Risk Grok
Skim: How value fails even if pieces “work.” Stranded cheap fleet · city lock-in wins · hard-city surprise · disclosure cliff · densify stall · fatal cluster · China category error.
Opinion-heavy on purpose. Each card: failure mode, second-order effect, what would change your mind.
Regulatory FMVSS self-cert vs purpose-built AV
Fact NHTSA opened AQ26002 after Austin Cybercab commercial deploy, then escalated to a formal Special Order (Sep 30 sworn deadline passed — no public outcome yet) on FMVSS self-cert for a vehicle lacking permanently attached wheel/pedals/mirrors — including foot-brake No. 135, temporary controls, and legality without a Part 555 exemption (Zoox holds one).
Second-order · check
Second-order: if self-cert does not hold, Cybercab scale path hits exemption caps / redesign / delayed utilization — cost-edge thesis stalls even if software works. Check: Special Order response / AQ disposition, any forced Part 555-style exemption path (and caps), and whether paid Austin unsupervised hours continue through the inquiry.
Capex Cheap vehicles, no utilization
Judgment Classic Autonomy risk: build the low-cost fleet before unsupervised demand and permission arrive. Lots fill; regs tighten; capital sits in metal. The cost edge becomes a carrying-cost edge.
Second-order · check
Second-order: lot inventory without unsupervised miles trains the market to treat robotaxi Capex as optionality burn. Check: lot growth vs unsupervised mile growth — divergence is the tell.
Moat City lock-in wins the decade
Judgment If generalization fails, Waymo-style densification compounds while camera-only stays stuck in easy geographies. Tesla’s equity story then depends on other businesses; robotaxi optionality decays.
Second-order · check
Second-order: markets may keep pricing “Autonomy someday” long after the ops reality is city-by-city slog. Check: hard-city unsupervised transfer speed — Barriers §4.
Safety Hard-city local surprise
Judgment National or easy-city averages can look fine while one hard city prints worse-than-human severity. Local regulators and insurers don’t average nationally — they ground locally.
Second-order · check
Second-order: one city’s grounding becomes a narrative contagion across the stack. Check: city-level independent prints and sudden geofence shrinks after incidents.
Market Disclosure cliff
Judgment TSLA can trade robotaxi optionality on ambiguous metrics. Risk: a forced clarification (incident, subpoena, quiet quarter) collapses the Autonomy slice of the multiple without a matching cash miss — pure narrative unwind.
Second-order · check
Second-order: invite/demo confusion with paid unsupervised makes the cliff steeper. Check: whether language stays carefully vague after peers publish clearer unsupervised stats.
Ops Densify stall / thin expansion
Judgment Waymo path risk: metro logos rise while rides/car/day stall. Fixed costs don’t amortize; unit economics never close; the “proven stack” stops compounding.
Second-order · check
Second-order: thin expansion invites Tesla narrative that city lock-in is a trap. Check: rides/week trajectory in core metros vs new logos.
Tail Fatal cluster grounds a stack
Judgment One multi-fatality cluster — or a pattern that looks like one — can freeze unsupervised permission for years regardless of prior averages. Tail risk is not “PR”; it is the binding constraint on the whole race.
Second-order · check
Second-order: grounding one stack can temporarily gift share to the other — or poison the category. Check: regulator response speed and whether peers get collateral scrutiny.
Frame China category error
Judgment Importing China weekly rides into the US race ranking misprices both Tesla and Waymo. Risk is analytical, not ops: portfolios act on the wrong scoreboard and miss the actual gates (insurance, hard-city, disclosure).
Second-order · check
Second-order: bull decks that lead with China scale train false urgency. Check: whether investment memos separate control-group evidence from US clearance evidence — Gate F.