Physical-AI verticals vs. chip-fab Phase 2 — mostly counter-phased, none cleanly at peak capex (mid-2026)

The question

Which physical AI companies (manufacturing sensors, warehouse automation, precision agriculture, industrial vision) show capital-cycle signals analogous to chip-fab Phase 2 (peak capex, approaching utilization tipping point) as of mid-2026, and are any publicly investable today? Context: extends RDCO's named-phase chip-fab/memory capital-cycle framework (Phase 1 demand → Phase 2 peak-capex/approaching-tipping → Phase 3 oversupply) to adjacent physical-AI verticals; the Markov phase-tracker build is live for memory.

What we already know (from the vault)

The Markov pipeline runs a P1→P4 ladder: P1 tight-supply/rising-ASP → P2 capacity-announcements → P3 capacity-online/glut/ASP-rollover → P4 capex-cuts/rationalization. The DRAM/HBM cycle is currently pinned in deep Phase 2 with exactly one earliest-tripwire flicker (price-gain deceleration), zero Phase-3 triggers fired [[2026-07-07-dram-hbm-phase2-phase3-early-signals]], [[2026-05-27-markov-equities-pipeline-spec]].
Critical structural read from the memory work: this cycle's capex is composition-shifted (node migration + HBM packaging, not greenfield wafers), so the classic "capex dollar → capacity glut → ASP rollover" Phase-3 trigger has not engaged. Watch capex composition and the capacity-online date, not the announcement or the dollar level [[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]].
The physical-AI thesis exists as a project but is thin and demand/SMB-framed ("outcomes not platforms," physical-world AI), not equity-cycle framed [[2026-05-03-opportunity-map]]. Robotics coverage is weak-mapped: the "loop is the moat" vertical-integration frame [[2026-05-09-avedissian-loop-is-moat-robotics]] and ARK's promotional $26T robotics TAM / end-of-decade humanoid timeline [[2026-05-20-ark-robots-beyond-factory]].
Physical Intelligence / Sergey Levine material anchors the robotic-foundation-model layer — the genuinely capex-heavy frontier of physical AI is still overwhelmingly private and pre-revenue [[2025-09-12-dwarkesh-patel-sergey-levine-fully-autonomous-robots]].

What the web says

Precision agriculture is in a down-leg, not a peak. USDA forecasts 2026 net farm income at $153.4B (−0.7% nominal, −2.6% real); Deere guides North America large-ag industry sales −15% to −20% for 2026 and production/precision-ag sales −5% to −10%; Deere, CNH, and AGCO are all deliberately underproducing to destock dealer inventory, and CNH tells investors to "wait until 2027" for recovery (Manufacturing Dive, Feb 2026; DTN/Progressive Farmer, Feb 2026). Deere also carries ~$1.2B of 2026 tariff cost.
Industrial vision / factory automation is at a trough turning to early recovery. 2026 is described as "the beginning of renewed momentum" after several slow years, with Roland Berger projecting 6–9% annual growth through 2030; drivers are reshoring, factory modernization, semiconductor manufacturing, and robotics (Engineering.com, 2026). Factory-automation-and-machine-vision market sized ~$90B (2024) → ~$150B (2033), ~6% CAGR (Verified Market Reports). Cognex, Zebra, Keyence are the named vision incumbents.
Warehouse automation is re-accelerating (demand ramp), post its 2021–22 overbuild/2023–24 hype-cycle digestion. Global market projected to ~$64B by 2032 at ~12% CAGR from 2026 (TheLogisticsIQ); the "after the boom" framing confirms a prior overbuild was already digested (RoboticsBiz).
Symbotic (NASDAQ: SYM) is the clearest public pure-play and reads mid-ramp, not peak. GAAP-profitable the last two quarters, cash grown to ~$2B (from $1.25B in Q4 2025), 11 customers / 70+ system deployments, 14 new deployments started in Q2 2026 (up from 10 in Q1), software revenue $13mm (92% recurring); expansion via Fox Robotics (autonomous forklifts), SymMicro (next-gen fulfillment, "$5B+" TAM), and Exol warehouse-as-a-service (Monte Investments, Symbotic Q2 FY26 review). Adjacent public exposure: GXO Logistics (NYSE: GXO) piloting Agility Digit humanoids.

Convergences and contradictions

Convergence: vault and web agree the frontier of physical AI (humanoids, robotic foundation models) is capex-heavy but private/pre-revenue; the public physical-AI verticals are legacy-industrial cyclicals whose phase is set by their end-customer's capex, not by a commoditized supply cycle.
Contradiction with the premise: the question assumes a physical-AI vertical analogous to chip-fab Phase 2 exists today. The data says the opposite — as of mid-2026 the investable verticals are counter-phased to the memory cycle: ag is at Phase 3→4 (destock/trough), industrial vision at Phase 4→1 (trough/early recovery), warehouse automation at Phase 1→2 (demand ramp). None is at "peak capex approaching a utilization tipping point."

Synthesis for RDCO

The honest, calibrated answer: no publicly investable physical-AI vertical is at a chip-fab-style Phase 2 (peak capex, approaching utilization tipping point) as of mid-2026 — they are spread across the other three phases, and the vertical closest to Phase 2 (warehouse automation) is on the early side of the ramp, not the top. Mapping the named verticals onto the P1→P4 ladder: (1) Precision agriculture — Deere (DE), CNH (CNH), AGCO (AGCO) — is at Phase 3→4: deliberate underproduction, dealer destocking, large-ag −15–20%, precision-ag −5–10%, recovery pushed to 2027. This is a trough-approach / contrarian-mean-reversion setup, the inverse of Phase 2. (2) Industrial vision + manufacturing sensors — Cognex (CGNX), Zebra (ZBRA), Teledyne (TDY), Rockwell (ROK), Keyence (TSE 6861), Omron (TSE 6645) — sits at Phase 4→1: multi-year automation downturn troughing, 2026 the first "renewed momentum" year at 6–9% forward CAGR. Early-cycle accumulation, not peak. (3) Warehouse automation — Symbotic (SYM), GXO (GXO), KION/Dematic (FRA: KGX), AutoStore (OSL: AUTO), Ocado (LSE: OCDO) — is at Phase 1→2: demand re-accelerating (~12% CAGR), Symbotic profitable and expanding deployment count QoQ. This is the only vertical with a rising-capex signature, but the read is accelerating demand, not an approaching top. All named tickers are publicly investable today; the pure-play robotic-AI frontier (Figure, Physical Intelligence, Agility, Skild) is not — it is private and pre-revenue.

Where the chip-fab analogy breaks — and this is the load-bearing insight for extending the Markov framework: the memory cycle is a supply-side commodity-capacity cycle (a 3-firm oligopoly builds fungible bit-capacity; Phase 2→3 is that capacity coming online and rolling over a single commoditized ASP). The physical-AI verticals are demand-side, buyer's-capex cycles — ag equipment cycles on farm income/commodity prices and used-equipment inventory; machine vision on factory-automation capex and book-to-bill; warehouse automation on e-commerce/logistics penetration and per-system ROI. There is no single fungible output whose price rollover marks the top. So the phase ladder's driver flips from supplier-capacity to customer-utilization, and the observable trigger set must be re-specified per vertical rather than reused from DRAM: ag → dealer new+used inventory-days, used-equipment price index, net farm income (a Phase-3/4 destock tracker); industrial vision → factory-automation capex growth, ISM/PMI, incumbent book-to-bill (a Phase-1 recovery tracker); warehouse automation → e-commerce penetration, deployment backlog conversion, and deployed-system utilization/ROI (the closest to a genuine Phase-2 tipping-point tracker, because oversupply here shows up as falling incremental-system ROI, not a spot price).

The strategic implication is not "find the physical-AI vertical that rhymes with chip-fab Phase 2" — it's that these verticals are cycle-desynchronized from memory, which is a portfolio feature. A Markov tracker that labels all four (memory Phase 2, ag Phase 3–4, vision Phase 4–1, warehouse Phase 1–2) turns the physical-AI cluster into a counter-phased complement to the semiconductor bet rather than a correlated double-down. The lowest-effort, highest-signal next build is a warehouse-automation utilization/ROI tripwire around Symbotic (backlog-conversion rate + incremental gross margin per new system), because it is the one vertical where a real Phase-2→3 "utilization tipping point" is mechanically definable and Symbotic is the one clean, profitable public pure-play to instrument it on.

Why this is in the vault

This is the first equity-cycle-framed entry in the otherwise thin, demand-side physical-ai-thesis project, and it gives the Markov phase-tracker a concrete extension target: a per-vertical trigger set (ag inventory-days, vision book-to-bill, warehouse system-ROI) that re-uses the P1→P4 ladder while correctly flipping the driver from supplier-capacity to buyer-utilization — plus a named, investable universe (SYM, DE, CNH, AGCO, CGNX, ZBRA, TDY, ROK) to instrument it on.

Open follow-ups

Build a Symbotic (SYM) Phase-2→3 tripwire: backlog-conversion rate, deployments-started QoQ, and incremental gross margin per new system — the physical-AI vertical with a mechanically definable utilization tipping point.
Construct an ag-equipment destock tracker (Deere/CNH/AGCO dealer new+used inventory-days + a used-equipment price index) to time a Phase-4→1 contrarian entry into the ag downcycle.
Instrument a machine-vision book-to-bill / factory-automation-capex early-recovery signal across Cognex + Rockwell + Zebra to confirm or falsify the "2026 = trough turn" narrative before sizing.
Decide whether the physical-AI cluster is worth a dedicated Markov thesis vs. folding these four verticals in as counter-phased overlays to the [[2026-05-18-memory-cycle-v1.1|memory]]/NVIDIA supply-chain theses.
Map the private robotic-AI frontier (Figure, Physical Intelligence, Agility) for the public proxy that gives the cleanest exposure (e.g., component/sensor suppliers, actuators, or the vision incumbents) since the pure-plays are un-investable.

[[2026-05-27-markov-equities-pipeline-spec]] — the Layer-1 phase-tracker and P1→P4 ladder this brief extends to physical-AI verticals
[[2026-07-07-dram-hbm-phase2-phase3-early-signals]] — the chip-fab Phase 2→3 tripwire set this brief mirrors (and shows physical-AI is counter-phased to)
[[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]] — "watch capex composition, not dollar level" insight, re-applied per-vertical here
[[2026-05-03-opportunity-map]] — the physical-ai-thesis project doc this is the first equity-cycle entry for
[[2026-05-09-avedissian-loop-is-moat-robotics]] — weak-mapped physical-AI robotics coverage; vertical-integration moat frame
[[2026-05-20-ark-robots-beyond-factory]] — physical-AI robotics TAM/timeline context (treat as ARK book-talking)
[[2025-09-12-dwarkesh-patel-sergey-levine-fully-autonomous-robots]] — the private, pre-revenue robotic-foundation-model frontier

Sources

Vault

[[2026-05-27-markov-equities-pipeline-spec]] — ~/rdco-vault/01-projects/investing/2026-05-27-markov-equities-pipeline-spec.md
[[2026-07-07-dram-hbm-phase2-phase3-early-signals]] — ~/rdco-vault/06-reference/research/2026-07-07-dram-hbm-phase2-phase3-early-signals.md
[[2026-06-28-chip-memory-cycle-phase2-phase3-indicators]] — ~/rdco-vault/06-reference/research/2026-06-28-chip-memory-cycle-phase2-phase3-indicators.md
[[2026-05-03-opportunity-map]] — ~/rdco-vault/01-projects/physical-ai-thesis/2026-05-03-opportunity-map.md
[[2026-05-09-avedissian-loop-is-moat-robotics]] — ~/rdco-vault/06-reference/2026-05-09-avedissian-loop-is-moat-robotics.md
[[2026-05-20-ark-robots-beyond-factory]] — ~/rdco-vault/06-reference/2026-05-20-ark-robots-beyond-factory.md
[[2025-09-12-dwarkesh-patel-sergey-levine-fully-autonomous-robots]] — ~/rdco-vault/06-reference/2025-09-12-dwarkesh-patel-sergey-levine-fully-autonomous-robots.md

Web

physical ai capital cycle phase2 analogs