Kaz Nejatian on Opendoor, AI Financing and the Cost of Growth

In this September 2025 roundtable, Harry Stebbings, Jason Lemkin and Rory O’Driscoll first speak with new Opendoor CEO Kaz Nejatian, then discuss AI infrastructure, startup growth, incumbents and IPOs. Company figures, deal reports and forecasts are presented as the participants’ contemporary statements, not independently verified results.

Nejatian says he did not expect to leave Shopify, where he had served as an operating and product leader. He joined Opendoor because he sees making buying, selling and owning homes less difficult as an important problem. He believes the company can build a profitable business while pursuing that mission.

Stebbings challenges whether promising to solve a large problem while working out profitability is a boom-time mindset. Nejatian rejects that characterization: he says the company has revenue plans and is a for-profit business, but argues that profits should support a broader mission. His confidence is a forward-looking management view, not a guarantee.

Asked about Opendoor’s reputation as a meme stock, Nejatian says he is not a trader and expects to hold Shopify and Opendoor shares. He believes the public price reflects the company’s potential and says he bought Opendoor shares at a higher price. His position and compensation give him a direct financial interest in this assessment.

Nejatian argues that housing is a large market with room for longer relationships and additional services. He cites earlier market-share performance in some local markets and believes Opendoor can regain ground. These are management claims and ambitions from his first days on the job, not verified future results.

O’Driscoll distinguishes valuation from the activism around the company, noting that outside shareholders pushed for changes. Nejatian says a large opportunity still requires operational excellence and committed stewardship. He emphasizes his belief in execution rather than presenting a completed turnaround.

O’Driscoll asks whether a business holding real homes is really a software company. Nejatian argues that software can provide the leverage and says both asset-heavy and asset-light models may have roles. He expects to build products and attach services rather than rely solely on one approach.

O’Driscoll breaks the operation into attracting buyers and sellers, pricing houses and efficiently managing transactions and repairs. Nejatian, then roughly a day into the role, says he is impressed with parts of the operation but sees room for better software and AI. His assessment is preliminary.

Nejatian says the long-term model should offer a fair price and earn trust, then provide useful services around the transaction. He imagines eventually offering a return option or long-term support for a purchased home. These are proposed future possibilities, not products or guarantees established by the interview.

O’Driscoll says the model is difficult because houses have individual characteristics that can materially affect price and margin. He compares it with Carvana’s car business but questions whether residential property is as easy to price. Nejatian responds that AI and software can improve the analysis; their disagreement remains unresolved.

Nejatian draws an analogy with Shopify: an accessible core product can support a longer commercial relationship and earn revenue through additional services. He wants Opendoor to make the first transaction efficient and offer further products over time. The discussion does not establish the economics of those future services.

Asked about real-estate agents, Nejatian says too many intermediaries can add friction, but acknowledges that some buyers and sellers want expert assistance. His stated goal is an excellent product for both sides, whether or not another person helps in a particular transaction.

Nejatian says the company did not yet have an asset-light model and that he would not restrict it to holding assets. He stresses that he has only just started and does not have a finished plan. The team will try products, learn from failures and look beyond a single home-purchase margin.

Stebbings asks about compensation. Nejatian says his salary is nominal and describes performance compensation tied to stock price and vesting conditions, with a complex mix of instruments. Headline potential compensation is contingent; it is not money already earned, and the interview is not a substitute for the company’s detailed disclosures.

Nejatian says the return of Keith Rabois and Eric Wu was a condition of his taking the role. O’Driscoll views board support as important for a difficult public-company turnaround. Nejatian says he expects board members to work closely with management, including reviewing homes and company expenses.

Stebbings asks about the substantial Shopify compensation Nejatian left behind. Nejatian says he values making an impact but is also bullish on his new investment and expects the effort to pay off. His stock ownership and incentive package matter when interpreting that optimism.

Nejatian describes evaluating his weeks by whether the work is hard, valuable and fun. He says Opendoor must deliver value to customers and shareholders and invites them to hold the company accountable. The plans he describes are aspirations at the start of his tenure.

After Nejatian leaves, Stebbings turns to Oracle and OpenAI. O’Driscoll describes Oracle reporting more than $300 billion in remaining performance obligations and infers that a large OpenAI cloud commitment accounts for much of it. A backlog of future obligations is not the same as revenue already earned or cash collected.

To illustrate the valuation argument, O’Driscoll hypothetically spreads $300 billion over five years and applies a revenue multiple. He then points out the uncertainty: OpenAI would need to fund its purchases, and the actual timing, collection, costs and profitability are not established by that arithmetic.

Lemkin asks whether investors are paying enough attention to the profitability of GPU hosting rather than focusing on top-line growth. O’Driscoll agrees that upfront capital expenditure and depreciation assumptions matter. Their concerns about Oracle’s eventual margins are opinions, not independently established results for the contract.

The panel compares Oracle’s established software business with an infrastructure expansion requiring large investment. O’Driscoll sees the stock-market reaction as enthusiasm about possible AI growth, while Lemkin questions whether the prospective revenue will contribute enough to the bottom line.

Stebbings worries that investors in private AI companies increasingly emphasize revenue growth without adequately examining gross margins. O’Driscoll doubts the full future amount attributed to OpenAI will necessarily be collected on the imagined schedule. This is his prediction, not evidence of a missed obligation.

The guests revisit earlier Stargate announcements and ask how different headline commitments relate to one another. O’Driscoll says announcements can create momentum for both the provider and customer, while urging investors to scrutinize assumptions about revenue and funding rather than add headlines as completed cash flows.

Lemkin predicts that some investors will regret rejecting attractive acquisition offers during a period of excitement. O’Driscoll recalls earlier technology cycles in which companies turned down offers and later lost value. Both acknowledge that some founders may rationally continue if they have strong conviction and an unusually large opportunity.

Stebbings distinguishes a rising private valuation from liquidity: a secondary sale can involve a discount or only a fraction of a position. O’Driscoll adds that private investors may have fewer opportunities to reduce risk when conditions change. A paper mark is not the same as distributed cash.

Lemkin says he has encouraged founders with strong offers to consider selling, while leaving room for founders to explain why they want to keep building. O’Driscoll says advice should reflect the market environment and each founder’s circumstances. Neither presents a sale as universally preferable.

The panel discusses changes in Microsoft and OpenAI’s relationship. Lemkin refers to reports of Microsoft using Anthropic tools and anticipates changes to OpenAI’s restrictions. O’Driscoll says the arrangement discussed is interim rather than a final agreement; the guests do not claim to know the final terms.

O’Driscoll speculates that Microsoft could eventually hold a significant equity stake while buying model access and selling cloud capacity on a less exclusive basis. The ownership range and potential return he mentions are scenarios, not settled transaction terms or realized proceeds.

He asks whether Microsoft can build a durable AI position of its own as its partner grows more independent. He offers a similar, less certain interpretation of Amazon and Anthropic, while acknowledging he is not close to either agreement. The future balance of these partnerships remains unknown.

The guests discuss the leverage a founder can gain after a startup reaches sufficient scale and can seek multiple sources of capital or infrastructure. O’Driscoll’s assertion that model companies have become independent is his assessment of bargaining power, not proof that commercial dependencies have disappeared.

Stebbings moves to application-layer companies Replit and Higgsfield. Lemkin says he is a small Higgsfield investor and a user. The panel cites a reported $50 million financing and $50 million annual recurring revenue for Higgsfield, alongside other rapid-growth claims; those figures are not independently verified here.

Lemkin sees AI video and presentation tools as evidence that markets can grow when creation becomes accessible to more people. He also mentions Gamma and discloses an investment in OpusClip. His enthusiasm should be read alongside those interests and does not establish the companies’ future growth.

O’Driscoll agrees that AI can expand access to coding and creative work, but notes risks for applications built on someone else’s underlying model. Customer retention, differentiation and model-provider dependence still matter. A rapid increase in users does not establish durable economics.

Stebbings compares the difficulty of predicting durable AI demand with the uncertainty around pandemic-era growth. Lemkin says some applications show strong revenue retention, while also acknowledging that products can still fail. The retention figures mentioned in discussion may use different definitions and are not independently audited.

The panel debates margins: Stebbings worries about the cost of serving usage, and Lemkin says economics vary significantly across products. He states that Higgsfield is cash-flow positive, a claim he makes as an investor. Cash flow and gross margin are distinct measures; the episode does not independently verify either company’s accounts.

Stebbings discloses an investment in Airwallex and contrasts a more established financial-services market with experimental creative tools. O’Driscoll says an investor must tolerate losses and identify which fast-growing companies can sustain demand. Different business models call for different risk assumptions.

O’Driscoll suggests durable winners may need both room to expand into adjacent products and founders willing to keep building. He gives the example of adding services around applications people create. This is a proposed framework rather than a test that guarantees any particular company’s success.

Lemkin argues that competition moves much faster than before and offers a hypothetical in which Anthropic loses substantial coding revenue if customers switch models. He explicitly says he is not predicting that outcome. Stebbings adds that some founders report low switching costs, though customers’ price sensitivity varies.

Stebbings asks whether incumbents could use existing distribution to compete with new creative and coding tools. The guests discuss Wix’s acquisition of Base44 and speculate about its future annual recurring revenue. The projected figure is not a reported completed result.

Lemkin argues that an incumbent can combine a new product with existing distribution, identity and safety capabilities. O’Driscoll notes that a given amount of new revenue may move the needle at a smaller company more than at a very large one. Neither claims every incumbent has successfully adapted to AI.

Stebbings raises the reported acquisition of Sana Labs by Workday for $1.1 billion, and the panel discusses how a strong second-place company may attract a strategic buyer. Lemkin reflects on his own founder experience. The reported deal value and comparison with competitors are presented as discussion, not independently verified outcomes.

The guests continue weighing when to sell and when to build independently. Strategic demand can create a valuable opportunity, but timing, competitive position and the buyer’s own alternatives affect the price. The discussion remains a set of investor judgments rather than a rule for founders.

The panel turns to a busy IPO week, citing offerings by Figure Technology, Gemini and Via. Stebbings and Lemkin welcome the prospect of greater liquidity for investors and employees. Reported deal sizes and first-day stock moves describe that week, not the companies’ subsequent performance.

O’Driscoll describes Figure as a financial-services business using blockchain in loan processing and settlement. He says credit quality still matters: technology alone cannot eliminate the risk of making bad loans. The panel contrasts Figure with Gemini’s crypto exchange and Via’s transportation software and services.

Lemkin asks whether a long journey to an IPO and a roughly tenfold aggregate comparison of Via’s valuation with capital raised remains attractive in an AI-driven market. O’Driscoll says he would welcome such an outcome. That aggregate ratio is not the realized return of any specific fund or investor.

Stebbings says he hopes successful public offerings will eventually return cash to limited partners who can fund new venture firms, including his own. He acknowledges that actual distributions take time. His fundraising interest is relevant to his enthusiasm for an open IPO market.

The guests discuss Bending Spoons’ reported acquisition of Vimeo for approximately $1.38 billion. O’Driscoll describes a strategy of buying established brands and improving operations; Stebbings questions whether a higher entry price changes the economics. Lemkin suggests profitability or renewed growth could matter, without claiming either outcome is assured.

In a quick-fire round, Lemkin gives a speculative year-end Opendoor share-price target, while O’Driscoll thinks the near-term stock could rise but emphasizes how hard the housing business is. These are informal predictions from September 2025, not investment recommendations or evidence of subsequent prices.

The guests debate Adobe’s share price and competitive position. O’Driscoll sees a profitable, slower-growing business that remains vulnerable to AI disruption; Lemkin expects further weakness and questions the meaning of ‘AI-influenced’ recurring revenue. Their revenue, multiple and personnel interpretations are opinions and estimates.

O’Driscoll asks how a large incumbent can make an AI acquisition large enough to affect its overall results. Lemkin argues that seat-based pricing creates a difficult incentive when new tools can reduce the number of seats customers need. The panel closes with this unresolved challenge for established software companies.