OpenAI’s Restructuring, Mercor’s Valuation and the Math of Venture Returns

Harry Stebbings talks with investors Jason Lemkin and Rory O’Driscoll about OpenAI’s restructuring, venture fund scale, Mercor’s valuation and the economics of investing in fast-growing companies. Their conversation reflects the information and market expectations available in October 2025; their projections and descriptions of private companies should be understood in that context.

The panel starts with OpenAI’s restructuring. O’Driscoll describes the creation of a public benefit corporation alongside the nonprofit foundation and sees a structure that makes future investment easier while retaining a nonprofit role. He regards the change as a major advance for the company’s financing ambitions, although the discussion does not settle every legal or governance question.

The investors trace the structure back to OpenAI’s unusual origin and its need for capital-intensive computing. They argue that a constrained investment structure became difficult to reconcile with the sums needed to train and serve models, and that simplifying it could make the company easier for investors to understand.

The panel recalls the leadership upheaval in which Sam Altman was briefly removed and returned. Their interpretation is that the confrontation exposed a tension between nonprofit oversight and the commercial organization’s need for continuity and capital. This is their reading of events, not a complete account of the governance dispute.

Stebbings asks who wins. The guests discuss Microsoft’s position, the value of the nonprofit foundation’s stake and the potential for OpenAI to raise more money. They stress that governance rights and commercial arrangements matter alongside headline ownership percentages; the value of an illiquid stake depends on the company’s future performance.

O’Driscoll suggests that a foundation holding a large stake could eventually control substantial resources for its mission, but that a valuation on paper is not cash available to spend. How and when it could turn the stake into liquid assets remains a practical question, as do the obligations of the commercial company.

Asked about the losers, the panel points to Elon Musk’s opposition and continuing disputes over OpenAI’s direction. They describe the conflict as unresolved and distinguish the completion of a corporate restructuring from a final resolution of every objection or lawsuit.

The panel considers employee ownership and the effect of a very large valuation on recruiting. Even a small percentage could represent a substantial paper value, they say, but the payoff depends on dilution, liquidity and future performance rather than a headline valuation alone.

Stebbings asks whether OpenAI might go public. The guests speculate that the restructuring could make an IPO possible and that consumer familiarity could create unusually strong retail demand. None of them presents an IPO date or valuation as a settled plan.

The speakers discuss reported primary and secondary transactions at different implied values, including deals associated with SoftBank. O’Driscoll notes that a later private transaction can produce a rapid paper markup on an earlier investment; that is a valuation change, not a realized cash return.

Could OpenAI reach a trillion-dollar valuation in 2026? O’Driscoll says it is conceivable if strong growth and investor enthusiasm continue, while emphasizing that a slowing growth rate can sharply compress a high revenue multiple. The panel’s revenue and valuation figures are contemporary estimates, and the outcome is a prediction rather than a fact.

The guests also question whether an OpenAI browser will become a mass-market breakthrough. Lemkin says that investor excitement around a technology does not guarantee that ordinary users will change their habits; the answer depends on whether the product offers a reason to switch.

Stebbings turns to a16z’s reported $10 billion fundraising across growth, AI applications, AI infrastructure and defense. The guests consider whether a very large venture platform is becoming the norm and whether the individual funds, viewed separately, are as extraordinary as the aggregate headline suggests.

O’Driscoll argues that scale lets a platform serve companies at multiple stages and buy options on future winners. Smaller firms must be more selective about where to compete. A broad platform can build relationships early and then deploy much larger checks as the most promising companies mature.

The guests discuss how a small initial investment can create an opportunity to follow a company closely before a later round. They also note that the option is not free: investing across many companies requires capital and judgment, and participation in a later round is not guaranteed.

The panel credits a16z with developing a distinctive media presence and a suite of services for founders. Lemkin argues that reputation and operational support help the firm become a preferred partner, while the others debate how a large organization maintains quality as it expands.

Large funds also collect larger absolute management fees, which can finance experienced staff and founder services. The speakers debate the business incentives of a platform: for limited partners the important question remains whether the combination of fees, fund size and investment decisions delivers attractive net returns.

The panel discusses bundling multiple strategies under one brand. Offering limited partners access to growth, specialist and earlier-stage funds can create an attractive package, but Stebbings and his guests suggest investors should examine each fund’s economics rather than assume that a famous name guarantees performance.

Mercor is the next case. Stebbings cites a reported $350 million financing at a $10 billion valuation and a headline revenue figure around $500 million. He notes a possible 20VC connection to the company; the episode does not establish the precise investment relationship or independently verify the reported figures.

The guests describe Mercor’s business as sourcing and matching specialized people to work for AI labs, including expert evaluation and model-improvement tasks. They see demand for domain expertise as a consequence of the increasingly sophisticated work model builders want humans to review or supply.

The business’s headline revenue needs interpretation. The panel asks how much represents gross customer spending, what is paid to workers, and what Mercor retains after direct costs. Without clear and comparable figures, applying a software revenue multiple to the entire headline number could be misleading.

The investors also discuss concentration risk: a small group of AI labs can account for a large share of demand in this market. Their comments about customer mix are estimates, not audited disclosures. They ask what happens if labs change suppliers, automate tasks or reduce spending.

O’Driscoll cautions against treating early AI service revenue as proof of durable software-like margins. Labor marketplaces can have real value and rapid growth, he says, but investors need to understand gross profit, customer retention and whether demand persists beyond a particular model-training cycle.

The panel nevertheless sees a potentially large market. If frontier labs continue to demand expert human input for years, a company that recruits the right people and manages quality could build an important business. The duration and magnitude of that demand remain uncertain.

Stebbings asks whether a $10 billion entry price leaves room for a venture return. The discussion turns to the size of a plausible exit, dilution and the amount of capital a business needs. A company can be impressive while an investment at a particular price still has difficult return mathematics.

The guests distinguish the company’s operating performance from the growth in its valuation between funding rounds. A high mark can be attractive to existing investors, but new investors must look forward from their own entry price. Future exits and the multiples discussed here are hypothetical.

They broaden the point to large private companies and financing behavior. Stebbings asks how much cash a rapidly growing business really needs when valuations rise quickly. The answer depends on burn, profitability, ownership goals and whether another financing creates operational flexibility or simply more dilution.

The panel discusses Ramp as another example of rapid private-market appreciation, using reported valuation and revenue figures rather than audited results. O’Driscoll says a large private company may behave more like a public mid-cap stock: investors should not assume another immediate threefold increase simply because past rounds moved quickly.

The guests consider the difference between a high annualized paper return from a quick markup and a large multiple on invested capital. They argue that the former can look remarkable while the absolute gain remains modest, especially when a fund has a long life and capital is still tied up.

Stebbings introduces a set of Carta data on 547 companies that raised Series B rounds in 2018. The panel cites approximately 18% reaching more than five times and about 10% reaching more than ten times, with a large group below two times. These are figures as presented in the discussion, with cohort and methodology limitations.

The distribution shows why a few exceptional outcomes matter so much, the guests argue. They mention Figma as a particularly large outlier in the dataset. The data is not, by itself, proof that every diversified strategy succeeds: entry prices, follow-on decisions, ownership and exit values all change the result.

O’Driscoll runs an illustrative calculation of what the distribution might mean for a portfolio and its investors after fees. His gross and net estimates depend on assumptions about investment sizes and fund economics; they are a thought experiment, not reported net returns from the Carta cohort.

Stebbings asks whether the numbers support a spray-and-pray strategy. The guests distinguish buying many positions from systematically winning access to strong companies and allocating more to the best performers. A broad portfolio can provide options, but it still requires decisions about where to put additional capital.

The panel revisits option checks. An investor may place a relatively small initial bet, watch the company develop and then try to increase exposure when evidence improves. The price of that information is the early capital at risk, and the ability to follow on depends on competition and available allocation.

The math is sensitive to the size of the winner as well as the number of losers. A spectacular return on a tiny holding does not necessarily rescue a fund; ownership, capital allocation and whether the investor can hold through subsequent rounds determine how much of that upside reaches the portfolio.

The guests discuss the emotional difficulty of seeing companies they passed on become valuable. They say even experienced investors miss winners, and a viable investment process requires learning from mistakes without assuming that every missed deal was knowable at the time.

Stebbings turns to the choice between selling a promising company and continuing to build. In the panel’s account, Synthesia reportedly received a roughly $3 billion offer from Adobe and declined it. The discussion treats the offer and revenue figures as reported or asserted, not independently confirmed deal terms.

Lemkin considers how a board should weigh a large offer against the company’s potential. The decision is not purely a spreadsheet exercise: founder ambition, employee liquidity, the available growth runway and personal risk tolerance all influence whether a sale feels right.

The investors disagree about how much outside shareholders should push a founder toward a sale. They distinguish a founder who wants to keep compounding from one who is exhausted or lacks the capital to compete. Scale has invested in Tavus, a company the panel identifies as competing in a related area, a relevant interest in this discussion.

The panel argues that rejecting an offer creates a new obligation to build enough value to justify the risk. That does not mean an offer sets the company’s future ceiling: a valuable acquisition proposal can reflect both the buyer’s strategic interest and the seller’s genuinely large opportunity.

Continuing independently also introduces decisions about liquidity. The guests compare staying private, raising more money and eventually going public. Each path has tradeoffs for employees and investors, and each can leave a founder responsible for a much larger organization than the one they originally built.

Stebbings and the guests discuss the personal cost of being the leader who insists on a larger outcome. They note that determination may create exceptional value but also prolong uncertainty for employees and investors who might otherwise receive cash from a sale.

The conversation turns to the proposed Amazon acquisition of iRobot, which did not close amid regulatory opposition. The panel describes the abandoned deal as an example of the uncertainty between signing and completion, and debates whether founders should accept a strong strategic offer when one appears.

The speakers consider the downside of relying on a transaction that may not close. An acquisition agreement may include protections, but a company can still face a changed market and operational difficulties after a deal falls apart. The panel does not establish the legal rights of the parties in the iRobot case.

O’Driscoll explains why internal rate of return can flatter a short holding period. An annualized rate measures the speed of a return, whereas the multiple measures how much capital comes back. Both are useful, but the cash actually distributed and the timing of that cash matter to limited partners.

The panel turns to Amazon and reports of corporate layoffs. The guests debate whether the company’s AI position calls for stronger action while also recognizing that retail, cloud and AI infrastructure are different businesses with different competitive strengths. Their comments on management strategy are opinions and forecasts.

O’Driscoll sees two distinct challenges: adjusting retail operations after pandemic-era expansion and strengthening AWS’s position in generative AI. He points to logistics and robotics as potential retail advantages, while arguing that cloud leadership in the prior era does not automatically guarantee leadership in the next one.

The guests speculate about whether Jeff Bezos might return to a more direct role. They present this as discussion, not a report of an impending management change. O’Driscoll cautions against overreacting to a cluster of negative headlines and notes Amazon’s continuing strengths in distribution and cloud infrastructure.

In a final rapid-fire comparison, Stebbings asks whether investors should prefer Brex at a lower proposed valuation or Ramp at a higher one. The panel uses figures discussed in the episode and says the missing variable is growth: the premium that is reasonable depends on how much faster the higher-priced company can expand.

O’Driscoll frames the question as finding the growth rate at which the two investments would be equally attractive. Revenue, growth, margins and entry price interact; a lower multiple is not automatically superior if the other company compounds much faster, and the financial figures used are not independently verified.

Stebbings praises a16z’s partners and says he has invested alongside them, a relevant interest when assessing his view. O’Driscoll credits the firm’s founder services and ability to organize specialized teams at scale, while noting that there is still a question about how far venture can scale profitably.

Lemkin says the key competitive position is becoming one of a founder’s first choices at the relevant stage. Large platforms can be credible across many stages, but smaller firms can still win by becoming exceptionally valuable to a particular group of founders.

The final question asks whether the guests would prefer to own Anduril at the valuation mentioned in the discussion. Lemkin declines on personal grounds, saying that he is not interested in building weapons. He jokes about the social cachet of a fashionable investment, illustrating that investors’ motivations are not always purely financial.

The guests close by reflecting on self-knowledge and the choices that make investing personally meaningful. Their rapid-fire comparisons are expressions of preference and speculation, not recommendations or independently verified valuations.