David George on Venture Fund Scale, AI Economics and Backing Founders

Harry Stebbings interviews David George, the Andreessen Horowitz partner who leads its growth investing team. They debate fund size, private markets, AI company economics and founder selection. George discusses companies his firm has backed, so his assessments and forecasts come with financial interests.

Stebbings asks whether a very large venture fund can plausibly return five times its investors' money. George says his firm's larger funds have performed strongly and describes a $1 billion fund as its best performer. These are his representations about fund performance; the interview does not provide audited fund-level returns or a complete peer comparison.

George says Databricks and Coinbase generated large contributions in that fund, citing a five-times-fund distributed return for Coinbase. His description does not establish that every contribution is realized or explain the measurement dates and fund-level net return, so the figures should not be combined into an audited overall result.

He cites an internal analysis of 50 leading IPOs from 2017 through 2025, saying approximately 47% of value gains occurred through Series B and 53% from Series C onward. This is a selected sample and its valuation and selection methods are not explained; it is not a general return split for all venture investments.

His argument is that longer private-company lifetimes and large technology markets create room to invest more capital at later stages. He expects another wave of major companies to create substantial private-market value, but that future scale and the fund returns it might produce are predictions.

Stebbings asks whether companies that stay private longer can be overtaken before investors obtain liquidity. George says public status does not itself determine competition and points to Flock Safety, in which his firm led multiple rounds. His optimism about the company is an interested investor's view.

Andreessen Horowitz could sell some private positions but historically has generally held and continued investing, George says. He also cautions against treating all extended private-company timelines as one story: individual companies have different reasons for waiting to list.

George says the public-company CEOs he knows have not told him they regret listing, while Stebbings says he has heard the opposite from others. George sees access to capital as one possible benefit; the two accounts are personal samples, not a survey of public CEOs.

They disagree about the relative cost of capital. George believes some companies in his portfolio could raise more cheaply in public markets, while Stebbings points to private AI valuations that appear high relative to listed peers. Neither establishes a general rule, and George says he is not close enough to the specific examples to judge them.

Remaining private can reduce visible share-price volatility and make employee communication easier, George concedes. He cites Stripe, SpaceX and Databricks as examples in which he believes that approach has worked; his firm has invested in these companies.

George argues that investors must reconsider traditional asset-class labels as valuable companies remain private for longer. He cites an analysis of declining return on invested capital in a public small- and mid-cap index, but the interview does not provide the methodology or show that the decline applies to every company.

Asked how an endowment should allocate across assets, he explicitly acknowledges his bias toward venture. His case depends on an assumption that future technology leaders resemble the last generation; he also acknowledges that some institutions may already be heavily allocated to private assets. The conversation is not an individualized allocation recommendation.

George rejects the idea that investors should ignore Europe, pointing to ElevenLabs. Stebbings says he missed its seed round and Deel's. Both men praise Deel's founder for pursuing customers relentlessly; George's firm has backed the companies they discuss.

On early companies priced like mature businesses, George agrees that many prices do not make sense. He says a very small number of exceptional founders can justify unusual risk, citing his firm's investment in Character.AI. His assessment of the downside and upside is an investor's judgment, not a guarantee.

He describes the growth fund as a way to revisit companies the early-stage team missed. George says roughly half its investment dollars follow early-stage portfolio companies, another 15% follow investments originated in growth, and about a third go to new companies. These are approximate internal allocations he reports, not externally verified figures.

One example is Deel: George says the firm did not participate in one intervening round before later co-leading a Series C. Stebbings reflects that fear of what a model provider or large incumbent might eventually build can cause an investor to miss a strong founder.

George calls his firm's principle investing in a founder's strongest strengths rather than demanding an absence of weaknesses. He says theoretical future competition and underestimating market size can become reasons to pass on good opportunities. That framework does not eliminate the possibility that competition or a small market proves decisive.

He ranks potential disruption to software companies as changes in business model, user interface and workflow, then access to data. He cites outcome-based customer service as one example and notes that colleagues at his firm invested in Decagon, which benefits from the trends he describes.

Stebbings argues that ambitious AI valuations require spending to shift from labor budgets into technology. George agrees the shift may occur but says customers must experience enough value to pull it through, rather than merely being instructed by management to adopt AI.

George cites C.H. Robinson's reported improvement in shipments per person and operating margin as an example of a business changing its operations. He attributes the gains to AI, but the interview does not disentangle other drivers or supply the underlying methodology, so it cannot establish AI alone caused the entire change.

Rapidly growing AI revenue matters if it is accompanied by retention and engagement, George says. Because these businesses may lack years of renewal history, his team looks at shorter-term usage and customer behavior. He cites Gamma, ElevenLabs, Abridge and Harvey among companies with demand or efficient customer acquisition; his firm has investments in them.

He says not every company must grow from zero to $100 million at the same pace. The relevant comparison depends on the market and direct competitors, and he considers customer acquisition efficiency important. Fast growth can help build an advantage, but it does not by itself prove durable returns.

On investor 'kingmaking,' George says capital and a prominent investor brand can help a company already attracting talent, customers and resources. He doubts that writing a large check alone can manufacture a winner and argues that indiscriminate subsidies can conceal a weak competitive advantage.

He acknowledges that a firm's brand can help founders recruit, particularly early on, and describes Andreessen Horowitz as a scale player that offers portfolio resources. That is also a description of his own firm's commercial positioning, not a neutral assessment of every investor.

The crowded customer-service category attracts funding because he believes today's products can already provide value, not solely because of future model improvements. He highlights Decagon, an existing investment, and says the market could either be dominated by a leader or divided among several players; he does not claim to know which outcome will occur.

Stebbings challenges the math of paying high prices for AI companies: even strong growth could leave modest returns if public-market multiples fall. George says unusually good companies can grow faster or earn higher multiples than an abstract example assumes. Both are scenarios, not proof that any particular valuation is justified.

On gross margins, George says his firm is more tolerant of lower current margins than in earlier software investing. Model input prices have fallen while usage of reasoning tokens has risen, making the near-term picture uncertain. His expectation that costs will rationalize and margins rise is a forecast.

He also considers a future in which model providers resemble a concentrated cloud market. He says a company claiming software-like margins while marketing itself as AI-heavy deserves questions about actual usage. Neither a particular market structure nor future margins are established by the interview.

Stebbings asks how investing in prominent companies such as Stripe and OpenAI fits the idea of finding overlooked opportunities. George says the opportunity may be recognizing how much larger an already successful company can become. He cites investments in Anduril and OpenAI made before some later product developments.

George identifies Anthropic as a notable company his firm did not invest in and praises its progress. He expects several model providers to remain important, comparing the potential market to cloud infrastructure; that comparison is an opinion, not a forecast supported by disclosed market-share evidence.

He predicts OpenAI and Anthropic will retain different strengths but compete in enterprise services, with Google also participating. He sees OpenAI's consumer brand as powerful. His firm is an OpenAI investor, and these statements are forward-looking competitive assessments.

Asked when OpenAI's entry price becomes too high, George says the team must constantly reassess. He recalls underestimating what Databricks could become and argues that the best founders may expand beyond their initial markets. That does not provide a specific price, expected return or assurance that future rounds will be attractive.

At scale, investment conflicts are difficult, George says. His firm tries to avoid direct conflicts, especially where a partner sits on a board, but he admits it does not always get the judgment right and companies can change direction after an investment.

The largest disagreement he recalls with firm founders Marc Andreessen and Ben Horowitz concerned Waymo in early 2020. George thought the price high before a market launch; Andreessen and Horowitz saw a potentially enormous market and a leader. The firm invested a smaller amount initially and later invested more.

George is enthusiastic about Waymo and cites a published opinion that its safety performance compares favorably with human drivers. The interview does not provide the underlying study, exposure comparisons or uncertainty needed to assess a claimed safety multiple; his firm is an investor.

Stebbings asks why the firm backed Adam Neumann's Flow. George says Neumann has unusual product and hiring strengths and sees an opportunity to improve renters' experience. He acknowledges risks while presenting his firm's investment thesis, not independent proof of the company's future performance.

George argues that rent is a major household expense and that a differentiated experience might create a business. His spending estimate and views about branding are illustrative claims in the interview; a large theoretical market alone does not establish customer demand or investor returns.

In the quick-fire round, George says his team changed its earlier assumption that foundation models would subsume most application software. It now expects many application companies to build around models. That view aligns with the firm's existing AI application investments.

He uses radiology as an example of why a model's performance on one task may not replace an entire occupation: clinicians perform other work around the scan. His clinical performance and time-allocation figures are presented without study details and should not be treated as verified medical comparisons.

He recalls meeting Abridge's physician founder and valuing the combination of domain knowledge, product judgment and execution. Stebbings sees a similar combination in Harvey's founder. Both companies are investments discussed from an investor's perspective, not independently assessed outcomes.

George praises investors and colleagues he has worked with and explains that the firm has decentralized as it grew. Earlier partner meetings exposed the growth team to many sectors; today, he says, the team has to seek that information deliberately and coordinate with early-stage investors.

Looking ahead, George is excited about personal health tools and robotics, which he sees as early investment opportunities. His hopes for AI-assisted health management and widespread robotic assistants are speculative, not validated clinical benefits or a promise of investment returns.