Cursor Deal Claims, AI Valuations and the Enterprise Agent Race: A 20VC Roundtable

Harry Stebbings, SaaStr founder Jason Lemkin and Scale Venture Partners’ Rory O’Driscoll discuss a purported Cursor transaction, private AI valuations, design software, enterprise agents and a possible Cerebras IPO. The acquisition, financing, personnel and financial details are the speakers’ accounts or reports, not independently established by this conversation; their market and policy predictions remain opinions.

Stebbings describes an alleged $60 billion Cursor acquisition or option involving xAI and SpaceX, with a purported $10 billion payment if it does not proceed. O’Driscoll characterizes it as an option, while Lemkin says they have not read the deal and he is unsure what is actually being purchased. The transcript does not independently confirm an agreement, price, payment, timetable or closing.

Lemkin cites purported personnel moves, commercial ties and compute commitments between the parties and wonders whether technical milestones matter. He explicitly says he could be wrong and that the deal might not close. These are unverified claims and speculation, not evidence of contract terms or employee conduct.

O’Driscoll says that if the transaction closed at the reported price it would be exceptionally large for a venture-backed private company. He imagines a strategic fit between a coding product seeking compute and a provider seeking customers. His examples of revenue, gross margin, capacity and costs are rough claims or scenarios, not verified company financial statements.

O’Driscoll speculates that combining compute and a coding product could support a prospective SpaceX offering, and Stebbings stresses that the alleged transaction is not a closing. Their discussion assumes an option and a later IPO without seeing the documents. It does not establish either event or the proposed financing structure.

The guests consider whether a delay could give an acquirer time to test the fit and avoid disrupting a possible listing. O’Driscoll says he does not know the motives; Lemkin reads public promotion as a sign of intent. Both are interpretations of reported activity, not knowledge of contractual conditions.

Asked which side would benefit, Lemkin argues that founders could gain liquidity early in a demanding company-building journey. O’Driscoll sees possible operational benefits on both sides but says future shareholders bear valuation and integration risk. No proceeds or founder liquidity are established unless a transaction actually closes on known terms.

O’Driscoll illustrates how an acquirer valued at a high revenue multiple could use stock to buy a lower-multiple business with less apparent dilution. The calculation relies on alleged SpaceX and Cursor values and projected revenues. A favorable arithmetic comparison does not verify the deal or establish the value of the combined companies.

Stebbings asks what investors would receive and when. O’Driscoll guesses consideration would probably be stock, not cash, and discusses potential lockups and price volatility, explicitly saying he has not checked the documents. The actual form of consideration, transfer restrictions and public offering remain unknown.

The panel also wonders whether a previously reported Cursor financing would still occur. O’Driscoll says he does not know and offers a scenario in which it becomes unnecessary. Lemkin speculates about links between the transactions; those assertions are not evidence of a completed round or an agreement between investors.

O’Driscoll sketches different hypothetical returns for early and late investors and explains why a stock lockup could matter more to someone with a higher entry price. He refers to possible investors and holdings but says he does not know the complete cap tables. Paper multiples and possible future proceeds are not realized returns.

The guests debate whether the alleged combination could change an acquirer’s AI narrative. They disagree on who may have influenced the decision and concede integration and valuation risks. Their characterizations of individuals’ motives, control and internal discussions are speculative rather than reported facts.

O’Driscoll says highly valued stock can be an acquisition currency and warns that the conditions permitting such deals may not last. He also points to differences between building rockets and software as a matter of competitive differentiation. The projected prices, multiples and revenue are illustrative, not confirmed terms.

Lemkin predicts a $100 billion acquisition within 12 months, saying several large companies could afford a strategically important target. O’Driscoll takes the other side and thinks the alleged Cursor transaction could remain a high-water mark for a decade. Neither prediction establishes that this transaction closed or that any later acquisition will occur.

O’Driscoll explains his skepticism through the limited number of potential buyers, valuation multiples and regulatory constraints. He admits he can imagine exceptions and agrees that strategic fear can motivate large deals. Lemkin draws on his experience at Adobe to argue that boards may change their price expectations when a new benchmark appears.

The guests discuss how a large reported transaction might affect corporate development. Lemkin argues that leaders defending a business may spend heavily for a scarce capability; O’Driscoll agrees on the pressure but not necessarily the $100 billion threshold. Their comments about private boardroom discussions are general interpretations, not knowledge of specific negotiations.

O’Driscoll discusses reported Cursor investment rounds and estimates returns for several funds, while saying he cannot calculate all the figures. He emphasizes that deploying hundreds of millions at a lower multiple is a different product from investing a small seed fund at a higher multiple. The alleged returns remain estimates and cannot be treated as realized while the transaction and consideration are unconfirmed.

The panel compares the needs of a large endowment, which must place substantial capital, with those of a smaller limited partner seeking high multiples from emerging managers. Access, fund size and liquidity affect the comparison. O’Driscoll returns to the risk that any stock consideration could fluctuate during an unverified lockup.

Stebbings reports that Tim Cook will step down as Apple CEO and that John Ternus will succeed him. Lemkin speculates whether technology leaders are leaving because of AI challenges, while O’Driscoll favors an interpretation of an orderly succession. The interview does not independently verify the announcement, timing or personal motivations.

O’Driscoll praises Cook’s operating record and says he owns substantial Apple stock, a financial interest relevant to his assessment. He cites long-term revenue, market capitalization and share-price growth but the conversation does not independently reconcile those measures. A quiet stock reaction, in his view, may reflect confidence in succession; it does not prove why investors traded.

The guests consider how much urgency AI presents for Apple, Adobe and Netflix. O’Driscoll sees different exposures across hardware, creative software and media, while Lemkin worries competitors may erode usage before subscription revenue declines. These are competitive opinions, not proof that any company lacks an AI strategy or faces a particular outcome.

Lemkin calls his own reduced Netflix viewing a form of ‘silent churn’: he still pays while spending more time elsewhere. He argues that usage metrics can warn about future retention even before revenue falls. His experience is one anecdote, and lower engagement does not establish an individual subscriber or market will cancel.

Stebbings cites reports that Anthropic declined funding at an $800 billion valuation and that secondary indications reached $1 trillion. He relays strong demand from one prospective investor. Neither a declined offer nor a completed transaction at those prices is independently confirmed, and an indication or quote does not establish the value of all shares.

Lemkin says model performance and customer preferences can change quickly, and describes testing his workflows across APIs. He cites reported Codex usage and a prospective agent release, but says he cannot predict which provider will lead in 30 or 90 days. His own test is not a benchmark for all users.

O’Driscoll distinguishes current enthusiasm from a future risk-adjusted investment return. He predicts Anthropic could list while demand is strong, but neither a filing nor an offering date is established. A company can be attractive, trade at a high price and still provide disappointing returns to a later buyer, he argues.

Asked about IPO pricing, O’Driscoll offers an opinion about what investors might pay at the time of recording, not a valuation established by a transaction. He speculates that a public listing could help fund compute and suggests a possible fourth-quarter window. Lemkin asks whether ongoing access to capital would actually be easier in public markets.

O’Driscoll argues public markets could offer equity, convertible and debt financing at scale, although he acknowledges very large private rounds. Lemkin raises a counterexample: a company needing ongoing capital could suffer if its public stock falls sharply. Their descriptions of hypothetical financing are scenarios, not assurances about what any company could raise.

O’Driscoll disputes the analogy to Figma’s share-price path and believes the AI cycle is at a different stage. He says even a lower post-listing price might leave a newly public company with cash already raised. Both the comparison and his prediction about market enthusiasm are judgments, not a guarantee of listing or future financing.

Stebbings introduces Claude Design and asks whether it competes with Figma. Lemkin says his own test found it improved on earlier design capabilities and describes it as an application, while saying it does not match everything Figma or Illustrator can do. His evaluation is based on his experience, not a controlled product comparison.

Lemkin thinks an integrated design-to-production workflow could let some teams bypass parts of a traditional design process. He predicts this could gradually reduce incumbents’ growth without immediately appearing in their next quarter. That is a competitive hypothesis, not proof of customer losses or a forecast that existing products will disappear.

Lemkin recounts his organization’s difficulty with Marketo unsubscribe behavior and alleges a legal compliance issue and inadequate support. The conversation provides only his account; it does not establish a legal violation, the cause of the problem or Adobe’s response. He uses the experience to explain his frustration with incumbent software.

Lemkin says Claude Design can import from Figma and export to Canva, and describes product and engineering teams increasingly working in code together. O’Driscoll distinguishes digital interface design from physical-world design such as posters. Their discussion of product features and adoption is based on their observations and may change.

The guests consider how a bundled tool might take some usage from a specialist without destroying the entire business. O’Driscoll compares the potential range to the historical competition between Google’s office tools and Microsoft Office. He stresses that losing a small share of revenue and losing a large share would have very different implications.

O’Driscoll says the strongest integrated competition may arise when design feeds directly into an already automated software-development workflow. He cautions that previous platform launches have generated excitement without lasting impact. Lemkin believes proximity to an existing coding product could make this investment more durable, but recognizes that future execution is unknown.

Lemkin describes shipping features rapidly and using AI-assisted design before a human designer has time to refine it. He says this is his team’s preference, not a claim that generated work is always production-ready or superior. He expects both established and AI-native companies to face rapid competitive change.

Stebbings cites reported growth-fund raises by Accel and Sequoia. O’Driscoll says abundant capital may reflect the scale of later-stage opportunities but warns that returns can be harder when strong demand and high valuations are widely recognized. The amounts and the future performance of the funds are not independently established.

Lemkin thinks strong relationships may help established firms gain allocations in large private rounds, though he notes that investors are sometimes excluded. O’Driscoll describes intermediaries matching limited-partner demand with scarce access and discusses reportedly high SPV fees. Their comments do not establish the terms of any particular vehicle or a guaranteed allocation.

Stebbings cites reports that Rippling crossed $1 billion in revenue with 78% annual growth. O’Driscoll says, if true, the result challenges the idea that all SaaS companies face the same fate. Lemkin thinks the reported rate may also indicate acceleration, but the panel does not independently verify the figures, timing or revenue definition.

O’Driscoll argues that payroll and financial administration require dependable execution, legal compliance and established infrastructure, making them different from some creative workflows. He predicts demand for specialist providers will remain, while acknowledging AI can affect how software is built and used. That is a market thesis, not legal advice or a guarantee for any vendor.

Lemkin says his team plans an AI finance workflow and would compare providers based on how their APIs work with agents. He says a possible move between vendors depends on that evaluation; it is not an announced completed switch. O’Driscoll agrees interfaces can shift market share while arguing that underlying financial services still need to be provided.

Lemkin alleges problems with his existing marketing tool and says his team is considering moving its data. He describes using Salesforce through APIs without its ordinary interface and praises that experience. These are his organization’s account and product opinions, not independent verification of alleged compliance failures or a completed migration.

O’Driscoll explains ‘headless’ access as separating a user interface from the database and workflow services behind it, allowing another application or agent to interact with the backend. He sees potential value in preserving integrations and data even if people interact differently. His account is a conceptual explanation, not a confirmation of every product or pricing detail.

Lemkin says Salesforce already has a long-standing API and argues the larger opportunity is an enterprise ‘agent fabric’: infrastructure to manage agents, context, permissions and security. He praises Salesforce’s API based on his team’s experience but says he does not know whether the company can deliver the broader vision in time.

The guests discuss why an executive might need to see what autonomous software is doing, audit decisions and control access across systems. O’Driscoll compares the need to the rise of web analytics when retail moved online. These are proposed governance requirements, not claims that any named platform currently provides complete security.

Lemkin thinks large incumbents may have an advantage because they already serve enterprise customers, provided they can execute quickly. He cites HubSpot’s customer base and discloses that Stebbings has invested in Peak when discussing a related category. Their estimates of competitive positioning and customers’ preferences are opinions, not verified market outcomes.

The panel debates whether AI startups should seek acquisitions. Lemkin says a public-company CEO told him high startup prices made deals difficult, while O’Driscoll thinks buyers may find other capable products at lower prices. The anecdote does not establish a specific offer or prove that any company will sell.

Lemkin discusses a reported small acquisition by HubSpot and worries that fast-changing products can become stale during integration. O’Driscoll reframes the issue as preserving urgency and speed after a purchase. Their assessment of the company and transaction is not independently confirmed, and no generalized failure rate follows from the example.

Stebbings says Cerebras has filed again for an IPO and cites 2025 revenue of $510 million versus $290 million in 2024. O’Driscoll praises the company’s progress but corrects an earlier suggestion of profitability, saying an accounting reversal should be distinguished from operating performance. The panel’s figures and interpretation of filings are not independently verified here.

O’Driscoll describes Cerebras’s wafer-scale inference chips and says one of his portfolio companies has used its cloud service. He corrects himself about counterparties when mentioning reported OpenAI and AWS arrangements and says the value of a stated commitment is uncertain. His favorable product assessment and the company’s IPO prospects remain his views.

The guests compare a reported Groq transaction with the risk and potential upside of remaining independent. O’Driscoll says a potential Cerebras valuation could depend heavily on market narrative and risk appetite. Neither a sale price, IPO price nor future return is guaranteed by the comparison.

Lemkin and O’Driscoll discuss the founder’s choice between selling and continuing to build. O’Driscoll notes that the answer can look different in hindsight depending on outcomes; Lemkin highlights the pressure on employees when stock values change. These reflections are not an assessment of any individual founder’s actual wishes or finances.

The guests review an interview with Nvidia CEO Jensen Huang, including disagreements about China and access to advanced chips. O’Driscoll suggests the participants approached the question with different assumptions about national competition and the risks of frontier models. His characterization of their beliefs and commercial incentives is interpretation, not an independent finding about security risk or policy.

Stebbings cites reported Snap layoffs and a share-price move; the panel offers brief opinions about the business and dual-class voting. The conversation does not establish the cause of the cuts or show that the stock move was caused by them. Their comparisons with other social-media companies are time-specific observations.

Lemkin cites a claim that 91% of AI unicorns are in the Bay Area and asks Stebbings about London. The episode does not provide the source, definition or denominator for the figure, so it is not established as a reliable measure. Stebbings believes the Bay attracts leading talent but sees opportunities and less investor competition in Europe.

O’Driscoll suggests local access to knowledge may have helped the Bay Area during an early period, while talent and costs could make other locations competitive over time. Stebbings describes European investing as less crowded from his perspective. These are their explanations and predictions, not proof of a fixed geographic equilibrium.

The panel closes with a light exchange about the trade-offs of concentrating talent and capital. The alleged acquisition, funding, leadership and IPO developments discussed throughout remain subject to verification and future events.