GPT-5, Palantir and the Economics of AI Growth

In this August 2025 roundtable, Harry Stebbings, Jason Lemkin and Rory O’Driscoll discuss GPT-5, a proposed bid for Chrome, AI startup financing and the earnings of public software companies. They also debate lean teams and the potential effects of AI on workers. Their financial figures, market assessments and predictions reflect the conversation at the time and are not independently verified here.

Stebbings asks whether GPT-5 was underwhelming. Lemkin describes a disappointing initial experience but says other users reported improvements in document work and coding. He sees coding competition as commercially important even if the launch did not impress every consumer.

O’Driscoll welcomes a shift from grand claims about imminent AGI toward improving a useful product. He compares the process with successive generations of operating systems and phones. His view that AI progress may become more incremental is a hypothesis, not a settled forecast.

O’Driscoll says competing models and lower prices could help coding-tool companies that depend on outside suppliers. He estimates a substantial price difference from his own review, but actual cost and performance depend on the model, workload and pricing terms. The panel disagrees over whether Anthropic should feel threatened.

Stebbings says the launch presentation disappointed him and suggests Anthropic may benefit. Lemkin speculates about why OpenAI released when it did; O’Driscoll challenges the idea that the timing must reflect a fully mapped strategy. None claims direct knowledge of internal release decisions.

The guests debate whether a lower-cost competitor makes life harder for an incumbent model provider, even if the incumbent’s product remains stronger in some tasks. Stebbings expects prices and capabilities to change over time; O’Driscoll says customers could mix cheaper and more capable models to manage costs.

O’Driscoll also discusses OpenAI’s open-weight releases and a simpler model-selection experience. He interprets these as possible commercial moves, while admitting he does not know the company’s motives. His comments about rollout and user experience are early observations from 2025.

Lemkin offers a theory that OpenAI chose to ship rather than wait for a perfect release, emphasizing that he does not know the internal deadline. Stebbings mentions a possible financing context. These are their interpretations of timing, not confirmed explanations.

Asked about investing in OpenAI at a hypothetical $500 billion valuation, O’Driscoll says he is more confident in its commercial opportunity. He sketches a consumer subscription business and possible large outcomes. These valuations, adoption assumptions and returns are predictions, not established future facts.

Stebbings suggests chip suppliers could benefit regardless of which model company wins, citing Nvidia as his example. O’Driscoll says investors sometimes overcomplicate an obvious underlying opportunity. Both are expressing views on an uncertain market, not recommending a security.

The panel turns to Perplexity’s proposed $34.5 billion bid for Chrome. O’Driscoll says the possibility of a sale arose in the context of US antitrust proceedings, but neither a forced sale nor a completed Perplexity transaction is established in this episode. Funding, legal process and the outcome remained uncertain at the time.

O’Driscoll says a browser’s strategic value depends on the buyer’s ability to monetize its traffic and choose default services. He contrasts a search business with an AI subscription business. His payment and revenue examples are hypothetical, not an appraisal of Chrome or a statement of transaction terms.

The guests ask whether Perplexity could fund a bid and whether competing bidders might emerge. Lemkin also sees public attention as valuable in a crowded AI market, although he does not claim to know the bidder’s motives. A proposal’s strategic value and its likelihood of closing are separate questions.

O’Driscoll criticizes the timing of antitrust intervention and argues that technological competition can change while legal cases proceed. That is his policy opinion, not a legal conclusion. Stebbings adds that Google has significant distribution and may remain competitive with its own AI products.

Lemkin argues that the browser has regained strategic importance as an entry point to AI services. O’Driscoll suggests a hypothetical subscription conversion could change the economics of browser traffic, but cautions that applying another product’s conversion rate would be a leap. The usage and paid-user figures cited are estimates.

Stebbings introduces a reported n8n financing at a $3 billion valuation and cites reported revenue and growth projections. He says the workflow company was founded before the current AI boom. O’Driscoll recalls that his firm had spoken with it earlier and underestimated how AI might accelerate the category.

O’Driscoll says workflow tools may become more valuable if AI helps perform work rather than merely route deterministic steps. Lemkin describes using Zapier to connect applications and sees a larger potential market as more people build software. These are interpretations of the opportunity, not verified company financials.

O’Driscoll suggests investors revisit existing companies when a technology shift changes their potential. He emphasizes founder execution: recognizing a new capability, shipping it quickly and testing it with customers. Hindsight does not prove that any particular company’s current valuation is justified.

The guests discuss competition among prominent venture firms for deals. Stebbings says 20VC has an investment interest in Lovable, relevant to his comparisons. They debate whether paying a higher price to gain access can be rational if the eventual outcome is large enough, while recognizing that a fund’s size and ownership matter.

O’Driscoll distinguishes an investment producing a good multiple from one that meaningfully returns a large fund. Paying up for a strong asset might still earn money, but a small allocation may have limited effect on total fund performance. Future exit values and the multiples discussed are hypothetical.

The panel returns to price discipline and the psychology of following other investors. O’Driscoll says a high-quality company can still be a poor investment at an excessive price; a company’s operating success and the buyer’s entry point are different questions.

Stebbings asks why Datadog’s stock fell despite results the panel regards as strong. O’Driscoll says public-market reactions depend on what investors expected, including expectations that management cannot observe. He cautions against treating each day’s price movement as a definitive verdict on operations.

The guests discuss Datadog’s reported large OpenAI-related business and the possible concentration that follows. Lemkin says fast-growing AI customers can create meaningful demand, while costs and purchasing power may change over time. Contract figures and any possible renegotiation are discussed as estimates and speculation, not verified terms.

O’Driscoll says infrastructure providers can benefit from AI spending even if they do not build models. Chips, data centers, power and software all form part of the supply chain. His comments about the size of aggregate spending are broad estimates, not a complete accounting of economic growth.

Stebbings asks whether established technology companies are winning alongside the AI wave. O’Driscoll distinguishes direct exposure to AI infrastructure from companies using AI to strengthen an existing business. In his view, scale, customer relationships and execution can help leading software firms defend their positions.

The panel turns to Palantir. Lemkin highlights its reported acceleration from slower growth at substantial revenue to much faster growth and cites larger contracts. O’Driscoll admires the business but says a very high revenue multiple requires sustained performance for years. Their figures are contemporary estimates and not a valuation recommendation.

O’Driscoll uses a hypothetical five-year compounding exercise to show why a lofty multiple might eventually normalize if growth persists. He also says he has not checked the calculation fully and would find the entry price uncomfortable. The scenario is not a forecast that the company will achieve it.

The guests discuss Palantir’s reported US commercial bookings and the difference between bookings and revenue. O’Driscoll describes a combination of software platforms, AI features and forward-deployed teams that can help large customers implement projects. His account of margins and demand is an investor interpretation rather than audited unit economics.

Lemkin cites Alex Karp’s prediction that Palantir could grow substantially with fewer employees. O’Driscoll says he would not enter that assumption into a model simply because it was stated. The panel treats the future headcount claim as a prediction, not an achieved outcome.

Lemkin argues that technology companies can operate with much leaner teams and cites Shopify and Palantir. Stebbings asks whether leaders should want employees to fear losing their jobs. The guests disagree about the effects of uncertainty and how easily workers could find comparable roles.

Lemkin describes his experience introducing an AI sales tool and says a team member left soon afterward. His interpretation of the person’s motivation is anecdotal and unverified. He argues for greater accountability; O’Driscoll questions the social consequences of a workforce living in constant fear.

O’Driscoll initially suggests some workers may find comparable jobs, while Stebbings worries especially about early-career people without specialized skills. The panel does not have employment data sufficient to settle their disagreement. Their descriptions of workers’ prospects are predictions and personal views.

Lemkin cites Shopify headcount falling from roughly 11,600 to 8,100 while revenue rose 91% from a prior peak period. The comparison is his reported calculation; the discussion does not establish that AI alone caused the change or isolate acquisitions, divestitures or other factors.

Stebbings argues that losing a job can carry substantial downside beyond missing future equity gains. O’Driscoll revisits his assumption about replacement work, and Lemkin says experienced executives in his network are finding hiring harder. These are personal experiences and uncertain labor-market forecasts, not conclusions about every worker.

Stebbings asks why monday.com’s stock fell after what he views as a good quarter. Lemkin points to high expectations embedded in a strong growth valuation. O’Driscoll reiterates that market reactions can depend on forecasts and investor expectations that outsiders cannot fully see.

Asked to rank Datadog, HubSpot, Palantir and Shopify, Lemkin favors Shopify in his comparison. O’Driscoll distinguishes the size of a company’s opportunity from the return available at a given multiple and says he cannot responsibly rank the stocks on the spot. Their preferences are opinions, not investment advice.

Stebbings cites Carta’s second-quarter 2025 data on high seed valuations. Lemkin says the analysis also indicates fewer seed rounds and greater concentration. O’Driscoll compares enormous late-stage rounds with the smaller pool of early-stage enterprise deals his firm tracks; those figures reflect the speakers’ data and definitions.

O’Driscoll expects capital-intensive model and defense companies, and companies staying private longer, to continue raising large rounds even if the current extreme concentration moderates. Stebbings says larger outcomes have made him more open to large platform funds. These are expectations about future financing, not guaranteed fund returns.

The guests ask whether one person can build a billion-dollar company. O’Driscoll regards the idea as a metaphor for greater productivity and notes that accounting, customers and other functions still require work. Lemkin thinks smaller core teams, supported by AI and outside contractors, could build very valuable companies.

Lemkin says his own SaaStr business has two employees, contractors and outsourced accounting, and hopes to reach $20 million in revenue that calendar year. He explicitly distinguishes employee count from the other people who help deliver the work. The revenue figure is a target, not verified results.

Lemkin gives an early example of an AI business-development tool arranging meetings and says small teams may trade some growth opportunities for fewer management demands. O’Driscoll notes that contractors and event workers still perform substantial work and may not share in the same equity upside.

The guests return to labor risk. O’Driscoll revises his earlier view, acknowledging that some displaced employees might face a significant income gap, while saying widespread fear is undesirable. Lemkin describes time spent supervising AI tools; the future balance of jobs and wages remains uncertain.

In a quick-fire round, the panel considers whether Palantir could reach a $2 trillion market capitalization within five years. Both guests take the under while recognizing the company’s strong performance. This is a dated speculative prediction, not a statement about its later valuation.

Asked about Stripe’s IPO timing, O’Driscoll says he does not know and discusses how private liquidity and cash generation can affect the decision. The founders have not disclosed their plans to him. The panel then speculates about potential disputes between xAI and Apple without claiming a legal outcome.

The episode closes with Lemkin naming Alex Karp as a desired future guest. The discussion throughout reflects individual opinions and the information available in August 2025, including claims and projections that were not independently verified in the episode.