OpenAI’s Jony Ive Deal, YC’s Scale and the Math Behind Venture Returns

Harry Stebbings is joined by investors Rory O’Driscoll of Scale and Jason Lemkin to discuss venture returns, IPOs, Y Combinator and OpenAI’s hardware ambitions. Their conversation combines reported developments with estimates and predictions made at the time of recording.

The panel begins with Builder.ai, which Stebbings says raised around $500 million before shutting down. He refers to reports of missed revenue projections and raises the possibility of fraud; the discussion does not establish that allegation as fact.

Lemkin asks how to assess a reported loss of more than $100 million for Insight Partners. O’Driscoll notes that a check of that size is painful but represents a small fraction of a multibillion-dollar fund, and a large investor should expect some investments to fail.

The same investor’s reported Hinge Health outcome invites a different question: how much does a successful exit move a very large fund? O’Driscoll argues that a later-stage strategy may depend on several strong returns rather than requiring each winner to return the entire fund.

O’Driscoll illustrates a model in which a portfolio of roughly 20 deals has losses, modest wins and a small group of large winners. The example is a planning framework, not a promise of returns, and unusually large outcomes can make a fund much better than the baseline.

The strongest deals have the greatest effect on the overall result. A slightly better recovery on a weak investment helps less than a major difference in the tail of a winning investment, while assuming an exceptional outlier in every fund would be dangerous.

For the biggest funds, O’Driscoll argues, the limited number of very large exits makes the arithmetic harder. One response is to put more capital into the strongest apparent winner, creating concentration risk that a seed investor with less information may be unwilling to take.

The panel discusses the attraction of fast growth to $100 million in annual recurring revenue. O’Driscoll calls traction a useful proxy for commercial success but not conclusive on its own; churn, durability and the economics behind revenue still matter.

Lemkin adds that hot AI companies have an advantage in recruiting and raising capital. A slower-growing business may need its own talent pool and financing strategy if investors and engineers gravitate toward the most visible hypergrowth opportunities.

O’Driscoll says boards should assess employee dilution alongside attrition and offer acceptance rates. Equity grants are costly to existing holders, but if a company cannot hire or retain people, simply minimizing dilution can also be a mistake.

The recent Hinge Health and MNTN IPOs lead the panel to conclude that a public market existed for some growing companies at the time of the conversation. They describe businesses with hundreds of millions in revenue and meaningful growth, rather than a universally open window for every startup.

O’Driscoll studies how liquidation preferences and conversion terms affect an IPO below a late private valuation. He says Hinge Health’s 2021 financing at $6 billion created a complicated negotiation for investors whose preferred shares did not automatically convert on the same terms as common stock.

He describes some late-stage holders negotiating an economic arrangement and others retaining preferred shares. The precise rights in the financing documents matter: a high previous valuation alone does not tell an investor what will happen at the IPO.

For Chime, O’Driscoll says the documents he examined provide for mandatory conversion of the later preferred rounds above a specified threshold. He uses a possible $12 billion valuation against a $25 billion entry price as an illustration, then stresses he does not know where the shares will trade.

The possible outcomes therefore range from holding preferred shares to negotiating a conversion or accepting a mandatory conversion and a lower mark. An accounting markdown and a realized loss on a sale are different events; the ultimate result depends on price and subsequent trading.

The panel turns to research suggesting accelerators and incubators participate in a substantial share of venture deals. Lemkin argues that Y Combinator has become larger and more attractive to founders, particularly as it has embraced AI.

O’Driscoll calls YC an unusually strong equity business because it has built an enduring system for helping founders become fundable. He argues that its brand and operating model can continue attracting companies beyond the performance of any single investor.

He estimates that YC’s structure could give it a considerable return advantage over a conventional seed fund, using a hypothetical comparison of 6x and 3x. This is his model based on assumptions about pricing and success rates, not a reported audited result.

Stebbings discusses Project Europe, a program in which he has an interest, and says it received thousands of applications. The guests see a need to make it easier for first-time founders to start companies, even though many of those companies will not succeed.

Stebbings likens YC to both Chanel and Walmart: it combines an aspirational brand with broad scale. O’Driscoll agrees that this combination is hard to build and contrasts an enduring product for founders with the interchangeable nature of many venture funds.

Stebbings offers an explicitly unsubstantiated hypothesis that Series A is especially difficult. He contrasts a seed round based on a compelling story with the need at Series A to demonstrate sustainable, high-quality commercial traction.

O’Driscoll agrees that competition for the most sought-after AI deals is intense even as many other companies struggle to raise capital. He wonders whether less obvious opportunities could be overlooked while investors crowd into the same favored categories.

Lemkin compares his first investment in RevenueCat with a more recent early-stage deal at a much higher entry valuation. He and Stebbings both disclose that they are RevenueCat investors; the comparison illustrates his concern about changing seed-stage economics, not an identical risk assessment.

O’Driscoll suggests that inflation and economic growth partly offset the comparison, though he agrees that higher entry prices can increase the risk per dollar. Maintaining the same ownership percentage often requires writing a larger check.

Lemkin says he underestimated how employee equity dilution compounds while companies stay private for many years. O’Driscoll agrees that small annual grants add up, but says companies also need to retain employees and motivate founders.

The board’s task is to balance refresh grants and hiring needs against the cumulative reduction in existing ownership. The panel treats high AI-sector grant rates as examples and claims from conversations, not a universal measured rate.

The panel considers OpenAI’s announced acquisition of Jony Ive’s hardware venture, valued in the discussion at roughly $6 billion. They debate whether the bet will create a new consumer interface for AI and what the deal says about talent, financing and competition.

O’Driscoll thinks a platform company may reasonably explore its own hardware but predicts that, like many earlier attempts, the product could fizzle over several years. He frames this as a base-rate judgment, not knowledge of the unreleased device.

Lemkin is much more optimistic. He predicts a launch within a year, a heavily subsidized device and widespread daily use, while acknowledging he does not know the form factor. Those prices and adoption figures are his speculation.

Lemkin says his own AI usage and experiences with note-taking tools make an always-available assistant plausible. O’Driscoll counters that a tool running on a phone does not establish that consumers will buy and routinely carry a separate device.

The panel discusses the reported size of the acquisition relative to OpenAI’s valuation and the dilution involved. O’Driscoll contrasts compensation for scarce design talent with the capital an investor might contribute, using it to illustrate bargaining power in the AI market.

Stebbings argues that hardware also gives OpenAI a persuasive story for raising further capital. Lemkin imagines AI expanding from minutes of use to much more of the day; the panel treats that scenario as a possible future business case, not current revenue.

The conversation moves to European and American technology ecosystems. The panel compares fintech opportunities, noting that the strength of incumbent banks and the economics of payment networks can differ across markets.

Lemkin describes the pull of San Francisco’s dense AI founder community. Stebbings argues that London has strong AI talent and may offer lower hiring costs; he cites ElevenLabs, Synthesia and Granola as examples of a local ecosystem.

Lemkin says the Bay Area’s intense comparison with peers can motivate founders. Stebbings replies that he values internal drive, while O’Driscoll argues that differences in financing, networks and the ability to recover from failure matter more than innate talent.

The panel discusses public statements from Duolingo and Klarna about AI and staffing. Lemkin predicts significant layoffs over the next two years while suggesting aggregate headcount could remain flat; he bases the view partly on conversations with executives.

O’Driscoll agrees that executives face competing pressures from investors and employees, but expects a more gradual employment effect than Lemkin does. He forecasts slower hiring and organizational changes over several years, not necessarily immediate mass layoffs.

In a prediction round, O’Driscoll says the definition and timing of AGI are disputed and speculates that contractual incentives could shape a declaration. Lemkin explicitly says he is not an expert and predicts AI could feel like AGI in 2026 and be broadly recognized around 2028.

The guests also discuss possible corporate and personal tax changes under legislation being debated at the time. Their comments are contemporary interpretations and forecasts, not a definitive account of enacted law or individual tax liabilities.

On whether anyone will reach a $500 billion net worth, the guests disagree about timing and public-market returns. O’Driscoll notes that private-company valuation changes could produce a large paper net worth even without a comparable rise in public stocks.

The panel cites an SVB analysis of US technology unicorns and estimates that only a minority meet a demanding combination of revenue, growth and profitability. Their further claim that many are worth less than $1 billion is an inference, not a valuation of each company.

Stebbings and the guests discuss the gap between hundreds of private unicorns and the number of attractive exits. They note that the IPO market absorbed far more deals in 2021, but disagree about how many current companies can meet investors’ present expectations.

O’Driscoll concludes that public-market demand can return, but the key question is how many companies have the scale, growth and profitability to meet the new bar. The panel closes by acknowledging that the answer will determine how quickly the backlog clears.