This conversation between investors Bill Gurley and Michael Mauboussin is about spotting companies that get better as more people use them (network effects), and why this AI boom may be less disruptive than expected. Gurley says AI is too visible—big players like Microsoft and Google are all in, making it harder for startups. Three key examples: Microsoft—quickly adding AI to its products gives it an edge; Uber—more drivers mean shorter wait times, which attracts more riders; CrowdStrike—its security software shares threat data, so the biggest network is the safest. No jargon.
Benchmark partner Bill Gurley and Counterpoint Global's Michael Mauboussin shared a stage to discuss the various types of increasing returns to scale, the hindrance of regulatory capture to innovation, the AI investment frenzy, and the structural challenges of the venture capital ecosystem. Bill Gurley believes that the investment opportunities in AI are "highly orchestrated," which precisely weakens the window for disruptive innovation by startups—when everyone knows the direction, first movers usually defend their advantage.
Bill Gurley believes that the biggest difference between this AI wave and historically disruptive innovations is "high orchestration." Historically, innovation windows that truly opened the door were almost never anticipated in advance, but AI – especially LLMs – has been followed globally and simultaneously, prompting giants like Microsoft, Amazon, Google, and Meta to go all-in from the start, with no case of "falling asleep during the mobile internet era."
Gurley stresses the need to distinguish LLMs from general AI: Tesla's shift to a pure AI path for autonomous driving is a real breakthrough, but it does not involve LLMs at all; meanwhile, the core strength of LLMs lies in text processing and code generation, not in being a panacea. "When people say, 'When computers are idle, they'll find a cure for cancer' – that came from a senior person at a foundation model company, but it's not going to happen any time soon."
Key assessment: Current LLMs have a structural flaw in "personal memory." Existing models are more like "an advanced version of Wikipedia," lacking continuous memory of a user's personal information. Gurley believes that whoever can solve the problem of "tuning the model for each individual" (e.g., Apple achieving local models by mastering phone context) will completely change the game – he cites the movie Her as an analogy.
From Michael Mauboussin: From a historical perspective, new technologies are usually first "attached" to existing processes to improve efficiency, and only later redefine workflows. This was true for the electric motor era, the internet era, and will likely be true for AI as well – "Everyone is talking about AI now, but very few companies have fully integrated it."
Michael Mauboussin lists five sources of increasing returns to scale: economies of scale, international trade, learning by doing, network effects, and recombination of ideas. The latter two are the core of this discussion.
Network Effects: Gurley's "Perfect Business Model" Framework
Gurley recalls that his 2003 article "In Search of the Perfect Business Model" was deeply influenced by Brian Arthur's "theory of increasing returns." "When more drivers join, pickup times drop; as pickup times drop, consumer value increases; more consumers use it, more drivers join – this is a win-win cycle for all participants in the system." He believes this structure is extremely rare, but once established, it provides an "unfair competitive advantage."
He offers a litmus test: "If you become the N+1,000th customer instead of the Nth customer, does your value become higher?" He cites CrowdStrike's security cloud network effect – shared threat intelligence means the largest network is the safest, and customers never want to join the second or third network.
Mauboussin adds: "Pricing power is not the key metric; the real key is: are you increasing the customer's willingness to pay every day?" If willingness to pay rises while the price remains constant, you are creating consumer surplus; if you need to raise prices in the future, you have room to do so.
Learning Curve: The Power of Wright's Law
Mauboussin introduces Wright's Law (proposed in the 1930s): For every cumulative doubling of production, unit costs decline by approximately 20%. The Santa Fe Institute tested 60 technologies, and Wright's Law was the most accurate model for predicting cost declines – solar cells and lithium-ion batteries fit almost perfectly. "Tesla's cost advantage per vehicle fundamentally stems from cumulative production far exceeding that of its competitors."
Recombination of Ideas: Open Source as the Best Example
Mauboussin cites Paul Romer (Nobel laureate) and his endogenous growth theory: Innovation is essentially the "recombination of existing knowledge modules." Digitization enables these modules to be quickly searched, manipulated, and recombined.
Gurley believes open source is "one of the most powerful innovations for human flourishing." "In Michael's framework, the marginal cost of open source is zero, and no one can prevent you from using it." He cites historical examples: Google used Android (open source) to counter Apple's closed ecosystem with the iPhone, and Kubernetes (open source) to counter AWS's cloud monopoly – both are cases of "using defensive open source to attack closed systems."
Bill Gurley argues that regulatory capture is the most underestimated structural problem in the United States. He cites Nobel laureate George Stigler: "As a country ages, more people learn how to influence Washington – regulation becomes the friend of the incumbent."
Supporting data: After the Dodd-Frank Act was passed, industry competition actually declined. "Regulation that serves incumbents has been layered on over the past 40 years, and we can no longer reverse course."
Vivid examples:
Comparison with China: "China can clear the table in one go, while we treat making an exception as a victory." Gurley believes that in areas like nuclear energy, which require long-term capital investment, this gap is a huge disadvantage.
Mauboussin adds: Nuclear energy is the most extreme example – "If you think from scratch, any visitor from Mars would say: 'Nuclear energy is clearly the energy source that humanity should develop vigorously.' But because we have accumulated fifty years of political, emotional, and regulatory baggage, we are actually moving backwards – Germany is a typical case."
Michael Mauboussin points out that the venture capital industry faces a structural dilemma of “low barriers to entry, high barriers to exit.” He cites data: the number of U.S. listed companies has decreased by 46%, and companies stay in private markets longer. This has led the SEC to attempt to “let ordinary investors participate in private markets,” but Gurley warns this could lead to the absurd situation of “using social security funds to buy an Andreessen fund.”
On capital misallocation during the ZIRP era
Mauboussin's research found: during the zero-interest rate period of 2009-2021, corporate actual behavior was completely opposite to textbook expectations. Expected: investment increased, cash decreased, debt increased; Actual: investment declined (except for intangible assets), cash balances rose, leverage fell. The only thing that increased was buybacks — because low interest rates had a clear boosting effect on EPS.
Gurley adds from an insider perspective: “When everyone understands the rules of the ‘increasing returns game,’ the game changes.” He cites the Uber/Lyft price war and the “arms race” among foundation model companies — “These companies burn $100-200 million a year, and there is no way the capital allocation math works. But they may not be able to stop — that’s the trap.”
“Missing the biggest” is a hundred times more important than “avoiding the smallest”
Gurley points out that the two types of errors in venture capital are asymmetric: “Only ‘missing the big one’ is fatal. Correctly judging which ones will fail is almost worthless.” He cites data from Hendrick Bessembinder (supplemented by Mauboussin): Since 1926, 60% of U.S. stocks have failed to beat Treasury yields, but 2% of companies have created 90% of the wealth.
Benchmark's response strategies:
| Company | Guest's Stance | Key Data |
|---|---|---|
| Microsoft | Bullish on its AI first-mover advantage | Rapidly integrating AI across full product line (Gurley notes a stark contrast with the mobile era) |
| Amazon | Bullish on its dual cloud/AI advantages | AWS dominant position, participating in open source mapping project |
| Neutral to positive | Android open source defensive strategy successful; faces open source mapping challenges from Meta/Amazon | |
| Meta | Bullish on its open source strategy | Participating in open source mapping project; Llama model contributions |
| OpenAI | Risk warning | Burning $100-200 million annually (Gurley); Gurley questions its capital allocation efficiency |
| Uber | Classic example (positive) | Perfect cycle in value network: "more drivers → shorter pickup times → more users" |
| CrowdStrike | Positive example | Security cloud network effect: shared threat intelligence makes the largest player the safest |
| Tesla | Positive example | Pure AI path for autonomous driving; Wright's Law cost advantage |
| Shopify | Retrospectively positive | Gurley admits he missed it due to bias against "anti-small business model" |
| HubSpot | Retrospectively positive | Same as above |
| Twilio | Retrospectively positive | Same as above |
| Stripe | Neutral (expensive valuation) | Post-valuation of $90 billion; Gurley believes returns unlikely to reach the top 2% tail level |
| NVIDIA | Cautious (concentrated power) | Holds strong bargaining power in the TSMC ecosystem, creating barriers for startups |
| Apple | Neutral (AI potential unclear) | If it launches local/personal context models, could be a game changer |
| Boeing | Negative example (regulatory capture) | Metaphor of headquarters moving from Seattle → Chicago → D.C. |
1. Bill Gurley: “This AI wave is ‘highly orchestrated,’ which precisely undermines its disruptive nature.” — Historically, disruptive innovation is almost never anticipated in advance; when everyone knows the direction, incumbents will not fall behind.
2. Michael Mauboussin: “Pricing power is not the key; the real key is: are you increasing the price your customers are willing to pay every day?” — Raising Willingness to Pay either creates consumer surplus (a moat) or leaves room for future price increases.
3. Bill Gurley: “Missing a big one is a hundred times more fatal than avoiding a small one.” — Venture capital has only two types of errors, but “correctly judging which will fail” has almost no value, while “missing a big one” is a total loss.
4. Michael Mauboussin: “Wright’s Law says that each time cumulative production doubles, costs fall by 20%. It is the most accurate predictive model across 60 technologies.” — Solar energy and lithium batteries fit perfectly. Tesla’s cost advantage is essentially a lead in cumulative production.
5. Bill Gurley: “Open source is one of the most powerful innovations for human flourishing.” — Marginal cost is zero, and no one can be excluded from using it. Google used Android (open source) to fight iPhone (closed), and Kubernetes (open source) to fight AWS (closed) — both “defensive open source” strategies.
6. Michael Mauboussin: “Innovation is fundamentally a recombination of existing knowledge modules.” — This is the core of Paul Romer’s endogenous growth theory. Digitization exponentially accelerates the speed of search and trial in this process. Matt Ridley’s metaphor: “Let ideas have sex.”
7. Bill Gurley: “The core problem with regulatory capture is — we celebrate ‘exceptions that remove bureaucratic obstacles,’ and that itself shows we have already painted ourselves into a dead end.” — The I-95 bridge repaired in 12 days, the “exemption” for TSMC’s Arizona plant — these exceptions precisely prove that the normal state is already unsustainable.
8. Michael Mauboussin: “In the zero-interest-rate era, companies did not do what the textbook said: investment did not increase, cash did not decrease, leverage did not rise. The only thing that rose was buybacks.” — In reality, companies use a fixed 15% hurdle rate for investment and financing (rather than the true cost of capital) to make decisions — this is “anchoring” in the behavioral finance sense.
9. Bill Gurley: “When everyone understands the increasing returns game, the game becomes corrupted.” — The Uber/Lyft price war and the foundation model arms race are cases where “people who know the rules” push the game to extremes, and ultimately no one may win.
10. Michael Mauboussin on Danny Kahneman: “He is not only willing to be proven wrong, he actively seeks out dissenting opinions. If there is a truth out there, and I no longer hold a false belief, I have moved closer to that truth.” — This is cognitive “zero-based thinking,” and it also applies to areas suppressed by political baggage, such as nuclear energy and psychedelics.