This interview covers how a16z's David George picks AI investments. He believes markets systematically undervalue companies growing over 30% because it's hard to model sustained high growth. He prefers 'technical terminators'—founders who start as engineers and become business leaders, like Databricks' Ali Ghodsi. Key holdings: Databricks (a top a16z bet, a data platform), Figma (design software, won after two years of relationship-building), and Waymo (self-driving, a16z invested small in 2020 and added later; just 400 cars in SF already beat Lyft's market share).
David George, General Partner at a16z, shared insights on the construction and strategy of his growth investing business during a program. The core argument is that the market often underestimates the value of sustained growth, and most high-quality tech markets ultimately exhibit a "winner-takes-al
David George is a General Partner at a16z, leading the firm's growth investment practice. The core of this interview focuses on how he builds team culture, designs investment processes, and positions across the stack amid the AI wave. David George argues that the market systematically undervalues sustained high growth—companies growing at over 30% often have their long-term value severely underestimated by traditional valuation models, because no one can naturally construct a financial model that sustains such high growth for 5-10 years.
David George argues that investors are naturally unable to model "sustained high growth," leading to systematic undervaluation of such companies.
He points out that for companies with growth rates exceeding 30%, the market has yet to fully price in the value of their growth rates. "I've studied all these companies — I call them 'model breakers' — and it's extremely unnatural for any investor to build a model where high growth persists for 5 or 10 years." He cites examples: no one modeled Google or Visa to maintain 15-20% growth 20 years after their founding.
Data supports this view: the consensus estimate for Apple's 2013 revenue in 2009 missed the actual figure by a factor of 3x — and this was the most widely covered company globally. David notes that if a company's growth declines from 80% not linearly to 65%, 50%, 30%, but instead holds at 75%, 65%, the valuation difference can reach 3x.
He compares the current portfolio: the dollar-weighted average growth rate of a16z's growth fund holdings is 112%, with an entry valuation of 21x revenue. "If I could spend my entire career investing in companies growing at 112% and entering at 21x revenue, I wouldn't hesitate — this is less risky than buying a company growing at 12% at 15x EBITDA, because growth itself resolves a great deal of risk."
David George defines his most preferred founder type—the "Technical Terminator": someone with a technical background who later evolves into an exceptional business leader.
The characteristics of this archetype: starting in technology, uncertain whether they can become a business talent, but ultimately excelling at both. Typical examples include Ali Ghodsi of Databricks (initially not the CEO, but one of seven co-founders), Mark Zuckerberg, and Elon Musk. David emphasizes: "These individuals are more likely to identify the next product direction because they understand the technology and are deeply involved in the product."
He provides a counterexample: Travis Kalanick of Uber—non-technical background, but suited for a "pure combat" market (fighting with mayors, battling competitors), requiring "ruthless competition, drive, and operational intensity."
Other examples: George Kurtz of CrowdStrike, Dave Baszucki of Roblox (outwardly quiet, but "ruthlessly competitive"), Dylan Field of Figma ("one of the nicest people, but competes with brutal ruthlessness"), Shiv Rao of Abridge (a practicing cardiologist turned tech founder, who keeps a bed in the office to sleep).
David George firmly believes that most technology markets ultimately become winner-takes-all, with second place often being unviable.
He borrows a scene from the film Glengarry Glen Ross: first prize is a Cadillac, second prize is a set of steak knives, and third prize — you're fired. "We strongly believe that the vast majority of market capitalization creation will flow to market leaders. This is evident in consumer companies driven by network effects, but it is equally common in enterprise companies — there is no second place for Salesforce, none for Workday, and none for ServiceNow."
Exception: The model layer (foundation models) may resemble the cloud computing market — large enough to accommodate multiple winners. "Early on, we debated: is this aircraft manufacturing (high margins) or airlines (brutal competition)? It now looks more like cloud computing — AWS, Azure, and GCP are each excellent businesses." However, in verticals such as consumer chat interfaces, second place may not be viable.
David George believes that "pull-based" businesses are the rarest and most powerful business models — where the market actively demands more of your product.
He keeps a sticky note on his computer that reads: "Is the market asking for more of your product?" When this happens, especially in the consumer space, it often gives rise to the most exceptional companies. ChatGPT is a prime example: 1 billion users, entirely organic growth, all driven by brand — and without network effects, which he finds surprising.
In contrast, "push-based" businesses require active sales or marketing, and they often become harder to scale as they grow. The cybersecurity industry is a typical case. "TikTok might be an exception — it aggressively promoted itself via Facebook ads in its early days, but push-based businesses almost never get easier."
When evaluating AI companies, he focuses on three metrics:
1. Ease of customer acquisition — Cursor grew virally; Abridge requires sales, but hospital systems are "eager" for it
2. Customer behavior/retention/engagement — Avoid "experimental" fads; Harvey saw a "step change" in usage after breakthroughs in reasoning models
3. Gross margin — Currently given a "pass": if an AI company has a 75% gross margin, "that means no one is using your AI product." He expects inference costs to continue declining, with gross margins eventually settling around 50%, rather than the 80% typical of SaaS
David George describes how a16z wins competitive deals through "multi-year relationship building" and a unique decision-making process.
The competitive landscape has fundamentally changed: technology companies now account for 8 of the top 10 global market capitalizations, with the private market valued at $5 trillion (a 10x increase over the past decade), equivalent to nearly one-quarter of the S&P 500. In the public market, fewer than 5 companies in the software, consumer, and fintech sectors are growing at over 30%, while the average growth rate of a16z's portfolio is 112%.
The key to winning deals is not "crazy storytelling" but "multi-year relationship building." He cites Figma's Dylan Field as an example: from day one at a16z, the team applied "full-court pressure"—inviting him to summits, giving him "bear hugs" from Mark and Ben, and helping find board members. Two years later, when COVID hit, Dylan proactively reached out.
Decision-making process design: Single Trigger Puller system, rather than a traditional investment committee. "You don't need to lobby or politick for votes. We openly expect disagreement, but after disagreement, you must commit." The team is only 10 people, but from the entry level, everyone is expected to "contribute to collective investment judgment."
David George believes the optimal combination for AI startups to defeat incumbent giants is: a completely redesigned UI/UX + a new data source + a business model shift.
He uses Salesforce as an example: the current product is essentially a "complex form checker" with a painful user experience. The future of AI will be proactive—"you log into Salesforce, and it tells you: these five clients you should follow up with; I have already monitored their online behavior and drafted a call script."
Key advantages: The new data source is no longer Salesforce's powerful structured database, but unstructured data from all interactions. A business model shift (e.g., charging per task completion rather than per seat) makes it difficult for incumbents to respond.
He cites the history of the SaaS/cloud wave: market revenue grew 7 times, with incumbents and startups each capturing half of the new share. "The more drastic the shift, especially in the business model, the more it favors startups."
| Position | Guest Stance | Key Data |
|---|---|---|
| Databricks | Bullish (Invested) | One of the largest holdings; key role in the data layer |
| Figma | Bullish (Invested) | Invested after 2 years of relationship building; market underestimated (design + front-end engineering convergence) |
| Stripe | Bullish (Invested) | Cross-fund investment |
| SpaceX | Bullish (Invested) | One of the largest holdings |
| Anduril | Bullish (Invested) | "Market desperately needs its products"; AI + autonomy + geopolitical demand convergence |
| OpenAI | Bullish (Invested) | Cross-fund investment |
| xAI | Bullish (Invested) | Cross-fund investment |
| Waymo | Bullish (Invested) | Small investment in 2020 (a16z was the only VC); added position in late 2024; only 400 vehicles in San Francisco surpassed Lyft's market share |
| Cursor | Bullish (Invested) | Viral growth; "Product-Market Fit" (PMF); enterprise sales immediately enter POC |
| Harvey | Bullish (Invested) | Step-change increase in usage after reasoning model breakthrough |
| Abridge | Bullish (Invested) | Founder Shiv Rao (physician-turned); hospital systems "eager" for the product |
| Roblox | Bullish (Invested) | Dual network effects; founder Dave Baszucki is a "technology terminator" |
| Coinbase | Bullish (Invested) | Cross-fund investment |
| Flock Safety | Bullish (Invested) | Mentioned as an existing holding |
| Decagon | Bullish (Invested) | Customer support space, advantage from business model shift |
| Salesforce | Risk Warning (Potential Disruption Target) | Existing giant, AI may disrupt its UI/UX and data advantages |
| Zendesk | Risk Warning (Potential Disruption Target) | Existing giant, facing competition from per-task pricing |
| GitHub | Positive Case (Not a Current Holding) | "Unique product leads to unique distribution"; once sold $400K to Walmart with no one making a phone call |
| Uber | Positive Case (Not a Current Holding) | Travis Kalanick is a counterexample of a "non-technology terminator" |
| CrowdStrike | Positive Case (Not a Current Holding) | George Kurtz is a "technology terminator" |
| TikTok | Neutral Observation | Exception to push-based business; aggressively promoted via Facebook ads in early days |
1. "The market systematically undervalues sustained high growth" (David George) — For companies growing at over 30%, investors cannot naturally model sustained growth, leading to severe undervaluation. Apple's consensus estimate in 2009 was three times lower than its actual 2013 performance.
2. "The 'Tech Terminator' founder archetype" (David George) — Founders who come from a technical background and later grow into business leaders (Ali Ghodsi, Mark Zuckerberg, Elon Musk) are more likely to identify the next product direction than purely business-oriented founders. Counterexample: Travis Kalanick (non-technical, suited for "pure combat" markets).
3. "Third place, you're fired" — The winner-takes-all rule (David George) — In most tech markets, the vast majority of market cap creation flows to the market leader. Salesforce, Workday, and ServiceNow all lack viable second-place competitors. The model layer is an exception, potentially accommodating multiple winners like the cloud computing market.
4. "Is the market demanding more of your product?" — The criterion for pull-based businesses (David George) — This is the rarest and most powerful business model. ChatGPT's 1 billion users, entirely organic growth with no network effects, is a "surprising" case of a pull-based business. Push-based businesses (requiring active sales) often become harder to scale as they grow larger.
5. "If an AI company has a 75% gross margin, it means no one is using your AI product" (David George) — The current market gives AI companies a "pass" on gross margins, expecting inference costs to continue declining. Ultimately, gross margins may settle around 50% rather than SaaS's 80%, but the scale of impact will be sufficient to compensate.
6. "We win deals through years of relationship building, not crazy stories" (David George) — The Figma case: From day one of joining a16z, a "full-court press" lasted two years until the founder proactively reached out during the COVID outbreak. The decision-making process uses a single decision-maker system rather than a traditional investment committee.
7. "90% of the residual value from technology will flow to end users" (David George) — The steam engine was not priced as "replacing 50 workers"; competitive forces compressed it to a "fair return on capital." The same applies to AI enterprises, but even so, it can still create the largest companies in history.
8. Three elements for AI startups to beat incumbents (David George) — Completely redesigned UI/UX + new data sources (unstructured data replacing structured databases) + business model shift (pricing by task completion rather than by seat). The more drastic the shift, the more favorable it is for startups.