Circle’s IPO, Anduril’s Funding and the AI Budget Debate

Harry Stebbings and investors Rory O’Driscoll and Jason Lemkin discuss IPO pricing, late-stage venture funding, AI's effect on software spending and the outlook for Elon Musk's companies.

The panel starts with Circle's IPO and its sharp early trading rise. O’Driscoll says the offering reopened the debate over whether sellers and the company could have obtained a higher price; the figures reflect June 2025 trading, not current prices.

He estimates that a substantial number of shares sold in the offering were secondary sales and illustrates the difference between the offer price and a later market price. The arithmetic is hypothetical and does not mean every share could have sold at the peak.

Lemkin describes Circle and CoreWeave as strong companies whose shares may also have attracted speculative enthusiasm. He contrasts them with other recent listings whose initial moves he sees as less extreme.

O’Driscoll argues some discount may be needed to induce investors to buy an untraded stock and bear risk. The difficult judgment is how much institutional and retail demand has been underestimated.

He describes an information imbalance: bankers price offerings repeatedly, while founders may experience the process only once. Their recommendations about investor demand and who will hold shares can be hard for a company to test.

The panel considers direct listings, auctions and SPACs as alternatives, each with practical limitations. O’Driscoll says the traditional IPO process has flaws but is hard to replace for a company that needs new capital.

Lemkin cites reported interest in the upcoming Chime IPO and cautions that oversubscription figures reflect orders that can exceed what investors actually expect to receive. This was a pre-listing discussion.

O’Driscoll agrees that headline order multiples can be misleading. Equity-capital-markets teams must judge which orders are genuine and which investors may sell quickly after trading begins.

He says a very large company may have more flexibility to use an alternative listing process, while smaller issuers have less room to risk a failed debut. His view is about negotiating position, not a guarantee of demand.

Stebbings asks whether stronger offerings will bring Stripe, Databricks and Figma to market. O’Driscoll thinks the largest companies have more discretion over timing, while Figma might view a listing differently after its abandoned Adobe deal.

Lemkin says positive market sentiment can change board discussions even for companies that could have listed earlier. O’Driscoll cautions that timing and demand are difficult to predict.

They cite CoreWeave's changing valuation before and after its listing as an example of how sentiment can shift. O’Driscoll says investors should recognize how little certainty they have about near-term trading.

Lemkin says the listing and subsequent financing helped address a debt concern at CoreWeave. The panel offers a contemporary interpretation; it does not establish that all financial risk disappeared.

O’Driscoll argues US public markets can provide substantial equity and debt capital to companies at scale. He notes that he did not buy CoreWeave shares despite their subsequent rise, illustrating hindsight's limits.

The group turns to companies choosing US listings. O’Driscoll cites rough population, GDP and stock-market shares to argue that US capital markets are unusually deep; these are his broad comparative figures, not a precise common-period dataset.

They discuss why international companies might seek US investors and liquidity. Public-market depth can support growth, but listing location also involves company-specific choices and trade-offs.

The investors consider private companies that remain highly valued without clear liquidity. Prolonged holding periods change the expectations of employees and investors who once anticipated a quicker exit.

O’Driscoll sketches an illustrative distribution: some unicorns may fail, some may be acquired or combine with others, and a smaller group may become public. His percentages are rough guesses, not measured forecasts.

He says employee equity may remain inaccessible for years at companies unable to run tenders or list. That can make startup compensation less attractive for people who need cash for family and housing.

The panel notes that tender offers can provide partial liquidity for strong private companies, but availability may disappear if growth slows. O’Driscoll says liquidity itself can become a recruiting advantage.

Stebbings cites Anduril's roughly $2.5 billion financing and a reported $1 billion Founders Fund investment. The guests discuss it as an example of concentrated late-stage investing, using figures available at the time.

O’Driscoll says Founders Fund has long articulated a concentrated strategy and an interest in national security. He admires the consistency while recognizing that large individual bets create concentrated exposure.

He explains that his own firm has a 10% concentration limit and typically targets around 20 investments. He says its focus on earlier enterprise software deals differs from a fund placing very large late-stage checks.

The panel sketches hypothetical returns if an investment increases substantially in value. The arithmetic depends on the assumed entry price, future valuation and carry; it is not a forecast of Anduril's actual outcome.

Asked what he would do without limited-partner constraints, O’Driscoll says he would mostly keep the strategy his firm knows how to execute. He is considering modestly greater variation in position sizes while maintaining discipline.

He argues that adding seed, growth and late-stage strategies can stretch a firm's capabilities. Founders Fund's breadth impresses him, but he does not assume every manager can replicate it.

Lemkin describes a hypothetical strategy of investing in each round of the strongest seed companies, provided later risk could be shifted to other vehicles. This is a thought experiment, not a recommended allocation.

O’Driscoll responds that very few portfolio companies offer enough later-stage upside to absorb large follow-on investments. Higher entry prices reduce the multiple available even when an early investment was a major success.

He says a growth fund may need to find new companies or increase the number of early investments to generate enough late-stage opportunities. Either choice changes the fund's strategy and risk.

Reflecting on DocuSign, an investment associated with his firm, O’Driscoll says in hindsight he could have invested more in a later round. He warns against learning only from the successful deals where more investment would have worked.

The investors discuss how to learn from both winners and mistakes. A consistent decision process must account for outcomes that could not be known when a follow-on check was written.

Stebbings asks how investors should respond to founders whose business may be attractive but not suited to venture returns. The guests distinguish a judgment about a fund's economics from a judgment about the value of the business.

Lemkin says he might back a slower-growing company if he believes in its founders, market opportunity and entry price. The decision depends on more than a single early revenue number.

O’Driscoll describes his firm's retrospective analysis of investments by growth quartile. He says businesses in the second-highest growth quartile sometimes produced outcomes comparable to the fastest group; this is his firm's experience, not a universal law.

He cites portfolio examples including Bill.com and HubSpot to argue that capital efficiency, market, founder and persistence matter alongside growth. His firm has an investment interest in those examples.

Stebbings raises research suggesting software spending growth slowed in the first half of 2025. The group discusses possible causes and does not establish a single cause from the cited data.

Lemkin worries that AI tools may compete for budgets previously allocated to established software, while consolidation continues. His estimate of spending diverted to Cursor is an interpretation, not measured proof of displacement.

O’Driscoll argues that some software markets were already maturing as adoption increased. He uses a simplified cloud-adoption calculation to illustrate why percentage growth cannot compound forever in a finite market.

He says AI could add to the addressable market if it replaces some labor spending, rather than merely swapping one software subscription for another. The extent of that expansion remains an open question.

Lemkin cautions that there is not yet enough evidence to assume a massive transfer of labor budgets to software companies. He expects adoption and pricing to vary across use cases.

Drawing on his contact-center portfolio exposure, Lemkin claims some customers have automated substantial portions of work without paying proportionately more in software fees. The example is his account and does not establish market-wide job losses or revenues.

O’Driscoll proposes an illustrative two-to-one labor-cost comparison and estimates the potential size of contact-center software and labor markets. He stresses that AI will not automate all labor and expects humans to remain involved.

Lemkin notes that some AI service companies also acquire or operate outsourcing businesses, making revenue categories harder to compare. He wonders whether competition could push AI sales and support prices much lower.

O’Driscoll compares the transition to the long adoption of SaaS: obvious technological potential does not imply every customer changes at once. He expects sequencing to depend on value and organizational readiness.

Stebbings asks about SpaceX, Tesla and Neuralink after the public political dispute involving Musk. The investors consider how customer dependence, brand exposure and government relationships differ across the companies.

O’Driscoll argues SpaceX's launch capabilities may make it difficult for some customers to replace, while Tesla faces a different consumer-brand exposure. These are his competitive assessments, not predictions of specific contract decisions.

Lemkin suggests public attention may move on over time, although he acknowledges uncertainty. The panel discusses Musk's record and future focus as opinions rather than established outcomes.

The guests debate whether wealth or an angry social-media persona affects an entrepreneur's ability to innovate. O’Driscoll cautions that public behavior does not reveal whether a person is happy privately.

In a prediction-market round, the guests consider whether Sundar Pichai might leave Google. They speculate about leadership and succession without claiming knowledge of private plans.

They discuss the New York Times lawsuit against OpenAI, including the possibility of settlement. Their predictions are not a legal finding; settlement and a court victory are different outcomes.

Lemkin says he has an interest as a publisher of SaaStr content and questions how creators should be compensated for AI use. His remarks describe his position and a disputed legal issue, not a court ruling.

O’Driscoll considers whether AI companies need licenses from every news outlet or could rely on a smaller number. He frames this as a commercial question distinct from legal rights and liability.

The final prediction concerns whether Linda Yaccarino will leave X. The guests acknowledge limited visibility into internal management; their judgments about personnel are speculative.

Stebbings thanks the guests and closes the conversation.