Cursor, Figma and the Race to Build Across AI Workflows

Harry Stebbings, Jason Lemkin and Rory O’Driscoll discuss venture fundraising, AI infrastructure, the convergence of design and coding, and a possible SpaceX IPO. Their investment ideas, company figures and future scenarios reflect the recording date and are not independently verified or financial advice; the speakers disclose several relevant investments and business relationships.

Stebbings says Lightspeed raised $9 billion across six funds and estimates roughly $2 billion is for venture and early stages. Lemkin says the distinction matters: the entire amount is not a seed fund. Both figures are presented as their understanding, not an independently inspected fund breakdown.

Lemkin thinks a multi-stage firm can pay more at seed to gain access to later opportunities, putting pressure on smaller funds. O’Driscoll points to early investments and reported later investment in Anthropic as examples of the two things a multi-stage limited partner may seek: long-term company building and concentrated growth bets.

The group discusses whether very large funds can return capital without participating in a few unusually large outcomes. Stebbings cites reported fund returns from Databricks and Coinbase; O’Driscoll says the math can work if investors execute and reach such deals. Their examples are not audited fund performance.

Stebbings worries multi-stage firms erode seed economics. Lemkin disagrees that appetite for small emerging managers remains as strong as before, while O’Driscoll acknowledges large funds may make it harder to compete at the margin. None provides a systematic estimate of the effect.

O’Driscoll says a competitive late-stage market must be assessed against potential returns, not fundraising alone. He recalls that 2021 looked attractive until growth slowed for many companies, and says AI firms may absorb large amounts of capital. Whether this cycle will be less dangerous remains unknown.

Lemkin calls leading companies staying private a major opportunity for growth investors, because much compounding can occur before public investors can participate. O’Driscoll adds that private-market cost of capital may now be preferable for some issuers; neither claim ensures future returns or says all IPOs would have benefited retail buyers.

The speakers compare Tesla’s early listing valuation with hypothetical much larger SpaceX IPO prices. O’Driscoll frames the decision to list partly as a cost-of-capital choice. The comparison is illustrative across different businesses and periods, and the future IPO and price were not confirmed in this conversation.

Turning to Disney and OpenAI, O’Driscoll describes the reported $1 billion investment and licensing arrangement as an experiment and offers his understanding of the economics. Lemkin sees a possible template for rights owners licensing intellectual property to AI platforms. Their descriptions of terms and future prices are not a contract analysis.

They ask whether licensed characters will attract users and produce sufficient economic returns compared with user-generated content. Stebbings suggests intellectual property might help retention, while O’Driscoll says this is an open question. Lemkin’s comments about future employment and leisure are speculative.

The group discusses reported ChatGPT app downloads and competition with Gemini. O’Driscoll’s preference is explicitly a gut reaction; Lemkin says growth in users may slow and asks whether a broad consumer app can monetize existing users. The episode does not establish current download rankings or a forecast of market share.

O’Driscoll distinguishes user growth from monetization and warns that a high-growth company’s valuation can suffer if revenue growth falls. He says he is not sounding an alarm about OpenAI’s revenue; the downside is a hypothetical risk, not a finding that its business has stalled.

They explain a conventional four-year stock vesting schedule with a one-year cliff and discuss reported changes in hiring practices. The speakers speculate that large equity awards and competition for talent could make candidates resist a cliff. This is an explanation of incentives, not verified terms for every employee.

Stebbings cites Oracle’s share-price decline and higher data-center capital spending. O’Driscoll argues investors had previously celebrated contract commitments without fully pricing the cost of meeting them, and questions the economics of capital-intensive infrastructure serving a few large customers. These are his investment judgments.

Lemkin thinks Oracle and CoreWeave could rebound if AI infrastructure demand continues, despite potential margin pressure. O’Driscoll agrees that returns depend on the future capital-spending cycle. They discuss multiple possible outcomes; the conversation is not a recommendation to buy or sell either stock.

O’Driscoll suggests markets may differentiate between companies with clear AI investment plans and marginal providers, rather than signaling either a universal collapse or rebound. He emphasizes that an intermediate outcome is plausible, though no short-term path is established.

Stebbings raises Broadcom’s reported market-cap decline and an Anthropic chip order. O’Driscoll reasons that custom-designed chips may carry different margins from Nvidia-branded products, but the cited order value, margins and share-price drivers are not independently verified in the interview.

Lemkin observes that investors tolerate weaker gross margins for some AI businesses but not others and says he cannot reliably predict who will receive that benefit. The group distinguishes a fall in market capitalization from a realized operating loss.

O’Driscoll discusses a cited long-horizon public-equity return forecast and cautions that expensive starting valuations may imply lower expected long-term returns without predicting the next year. He recommends thinking in terms of overall allocation rather than a binary market-timing decision; this is commentary, not personal investment guidance.

He adds that private valuations compared with public peers at high multiples can look different if the public market reprices. His example of a possible decline followed by recovery is a scenario, not a prediction of a specific market move.

Stebbings initially attributes a design-oriented coding feature to Anthropic, and Lemkin corrects him: the subject is Cursor. Lemkin argues design, coding and other software categories may converge as tools can perform more tasks; this is his forecast, not evidence that Figma has lost its market.

Lemkin uses e-commerce marketing, sales and support as an example of potential convergence and imagines a shared agent across the customer journey. O’Driscoll agrees that AI products may expand across departmental boundaries, while acknowledging he does not yet know what winning applications will look like.

O’Driscoll says a single agent could preserve customer context across prospecting, sales and support, rather than forcing a customer to repeat information. He notes investments in support and AI sales businesses. His claim that the categories will converge over one to three years is a prediction.

Lemkin distinguishes the roles of designers from the platforms they use: he expects designers to remain but thinks shared tools could reduce handoffs. He discloses an investment in Alloy App, a startup working between existing products and AI-assisted development; reported demand is his account.

Asked who will win among Cursor, Claude Code and Figma, Lemkin says execution will matter and acknowledges that he cannot predict the outcome. He thinks Cursor has momentum but also notes it lacks Figma’s designer base and customers. His work with Cursor is a relevant interest.

They speculate about whether Cursor might move toward consumer-oriented app creation. O’Driscoll argues enterprise software budgets may be larger, while Lemkin cites retention and growth estimates without a disclosed methodology. Neither estimate proves a company’s financial performance or future strategy.

The panel cites a report suggesting coding represents a large share of enterprise AI application spending. O’Driscoll compares estimated end-user spending with infrastructure investment and asks whether customer budgets can grow enough to support capital expenditures. The cited categories, totals and timing are estimates and not directly interchangeable cash flows.

Lemkin says a competitor need not eliminate an incumbent to slow it: existing customers might renew but buy fewer seats, while new customers defer purchases. He uses Figma as a hypothetical and calls this a risk to growth and net retention, not a finding that Cursor has already caused such losses.

The speakers discuss established software companies and argue that budget priorities can move toward AI. Their examples and comparisons are interpretations, not causal proof that a specific company lost growth to AI. O’Driscoll notes that infrastructure suppliers may benefit if they serve AI workloads.

UiPath becomes an example of a company the panel believes must adapt robotic process automation to AI agents. O’Driscoll suggests incumbents can build or acquire capabilities but must change customer perception and execute organizationally. Their product assessments and reported company metrics are not independently checked.

Lemkin points to reported high gross retention as giving an incumbent time to sell new products to existing customers; he frames a return to faster growth as a possibility. O’Driscoll cautions that lifting growth at a large revenue base can take years and may not immediately be rewarded by markets.

The topic turns to Boom Supersonic’s reported data-center power initiative and financing. O’Driscoll explains that aircraft-engine and stationary turbine technologies can overlap, making the idea less surprising than it first sounds. He expresses admiration, not a technical validation or assurance that the project will succeed.

Lemkin describes reported changes in Boom’s private valuation and financing terms, which are not independently verified. The panel distinguishes ambitious hard-engineering risk from an established software business reaccelerating, and acknowledges that neither aircraft nor turbine success is guaranteed.

Stebbings introduces reports of a possible SpaceX IPO at $1.5 trillion after an alleged secondary valuation around $800 billion. O’Driscoll says reported revenue and ordinary multiples do not by themselves explain the higher number. Neither a transaction nor a future IPO price is established by the interview.

O’Driscoll calls the gap between conventional valuation and the reported price an “Elon option value”: investors may credit Musk’s ability to create additional businesses. This is his interpretive framework, not a measurable accounting item or proof that a premium is justified.

He considers potential Starlink and launch revenue but says he still struggles to derive the reported valuation. Lemkin speculates that space-based data centers could become a future opportunity. Neither a viable commercial plan nor exclusive technical capability is established in the discussion.

Asked whether a $1.5 trillion IPO would attract demand and trade well, O’Driscoll says he cannot quantify investor willingness to pay for intangibles. He distinguishes possible outcomes for existing sellers from returns for new buyers; his assessment is not investment advice.

The group illustrates the potential fundraising challenge: the proceeds required depend on how much equity would be sold. O’Driscoll asks which institutions or retail buyers might provide that capital and notes the valuation might change. All numbers are hypothetical.

Lemkin invents a scenario in which Google or other institutions anchor a SpaceX offering and create demand. The speakers treat it as a story about how investor narratives could develop, not a reported agreement, confirmed investment or announced IPO allocation.

Lemkin recounts a story about Peter Thiel, Musk and an early SpaceX investment as a lesson in treating people well. O’Driscoll questions how much of the outcome can be attributed to kindness. Their recollections and reported ownership and returns are not independently verified.

For a hypothetical choice at stated valuations, O’Driscoll selects Cursor and Lemkin chooses Figma, citing uncertainty about the durability of AI leaders. These are personal, time-specific preferences and not evidence that either company is a better investment.

Asked to compare OpenAI, Anthropic and Google at hypothetical prices, Lemkin chooses Google based partly on his experience working with its teams through SaaStr. O’Driscoll considers Anthropic but notes that price matters and that future financing terms are uncertain. Their hypothetical valuations and company forecasts are not recommendations.

The panel closes by reflecting on how hard it is to ask and answer precise questions about fast-moving markets. The episode’s valuations, predictions and potential IPO scenarios should be read in the context of its recording date.