David George & Jack Altman on AI, Autonomy, and the Next $25 Trillion

David George argues that AI’s economic impact is unlikely to be winner-take-all: frontier and open models, AI labs and application companies, and Nvidia and newer chipmakers can all succeed. AI infrastructure spending has passed railroads as a share of GDP, reaching roughly 3%, and George expects it could grow tenfold over many years, reaching many trillions of dollars and contributing to GDP growth. Data-center capacity is hard to secure until 2028 because of supply-chain bottlenecks.

George estimates OpenAI, Anthropic, xAI, and Cursor together have about $120 billion in revenue and are adding more revenue per month than every hyperscaler except Amazon. But most AI revenue still flows to sellers of tokens, leaving a central test: whether enterprise customers achieve enough productivity gains to justify their spending. Roughly 30 million coders account for most current AI revenue, against about 1.5 billion knowledge workers worldwide, and AI has reached less than 5% of the broader B2B economy. Coding and writing are working use cases, while legal adoption is about a year behind coding.

The speakers expect demand to grow as AI spreads beyond coders, though enterprise adoption takes time because of sales, integration, implementation, and company-specific rules. George considers it very unlikely that coding will prove anomalous and AI will have little effect on other white-collar work, but identifies that possibility as a risk. Heavy users remain willing to pay about ten times more for frontier-model tokens, especially for coding; as usage expands, cheaper and open-source models should also gain demand. Application companies can select models based on the value and cost of each task. In legal technology, some clients now require law firms to use Harvey, and Kirkland’s announced plan to spend $500 million on its own technology stack was described as a major validation. The speakers argue that industry-specific products can become billion-dollar businesses because implementation, product details, and sales matter, and labs are unlikely to build every vertical offering.

Consumer AI already has more than a billion users, but most use it much like a search engine. The larger opportunity, the speakers say, is for proactive, multimodal assistants that interact naturally and take action; they expect models could improve 10 to 100 times over the next five to ten years. Consumer business models may combine subscriptions with advertising or a new ad-like format, although the eventual formats remain uncertain. They frame the previous technology wave—mobile, social, e-commerce, SaaS, and cloud—as creating about $25 trillion of “market gap,” and see AI, autonomy, robotics, health applications, and defense modernization together representing more than $25 trillion in opportunity.

Autonomous driving could expand ride-hailing by improving the service and lowering its cost. Taxi rides made up less than 0.1% of U.S. miles traveled, while San Francisco’s taxi market reportedly grew from about $100 million to $1 billion after Uber and Lyft changed the product and business model. One speaker says Waymo has driven millions of miles and is 10 to 14 times safer than a human driver, with autonomy potentially reaching 20 times safer. Fully loaded car ownership costs about $0.80 per mile, compared with more than $2 per mile for Uber or Lyft. If autonomous rides become cheaper than current ride-hailing, the speakers expect demand to grow more than tenfold. They estimate U.S. buyers would pay at least $10,000 per car for full autonomy, upfront or by subscription, against annual sales of roughly 17 million new cars. With Waymo operating fewer than 10,000 vehicles nationwide, production capacity may slow adoption even as a small number of major players serve a very large market.

Robotics already creates value in factories for repetitive, safe tasks. Mind Robotics is developing factory-floor robots for Rivian assembly, where deployment could generate revenue while providing feedback to improve models. The speakers think industrial and consumer robotics could ultimately exceed language AI’s market, with a “ChatGPT moment” for robotics possibly arriving within five years, though full realization may take more than ten. They expect productivity and quality-of-life gains, while warning that rapid automation could disrupt jobs and require workers to retrain or transition.

George rates the current product cycle 9–10 out of 10, driven by AI, autonomy, robotics, American dynamism, and biohealth, compared with 1 for 2021 and 8 for 2010; he rates the capital cycle about 6 and says product cycles matter more to venture and growth investing over a decade. As companies stay private longer, Benchmark has raised its first growth fund. One analysis cited by a speaker split private-market returns evenly between seed through Series B and Series C onward; he expects later-stage returns could rise to about 70%. George says his firm will pay high prices for exceptional companies and seeks outperformance mainly through revenue growth. He cites Databricks’ expansion into new products and SpaceX’s Starlink, whose economics once looked poor, as examples of outcomes exceeding early expectations; he also expects orbital computing to work. Company narratives, as with Palantir and CEO Alex Karp, increasingly shape fundraising, hiring, retention, and sometimes customer acquisition.