David George - Building a16z Growth, Investing Across the AI Stack, and Why Markets Misprice Growth

The central thesis is straightforward to state but difficult to execute: pay fair prices for great companies by identifying greatness before others fully price it in. The discussion identifies three sources of genuine edge in growth investing—product insights, market insights, and people insights—and argues that spreadsheet-based analysis of margins and unit economics is widely available and therefore not a source of differentiation. The real advantage comes from being closer to founders, products, and markets than competitors. A key structural belief is that most technology markets are winner-take-all, described through the Glengarry Glen Ross analogy: first prize is a Cadillac, second prize is steak knives, third prize is termination. The discussion points to Salesforce, Workday, and ServiceNow as enterprise examples where there is effectively no viable second-place competitor. This conviction shapes portfolio construction: the fund prioritizes market leaders and is skeptical of investing in second-place players even when the absolute business appears sound. An important exception is noted for very large markets. Cloud infrastructure is cited as a case where AWS, Azure, and GCP all coexist profitably—not because winner-take-all dynamics are absent, but because the market is large enough to support multiple scaled winners. The same logic is applied to AI foundation models: the expectation is that the model layer will resemble cloud rather than a single-winner outcome, though the consumer AI interface market (dominated currently by ChatGPT) may still consolidate to one dominant player.

The discussion introduces a specific founder archetype the fund actively seeks: the technical terminator. This is defined as a founder who begins with deep technical and product grounding and then develops commercial and operational excellence over time. Examples cited include Ali Ghodsi of Databricks (who was not initially CEO), Mark Zuckerberg, Elon Musk, George Kurtz of CrowdStrike, Dave Baszucki of Roblox, and Dylan Field of Figma. The logic is that technical founders are more likely to identify the next product opportunity and navigate complex, shifting markets because they remain embedded in the product. The counterexample offered is Travis Kalanick at Uber—a non-technical founder who was precisely right for a market defined by regulatory combat and operational intensity rather than product innovation cycles.

On the consumer side, the discussion is bullish but humble. ChatGPT is described as having reached Google-scale roughly four times faster than Google did, with approximately one billion users but fewer than 50 million paying subscribers. The monetization model is explicitly uncertain—the discussion rejects the assumption that advertising will be the answer, suggesting instead something more like a native affiliate or action-based model. The analogy to Facebook's feed-based advertising is instructive: no one predicted that format in advance, yet it became arguably the most effective ad unit in history. The more important consumer shift anticipated is from reactive AI (the current chatbot model) to proactive AI with persistent memory, multimodal capability, and the ability to execute tasks on behalf of users. Deep Research is cited as an early signal of this transition. On the enterprise side, the discussion expresses more skepticism about near-term business model clarity. The steam engine analogy is used: competitive forces drove the price of steam engines to a level reflecting appropriate return on capital, not the full value of the labor displaced. The expectation is that 90% of AI productivity gains will accrue to end users, not to AI vendors—and that this is not a reason for pessimism, since even capturing 10% of a very large market can produce the largest companies ever built. Customer support and coding are identified as the two enterprise verticals where AI business models have progressed furthest. Customer support works because task completion is discrete and measurable, enabling outcome-based pricing. Coding works because the developer ecosystem is already accustomed to consumption-based pricing.

The discussion draws a sharp distinction between pull businesses—where market demand organically drives adoption—and push businesses, which require active sales and marketing to grow. Pull is described as the most important qualitative signal in evaluating AI companies. Cursor (viral developer adoption), Abridge (hospital systems actively seeking the product because physicians demand it), and Anduril (defense procurement urgency driven by geopolitical need) are cited as pull businesses. Push businesses are not disqualifying, but the discussion notes they tend to get harder to scale over time, particularly in consumer markets, as customer acquisition costs compound. TikTok is offered as an unusual exception—a push business that used paid advertising on Facebook aggressively enough to establish organic pull. For AI companies specifically, three evaluation criteria are outlined: ease of customer acquisition, durability of customer behavior (retention and engagement over time), and gross margins—with an explicit pass given on margins in the current period, on the expectation that inference costs will decline.

The growth fund operates without a traditional investment committee. Decisions are made by a single designated investor (the "trigger puller"), with open disagreement encouraged but commitment expected once a decision is made. The rationale is that committee structures create incentives to politick for votes rather than rigorously assess investment merits. Winning competitive deals is described as a multi-year relationship process rather than a transactional event. The fund invests in helping founders before any capital is deployed—with recruiting, customer introductions, and board placements—to earn the right to a direct conversation when the founder is ready to raise. The Figma investment is cited as an example where years of relationship-building, combined with a willingness to override a narrow market-size analysis in favor of a broader product thesis, produced a successful outcome. The fund's current portfolio is described as dollar-weighted growing at 112% annually, entered at an average of 21 times revenue—a combination the discussion argues is less risky than a 12% grower acquired at 15 times EBITDA in private equity, because sustained high growth de-risks valuation math dramatically.

- **Markets misprice persistent growth**: Consensus financial models systematically underestimate companies that sustain high growth rates; Apple's 2013 actual results were 3x above 2009 consensus estimates, illustrating how large the valuation gap can be. - **Winner-take-all is the default in tech, not the exception**: The fund treats market leadership as a near-prerequisite for investment, with the primary exception being markets large enough (cloud-scale) to support multiple profitable players. - **AI monetization models remain genuinely open**: Neither advertising nor seat-based SaaS is assumed to be the answer; the most intellectually honest position is that the dominant consumer AI monetization format has not yet been invented. - **Pull is the most important early signal for AI companies**: Organic demand, durable engagement, and ease of customer acquisition matter more than gross margins in the current period, with the expectation that inference cost declines will improve margins over time. - **Growth-stage edge comes from people, product, and market insight—not financial modeling**: The fund explicitly deprioritizes margin and unit economics analysis as a source of differentiation, focusing instead on founder judgment, product trajectory, and market structure.