assessment of the evidence and predictions about individual consequences are opinions about a pending dispute, not a judicial determination or a complete account of the parties' positions.

He says discovery could matter to allegations about what senior personnel knew or encouraged. His further claims about Apple's motives and the impact on OpenAI's hardware plans are interpretations; the interview does not independently establish the facts or an outcome.

Lemkin distinguishes hiring people for experience from taking confidential material, while acknowledging that the boundary can be contested in practice. The speakers discuss employee mobility and California law generally; their conversation is not a legal ruling or legal advice.

Lemkin speculates that the dispute could contribute to a pause in OpenAI's hardware work. O’Driscoll agrees with his strategic preference for focusing on coding and enterprise AI. Neither has established that the company has cancelled or delayed a product.

The panel moves to Meta's Muse Spark release and its paid model API. O’Driscoll sees a new low-priced competitor with substantial resources; his account of Meta's business model and the consequences for rivals reflects his interpretation at the time.

He says he tried the product as a consumer but regards performance on coding tasks and independent evaluations as important tests. A product launch or an announced benchmark does not by itself establish lasting superiority or a profitable business.

Lemkin expects companies to use lower-cost models for some work as employees encounter token budgets. He sees a competitive battle for less expensive tasks, but says the eventual revenue and margins in that segment are unknown.

Discussing a Databricks paper, O’Driscoll highlights cost per completed task rather than cost per token alone. Models may use different amounts of reasoning and perform differently across tasks, while the surrounding harness can also affect total cost and performance.

Stebbings cites a company post describing a large increase in AI spending. Lemkin says complex workflows can consume substantially more tokens even if unit prices decline, and describes developers running multiple agents and alternatives. The cited growth is the speaker's account of a post, not an audited industry statistic.

O’Driscoll frames the management problem as comparing the value of employee time saved with the expense of the tools. AI use can be worthwhile, he says, but an unrestricted workflow can eventually cost more than the work it saves.

Lemkin describes exhausting his own credits while experimenting with a design tool and imagines generating many website alternatives overnight. His example illustrates potential demand and cost, not a measured prediction of how much every designer will spend.

Asked about Figma, Lemkin says AI design tools can help people get further without a full-time designer. He does not claim that enterprise Figma is already displaced; he worries that losing smaller users could weaken the future customer pipeline.

O’Driscoll compares estimates of AI company revenue with software engineering spending and asks whether enterprise budgets could become a constraint. He explicitly says he has no conclusion; his employee counts, wages and revenue estimates are discussion figures, not a verified market model.

Lemkin agrees there are limits to what businesses can spend, while O’Driscoll notes that very fast growth can continue even as the percentage rate slows. Their numerical examples are scenarios and do not establish current revenue or future market size.

The guests consider additional demand from AI features embedded in software beyond coding. Lemkin suggests a possible share of software revenue devoted to models; O’Driscoll treats it as another hypothetical market, while cautioning that the assumptions and limits still need examination.

Asked whether labs are moving into legal and life sciences because they have reached existing market limits, O’Driscoll says there is no evidence in the discussion that such a ceiling has been hit. He speculates about research motivations without claiming to know founders' private reasons.

The conversation turns to SK hynix's reported US listing and the memory market. O’Driscoll describes strong demand and stock volatility, but says trading moves can reflect technical factors as well as changes in the outlook for AI spending.

He contrasts a bullish view of low valuation multiples and sustained demand with the historical cyclicality of memory capacity and pricing. His figures and expectation that the cycle could last longer are market opinions, not independently audited financial results or guaranteed returns.

Lemkin refers to IBM's reported explanation for weaker results and wonders whether rising memory costs have crowded out other purchases. O’Driscoll agrees this is a plausible budget mechanism, while the interview does not prove the cause of any company's sales or share-price movement.

Stebbings asks whether an investor shifting from early-stage to later-stage opportunities signals a broader venture change. Lemkin sees it as a sign of the appeal of larger checks and liquidity, based on his interpretation of that investor's public comments.

O’Driscoll separates a structural expansion in late-stage private financing from cyclical valuation risk. More companies can raise large private rounds, he says, but an attractive recent period does not guarantee that entry prices will continue to support returns.

Stebbings says private secondary markets have become more liquid for some prominent companies and that fast-growing startups can reach large valuations quickly. These are his observations about accessible opportunities, not a guarantee that every holder can sell or realize a quoted valuation.

O’Driscoll adds that early and late investing demand different advantages. A firm skilled at being the first check may not automatically have an edge buying a mature private company, even if late-stage investing has recently performed well.

The guests discuss a public comment about exceptional monthly startup growth and the role of Y Combinator. O’Driscoll credits the accelerator's platform for making investment access repeatable, while the guests' opinions about individual investors and fee terms are not independently verified.

The panel turns to criticism of a shopping startup's affiliate attribution practices. Stebbings presents the reported allegation; the discussion does not establish liability, intent or the accuracy of every claim about the company.

Lemkin says the criticism should be viewed in the context of wider affiliate practices, while O’Driscoll says problematic behavior should stop. Their generalizations about other companies are opinions and do not prove that the same conduct is universal.

The guests broaden the discussion to data sourcing and business practices at startups. They debate whether widespread practices justify taking risks and distinguish hypothetical boundary-pushing from the allegations in the Apple case. Their comments do not establish illegality by unnamed providers.

Stebbings cites reported seed-round pricing data, including much higher valuations at the top of the distribution. O’Driscoll attributes part of the divergence to capital-intensive model and infrastructure companies; the figures are cited in conversation, not independently verified here.

Lemkin adds that large funds may combine what once would have been multiple rounds to reach an ownership target. The guests see both capital needs and competition for founders as possible explanations for high prices, while noting that valuation depends on supply and demand.

Stebbings mentions a reported financing for Chai Discovery and discusses investor demand for AI biology companies. O’Driscoll says it remains to be seen whether those companies follow the trajectory of model businesses or the very different development path of biotech.

Lemkin argues that announcing a large financing can help recruiting but may also attract competitors. Stebbings describes investor interest spreading from one heavily demanded round to adjacent companies; those are their interpretations, not evidence of a particular investor's private decision.

The panel discusses investing in companies that may compete. O’Driscoll distinguishes board-level early investing, where confidential information and conflicts can matter, from late-stage passive positions with limited information rights. Their remarks are general views rather than advice on a particular fund's obligations.

Stebbings raises Constellation's reported acquisition of TouchBistro and compares a reported purchase price with annual recurring revenue. O’Driscoll interprets the situation through slow growth and debt; the interview does not independently verify the full transaction terms or capital structure.

He argues that debt can create divergent incentives between senior creditors and equity holders when growth slows. Lemkin regards the reported transaction as a cautionary example and gives his own view of a low revenue multiple, not a universal valuation rule.

Discussing software rollups, the guests debate whether AI could accelerate decline in mature products. Lemkin cites his own experience with a marketing product and predicts a difficult future; these are anecdotes and forecasts, not proof that any named company's revenue will reach zero.

Stebbings mentions Greylock's reported new fund and praises its size relative to the partnership. The guests debate whether fund size and fundraising cadence reflect discipline, incentives or a choice about what investment strategy a firm can execute.

O’Driscoll argues that a firm should raise capital suited to its strengths and long-term strategy rather than simply match peers. Lemkin emphasizes the economics of deploying a fund and raising the next one; neither offers a guaranteed return or an independent assessment of Greylock's fund.

The guests close by reflecting on the importance of surviving market cycles and preserving an investment franchise. Their discussion returns to the tension between the current scale of opportunity and the risks of extrapolating favorable conditions indefinitely.