SpaceX Valuations, Harvey’s Growth and the Risks Behind AI Deals

Harry Stebbings opens with reports of a SpaceX secondary sale at an $800 billion valuation. Rory O’Driscoll considers SpaceX’s rocket and Starlink businesses impressive but says the reported price assumes a great deal of future performance. The company’s revenue, growth and valuation figures are the panel’s recollections, not independently verified financial disclosures.

Jason Lemkin wonders whether recognizable names such as SpaceX could attract unusually strong public-market demand. Stebbings asks whether large private secondaries might instead reduce the need to list; both are possible scenarios, not confirmed IPO plans.

O’Driscoll says a high private-market price can make a lower IPO valuation feel disappointing even if the listing succeeds. He distinguishes the company’s ability to continue operating from the losses a late private investor might suffer if the public market values its stake less highly.

Lemkin believes founders and investors have become more willing to accept the possibility of a down round. O’Driscoll agrees that price risk is part of late-stage investing, while noting that repeated markdowns could eventually make private investors more cautious.

The group questions traditional advice not to raise at an excessively high price. Lemkin thinks some founders now prioritize access to capital even if a later financing values the company lower; that observation does not remove dilution or investor-loss risks.

On the 2026 IPO market, O’Driscoll constructs a hypothetical combining SpaceX, Anthropic and Databricks valuations. His arithmetic is an illustration of potential scale, not evidence that those companies will list, that investors can immediately sell their shares or that the assumed valuation would become cash returned.

The panel says a few very large offerings could dominate annual IPO totals. O’Driscoll adds that even a strong year for liquidity could leave investors who purchased at the last private price below their entry point.

Lemkin admits he was too optimistic about previous IPO timing and predicts Anthropic and Databricks may list in the second half of 2026. These are December 2025 forecasts, not company announcements, and depend on market conditions.

O’Driscoll considers whether access to public capital can become strategically valuable for businesses with large financing needs. The guests’ revenue, growth and future-valuation examples are hypothetical and should not be read as verified company guidance.

The investors compare different revenue multiples and growth trajectories for SpaceX and Anthropic. O’Driscoll points out that the comparison embeds assumptions about persistence, profitability and a perceived founder premium; it does not establish what either company is worth.

Stebbings turns to Netflix’s proposed Warner Bros. Discovery transaction and a competing Paramount approach. O’Driscoll corrects the framing: at the time, Netflix was seeking a deal, not describing a completed acquisition. Completion, terms, competing bids and regulatory outcomes were uncertain.

O’Driscoll sees the proposal as an example of a technology-led distributor gaining leverage in an older media industry. He compares the bidders’ reported financial resources, but the figures and his conclusion about their relative strength are his analysis rather than transaction findings.

The guests discuss regulatory review and competing definitions of the relevant market. O’Driscoll says streaming concentration could be assessed differently from the broader entertainment market; that is a possible legal debate, not a prediction of an agency decision.

He also raises the possibility of political influence and concerns about fewer buyers for creative work. These are his interpretations of the public debate, not proof of improper conduct or a determination that the proposed transaction would harm consumers or creators.

Lemkin speculates about how Warner Bros. Discovery’s management might prefer one bidder over another. He does not know the executives’ private motivations, and the panel notes that the interests of shareholders, executives and competing bidders need not always align.

The investors discuss whether a buyer with a higher trading multiple and global distribution could justify a higher offer. These are theoretical financing and distribution arguments; neither establishes that the deal is accretive or that any particular buyer will prevail.

O’Driscoll places the debate in a wider history of digital distribution reshaping advertising, retail and entertainment. He wonders whether financial services and other industries might follow, presenting a long-term thesis rather than a fixed forecast.

Stebbings asks about Tiger Global’s reported smaller fund and slower investment pace. Lemkin argues that growth investors adapt their strategy to market conditions, while O’Driscoll interprets a more concentrated approach as a possible return to an earlier style. The fund size and deal count are reported estimates.

O’Driscoll praises the willingness to continue investing after a difficult period. His account of Tiger’s leaders’ personal feelings and the performance of earlier funds is speculation, not a disclosure from the firm or a verified fund-return calculation.

The panel discusses a reported general-partner commitment of roughly 20% and the tradeoff between investing one’s own capital and raising outside money. Their simplified arithmetic omits fees and other fund terms, and the actual commitment is not independently confirmed.

O’Driscoll says a large personal commitment can align a manager with limited partners and enable larger late-stage checks. Lemkin agrees that the ability to deploy a larger check can matter, while noting that the economics may look different for an early-stage strategy.

Stebbings raises a reported $500 million seed round at a $5 billion valuation for a startup associated with former Databricks AI leader Naveen Rao. The guests praise a founder’s track record but do not establish the actual financing terms or predict a particular return.

Stebbings says some headline seed rounds can follow earlier, lower-priced commitments. O’Driscoll explicitly says he does not know if that occurred in this specific case; the examples of blended entry prices are hypothetical and should not be assigned to the named company.

The guests debate whether staged commitments distort headline valuations or simply allow new investors to accept or decline a stated price. They compare the practice with earlier accelerator financing, while recognizing that different investors may have different cost bases.

Stebbings introduces a reported Harvey financing of $160 million at an $8 billion valuation, along with revenue and retention figures. Those figures are stated in the interview, not independently verified financial results; the panel also has interests in adjacent legal AI companies.

O’Driscoll calculates that $160 million is about 2% of an $8 billion valuation and discusses whether a small financing can set a meaningful mark. Lemkin notes that limited partners may apply rules when deciding whether to recognize a valuation increase; actual fund accounting depends on the facts and applicable standards.

The investors ask what size and competitive position would justify a return at that entry price. Stebbings points to competition and investment interests in adjacent legal niches; the claim that any vendor has won a geography or a market remains the panel’s opinion, not an established market-share finding.

Lemkin argues that unusually strong revenue growth and retention, if durable, can support an aggressive growth investment. His calculations depend on the interview’s reported metrics and a hypothetical continuation of growth; they are not verified forecasts.

O’Driscoll says the important question is how quickly a very high growth rate decelerates. A company can grow impressively and still disappoint an investor who paid too far ahead of future revenue. He cautions that a great company can nevertheless be a poor investment at a particular price.

Stebbings works through a hypothetical fall from tripling to doubling and then slower growth. The guests use the scenario to show that a valuation multiple can look very different depending on which future revenue year an investor chooses as the denominator.

Lemkin compares the risk with paying several years ahead during the 2021 market. O’Driscoll agrees that the earlier period shows why high entry prices can be dangerous, while arguing that market size and competition remain central to the current legal AI thesis.

The panel considers whether faster or more capable foundation models could make existing AI applications less distinctive. Lemkin warns of a possible step change; Stebbings argues that customer relationships, integrations and workflows can create defensibility. Both are describing competing hypotheses.

Lemkin says customers could switch if a future provider were dramatically better. His examples of latency, setup and error rates are personal observations, not measured comparisons; the guests do not demonstrate that any named product will become obsolete.

O’Driscoll agrees that model changes can force application companies to revisit product-market fit, while saying established workflows may provide some protection. He suggests teams evaluate new models quickly because capabilities and costs are changing.

Lemkin predicts that much faster deep reasoning could change business applications. His imagined reduction from minutes to seconds is a future scenario, not a demonstrated capability or a guarantee that incumbent vendors will lose customers.

The speakers discuss routing between model providers. O’Driscoll says APIs can make switching easier than moving an entire cloud deployment, though real integrations and customer requirements can still impose costs. Stebbings says some companies he has spoken with changed providers after model updates; those anecdotes are not market-wide measurements.

O’Driscoll challenges the idea that a replaceable model must necessarily be a bad business. He uses commodity industries as an analogy and speculates that high fixed costs or consolidation could produce a smaller number of viable suppliers. The eventual market structure is unknown.

Stebbings argues that consumer memory and habit might make an assistant sticky, while model choice in business applications can change more quickly. The panel disagrees about how easy it would be to replace a familiar consumer service; these are product judgments rather than measured switching rates.

Lemkin claims that some startups use Chinese-origin open models to reduce costs. Specific named-company usage and percentages are not independently verified, and model origin, hosting location, data handling and security are separate questions.

O’Driscoll corrects a statistic attributed to a venture firm: he says the reported 80% concerned the Chinese share of open-source model usage within a subset, not 80% of all portfolio-company AI usage. The guests do not provide the underlying dataset or methodology.

The panel distinguishes legal and security questions from business considerations when evaluating models. O’Driscoll says applicable rules and practical data handling should be assessed; his comments are not legal advice and do not establish any particular model’s security.

Stebbings discloses an approximately $7 million investment in Airwallex and a relationship with Keith Rabois before discussing public allegations about data flows. The program does not independently verify those allegations or show that customer data was improperly accessed.

The guests compare reported Airwallex and Ramp valuations and revenue, with different views on whether geography, product mix or investor preferences explain a discount. Those financial figures and like-for-like assumptions are unverified, and Stebbings has a disclosed financial interest.

O’Driscoll distinguishes a company’s place of incorporation and ownership from the locations of staff and infrastructure. He says cross-border operations can create legal and due-diligence questions, but the panel does not establish the company’s actual data flows or the legal obligations of any specific deployment.

Lemkin says a company facing such questions should be able to explain and substantiate its data practices. He also worries that ethnicity can shape the debate; O’Driscoll emphasizes security and regulatory concerns. Their disagreement does not establish discriminatory intent by any named person or prove a security incident.

Asked to choose between reported Airwallex and Ramp valuations, O’Driscoll prefers a hypothetical Airwallex investment under assumptions about comparable growth and a change in geographic exposure. Lemkin prefers trusting management to address the issue. Neither has demonstrated that the assumptions or proposed change are necessary or achievable.

The guests argue over a board’s role: O’Driscoll says boards should raise strategic risks that can affect a future sale, while Lemkin warns against investors prescribing operational decisions to capable executives. They agree that the discussion represents a substantive difference in judgment.

The conversation ends with prediction markets. The guests discuss anonymous trading patterns and possible information advantages, but the program does not establish who traded, whether anyone had confidential information or whether a violation occurred.

O’Driscoll distinguishes a trader possessing nonpublic information from someone who can manipulate an outcome. He anticipates scrutiny of prediction markets and sports micro-bets, but the precise legal rules depend on the contract and jurisdiction; the panel’s legal characterizations are not advice or findings.

The guests close by noting that trading on information and influencing an event can pose different integrity questions. Their imagined examples are scenarios for debate, not evidence of misconduct by any person or platform.