Autonomy and Robotics

The question

What does the market have to believe about the economics of physical autonomy — not just the existence of the technology — for this book to compound above SPY over the next three years? The bullish story for robotics and autonomy has been told continuously for a decade and has repeatedly been a bad trade at the index level. The interesting question is narrower: where within the physical-AI stack does the capital intensity of the enabling infrastructure, combined with the revealed cadence of industrial adoption, produce an imputed probability of cash-flow inflection that is lower than reality? The book is built to answer that question with a barbell: large concentrations in names with contracted revenue and large installed bases, and a tail of specialist positions where the reference class supports asymmetric payoff.

Base rates

The reference class for autonomy platform monetization is thin. Autonomous vehicle fleet economics, humanoid production economics, and general-purpose robotics margins are all estimated from a handful of data points. The base rate for a named autonomy program to hit its original commercial milestone on its original schedule is low — on the order of one in four based on the last decade of public guidance. The base rate for industrial robotics integrators to compound earnings at or above the broad market over rolling five-year windows is far more favorable — roughly two in three for names with installed-base service revenue above 40% of total. Defense autonomy primes have base rates closer to the industrial cohort than to the autonomy-platform cohort: contracted backlog, dollar-denominated pricing, and decade-length programs. The book allocates weight in rough proportion to these conditional base rates.

Why the consensus view is wrong (or incomplete)

Consensus treats autonomy and robotics as a single thematic basket and prices the basket on the expected date of the most visible platform milestone. That framing gets the reference class wrong. It over-weights the narrative-driven platform names and under-weights the industrial and defense beneficiaries who earn revenue regardless of whether the platform milestones hit on schedule. The expected-value error runs in both directions: consensus pays too much for calendar-specific catalysts on platform names, and too little for the steady compounding of the picks-and-shovels layer — automation controls, motion components, grid-side interconnect, test equipment, and vision systems. The imputed probability embedded in current prices for the industrial layer implies a slower adoption curve than the observed orders and backlogs support.

Position construction — name each holding's role

Tesla (TSLA, 20%) is the autonomy-platform position. Sized at the cap because of optionality value, not because of conviction in any specific calendar event. RTX (RTX, 20%) is the defense-prime anchor — diversified across missiles, avionics, and propulsion, with backlog visibility into the 2030s. Lockheed Martin (LMT, 16%) is the complementary defense prime, weighted behind RTX for program concentration risk. Parker-Hannifin (PH, 13%) is the motion-and-control position — the picks-and-shovels exposure that compounds regardless of which autonomy platform wins. Emerson (EMR, 10%) is the process-automation and controls position. Symbotic (SYM, 5.6%) is the warehouse-automation specialist; small weight reflects customer concentration and high expectations premium. Rockwell (ROK, 5.4%) is a factory-automation complement to EMR. Hubbell (HUBB, 3.1%) is grid-and-connector exposure — motion and automation need power delivery. Teradyne (TER, 3.0%) is semiconductor test, tied to autonomy compute. AeroVironment (AVAV, 2.4%) is small-UAS defense autonomy; asymmetric on contract flow. Cognex (CGNX, 1.1%) is machine vision — a toehold in an industry with high fixed costs. Iridium (IRDM, 0.3%) is satellite connectivity for autonomous systems, held as a nominal position.

Asymmetric payoff — rough EV framing

The three-year base case (55%) has the book compounding at roughly 10–12% as defense backlog and industrial automation earnings carry the outcome. The upside case (25%) has at least one autonomy-platform milestone landing within its guided window and delivers 20%+ annualized, driven mostly by TSLA and SYM. The downside case (20%) has an autonomy winter combined with a defense budget ceiling and tracks SPY within a few hundred bps. Expected value is low- to mid-teens. The asymmetry comes from the tail positions, not from the anchors; the anchors are there to make the tail affordable.

Three things that would change our mind

First, a defense budget outcome that caps topline growth on RTX and LMT below nominal GDP for two consecutive authorization cycles. Second, a slowing of industrial automation order books — measured at EMR, ROK, and PH — that persists for three quarters. Third, evidence that autonomy compute is concentrating inside a single vertically integrated platform to the exclusion of third-party suppliers.

What we are explicitly NOT betting on

We are not betting on any specific robotaxi launch date. We are not betting on humanoid unit economics at scale in this decade. We are not betting on a peace dividend or a defense spending cut. We are not betting on a single autonomy-stack winner; the book is deliberately built so that the industrial and defense layers carry the outcome if the platform layer slips.