I asked Claude to tell me which assets could benefit/lead.
Here's the tradeable map:
Win-either-way (the moat travels with them):
$NVDA — the bear case to your own thesis. They're porting the fortress, not ceding it: Jetson Thor is now generally available, a ~7.5x leap, already in Amazon Robotics, Boston Dynamics, Figure and Caterpillar, plus DRIVE for AV and the Isaac/Cosmos/GR00T simulation stack. The subtle counterpoint that supports you: edge Jetson is lower-ASP and more contested than datacenter — so even if Nvidia "wins" Wave 3, it may not compound the way Waves 1–2 did. The moat doesn't have to move to a rival; it can just thin. ExoswanThe Robot Report
$ARM — the toll booth. Energy efficiency is the literal pitch, and Arm sits under most edge devices. If the edge proliferates, royalties scale regardless of which chip wins. Arm Newsroom
$TSM — agnostic to the whole fight. Everyone's edge silicon is fabbed there, with CoWoS booked out. The jeans-in-the-gold-rush play. Institutional Investor
The actual Wave-3 bets (closest to "machines that act"):
$AMBA (Ambarella) — the cleanest liquid pure-play on your "smart cameras that decide" line. Edge AI inference SoCs for vision/automotive/security, with rising AI content per chip pushing ASPs up. Small-cap, volatile, most thesis-leveraged of the names you can actually trade. sec
$QCOM — the credible challenger to Jetson's developer lock-in. Acquisitions and a new processor line aimed at owning the edge stack silicon-to-software; the open question is whether it pulls developers away from the Jetson default. Exoswan
$NXPI — the automotive angle done differently. Zonal architecture shifts the car from many cheap control units to a few centralized compute chips, moving NXP from ~$5 commodity parts to high-margin platforms, hitting volume in 2026. Exoswan
The TinyML tail (billions of units, dollars each):
$CEVA,
$MCHP, Renesas — inference pushed down to the microcontroller, milliwatts and dollars, deployed in volume. Different risk/reward — high-volume, low-glamour, the unsexy endgame. Exoswan
Demand side (the robots themselves, not the silicon):
$TSLA — the wildcard that cuts against my moat thesis from the other direction: Optimus FSD on its own custom inference silicon. The vertically-integrated robot maker that builds its own edge chip is a third outcome neither "Nvidia wins" nor "merchant challenger wins."
$SYM,
$SERV and similar for warehouse/delivery robotics — demand pull, not chip exposure.