6:07 1. Name the asset the incumbent actually owns before pricing the threat
The repeatable method
- Write down what the company is popularly called, then list every input its product needs and mark who paid for each. What the company itself did not build is not its moat.
- Whatever is left — the coordination, the data, the network, the standard — is the real asset. Restate the business in those terms.
- Now re-run the disruption argument against that asset specifically. A threat to the popular description is not automatically a threat to the asset.
Here: UBER didn't build the roads (taxpayers), the fuelling network (CVX and peers), the cars (drivers' own, idle ~95% of the day) or the labour pool. What it owns is allocation — dispatch, surge prediction, pricing, reputation. So "a physical capital allocator," and the AV threat has to be measured against allocation, not against driving.
Watch for
- A bear case aimed at an input the company never owned; a business whose capex line is suspiciously small relative to its footprint; the popular label ("ride-sharing app") doing analytical work it can't support.
10:27 2. Find the local unit of competition — and the threshold that makes it work
The repeatable method
- Ask at what geographic or segment level the network effect actually operates. Global scale is often irrelevant; the economics are usually local.
- Define the threshold — the level of local supply at which the customer experience becomes reliable and the supplier's utilization becomes attractive.
- Score each competitor by whether they clear that threshold per market, not by their national or global size.
- Track whether the challenger has ever pushed the incumbent back below the threshold in any market. If not, headline expansion isn't yet competitive damage.
Here: "It's not scale… it's density." Critical density = any rider served promptly and any driver who logs on gets fares. LYFT never clears it, so "they don't have a chance"; and even in Waymo's most mature cities, "nowhere has Waymo knocked them out of critical density."
Watch for
- Market-share statistics quoted nationally when the flywheel is per-city; a challenger's city count rising while the incumbent's local wait times and utilization stay flat; the incumbent forced back into subsidising supply (the tell that density has broken).
17:57 3. Test whether the disruptor's proof point is a representative market
The repeatable method
- Identify the single market everyone cites as evidence the new model works.
- List the variables that make the new model's unit economics work, then measure that market against the national average on each one.
- If the beachhead is an outlier on several variables at once, discount the extrapolation — expect worse utilization, pricing and uptime elsewhere.
- Ask the obvious follow-up: why did they start there? Rational operators start where the economics are easiest, which is exactly why the first market overstates the general case.
Here: San Francisco is ~2× the population density of LA and 4–5× the US top-50 average (density favors AVs), roughly twice as affluent (riders pay the AV premium), and has about half the average trip distance. "It would be a mistake for investors to look at San Francisco… and to extrapolate that everywhere else."
Watch for
- A single flagship city or cohort carrying an entire TAM argument; expansion into ordinary markets quietly showing lower utilization; the disruptor changing which metric it reports as it leaves the beachhead.
20:34 4. Check whether the new model can serve peak demand — the base-load test
The repeatable method
- Chart demand by hour and by day of week. If it spikes rather than plateaus, the peak-to-trough ratio is a first-class competitive variable.
- Ask who bears the cost of the idle capacity between peaks. Owned capacity (fleets, plants, salaried staff) carries it; borrowed or on-demand capacity does not.
- State the resulting ultimatum explicitly: the owned-capacity operator must either hold a fleet idle most of the day or forfeit the busiest hours. There is no third option.
- Check which one the operator has actually chosen — that reveals the true addressable share.
Here: Friday demand spikes "like six times what an AV can handle." AVs "serve as a fixed base load supply"; human drivers "provide boost supply to meet demand spikes." An AV-only rival "must hold significant… underutilized supply to match Uber's reliability and price, or deliver worse consumer experience and leave significant demand on the table" — and so far they are choosing to skip the peak.
Watch for
- Demand that surges around events, weather, commutes or seasons; a competitor reporting rides per week without saying when; fleet-shuffling between cities for big events as a sign the peak isn't solved.
26:39 5. Ask which player can afford to arrive late — sequence the two build-outs
The repeatable method
- Break the end-state product into its components (here: the autonomous technology, and the demand network it runs on).
- For each competitor, mark which components they already have and which they must build, and how many years each build takes.
- The player missing only the fast component can afford to be years late; the player missing the slow component cannot make it up with capital.
- Convert that into an explicit deadline for the incumbent, and judge the stock against that deadline rather than against the next headline.
Here: Uber lacks AVs (buyable, licensable, and "no ASML machine"); Waymo lacks per-city density (multi-year, capital-heavy, red-taped). So Uber "can be very late to the game… and still be in the lead" — the Disneyland line-skip — and Carlson's stated deadline is 5–7 years, with a full decade of inaction the only losing branch.
Watch for
- The incumbent's first real AV integrations (or continued absence) against that 5–7 year clock; any city where the challenger reaches full density and the incumbent's wait times slip; technology that turns out to be harder to license than assumed.
37:32 6. When supply will commoditize, own the aggregator
The repeatable method
- Judge whether the contested technology is genuinely unrepeatable or merely difficult — benchmark it against a real monopoly technology you already understand.
- If several credible suppliers will exist within the investment horizon, treat the supply as commoditizing.
- Then position in the layer that aggregates demand across those suppliers, because that layer captures the economics once supply is plentiful.
- Sanity-check with precedents in other industries where the same shape has already played out.
Here: "AVs are impressive technology. But it's no ASML machine" — lidar plus cameras, replicable by others. So the profits accrue to the aggregator, "and that's what happens in all different industries. Like NFLX is an aggregator, and YouTube's an aggregator, BKNG is an aggregator, the App Store from AAPL and the Google Play Store."
Watch for
- A second and third credible supplier appearing (the commoditization tell); a single supplier achieving something no one else can replicate (the thesis-breaker); the aggregator being cut out of distribution rather than losing on technology.
38:24 7. Pre-commit to the headlines you'll ignore
The repeatable method
- Once the mechanism is understood, list in advance the news that will look alarming but is already inside the thesis (here: each new Waymo city).
- Separately, list the news that would actually falsify it — the metric, not the announcement.
- Hold the position against the first list; act on the second. This is what keeps a multi-year structural view from being traded out on a news cycle.
Here: "You're going to see constant headlines of the advancement of Waymo over the next five years. That's going to happen. And it might keep Uber stock down a little bit because investors get gloomy… But just know the timeline." The falsifier is different in kind: Uber failing to integrate any AVs for a decade, or Waymo pushing Uber below critical density somewhere.
Watch for
- Confusing an expansion announcement with a share loss; a falsifier you can't state as a measurable number; your own conviction moving with the stock rather than with the metric.