2026-06-18 👁️ — views

Physical AI Supply Chain Analysis — Hardware Makers, Bottlenecks, and the Ramp Constraints

Who makes the chips, sensors, and actuators powering physical AI fleets — and which supply chain bottlenecks could stall the ramp even if the software is ready.

Article 32 in the Physical AI Benchmark Series — The Hardware Behind the Ramp

You cannot build a Cybercab or a Waymo Gen 6 vehicle with software alone. Every autonomous vehicle and humanoid robot is a dense assembly of specialized hardware: compute chips, sensors, actuators, batteries, and a vehicle chassis that must be sourced, manufactured, and delivered at scale. The software race in physical AI dominates headlines. The hardware supply chain is where the actual constraints live.

This article maps the component supply chain behind physical AI: who makes the critical parts, which players have concentrated control over key inputs, and what supply chain bottlenecks are most likely to constrain fleet and humanoid scaling in 2026–2028 — even if the AI software is ready to deploy. All unit cost figures and lead time estimates are based on industry reporting and analyst research and are labeled as estimates throughout.

Section 1 — Component Supply Chain Map

The table below covers every major hardware category required to build an AV fleet or humanoid robot at commercial scale. For each component, it identifies key suppliers, Tesla’s and Waymo’s sourcing approaches, and a supply risk assessment.

Reading the table: Supply risk is not uniform. Two categories stand out as genuinely high-risk: humanoid precision actuators (scarce global supply, long lead times) and Waymo’s vehicle chassis dependency on Zeekr (geopolitical exposure). Everything else sits in the medium-to-low range, either because multiple suppliers exist or because one party has vertically integrated. The risk asymmetry between Tesla and Waymo on the chassis row is the single most important supply chain fact in this analysis.

Section 2 — The Three Critical Bottlenecks

Bottleneck 1: Waymo’s Zeekr Dependency (HIGH Risk)

Waymo’s Gen 6 vehicle is manufactured by Zeekr, a Geely subsidiary headquartered in China. This is not a minor sourcing detail — it is the structural foundation of Waymo’s fleet expansion plan, and it carries compounded risk on three dimensions.

Tariff risk: US-China tariffs on electric vehicles currently exceed 100% following escalating trade actions. Waymo has operated under commercial AV import classifications, but this exemption is politically vulnerable. If it is revoked, the landed cost of each Waymo Gen 6 vehicle increases dramatically — potentially disrupting fleet economics.

Geopolitical risk: Any significant deterioration in US-China trade relations could disrupt Zeekr vehicle production or export approval entirely. Waymo has no disclosed alternative manufacturing source for the Gen 6 platform. A supply interruption does not slow fleet growth — it stops it.

Concentration risk: 100% of Waymo’s new fleet vehicles come from a single overseas manufacturer. There is no disclosed dual-source or domestic backup production path.

This is arguably the most underappreciated risk in Waymo’s scaling plan. The company’s operational record — safety, dispatch efficiency, rider experience — is strong. But all of that operational excellence depends on a chassis supply line that runs through geopolitical terrain that neither Waymo nor Alphabet controls.

Bottleneck 2: Humanoid Precision Actuators (HIGH Risk for All Humanoid Makers)

Harmonic Drive actuators — the joint mechanisms that give humanoid robots precise, back-drivable motion — are:

• Manufactured primarily in Japan (Harmonic Drive Systems) and Germany (Harmonic Drive AG)
• Expensive at approximately $200–500 per joint (est.), with pricing varying by torque rating and configuration
• Capacity-constrained, with reported lead times of 12–24 weeks (est.) under normal demand conditions
• Required in large quantities: a humanoid robot typically requires 20–40 actuated joints

The scale implication is severe. If Tesla targets 50,000–100,000 Optimus units in 2027 (Musk’s stated target), that implies 1–4 million high-precision actuators per year. Current global production capacity for harmonic drive actuators is orders of magnitude below that number, sized for industrial robotics demand measured in thousands of units annually — not millions.

Tesla is reportedly developing in-house actuators for Optimus specifically to escape this constraint. The development timeline for a production-ready proprietary actuator that meets Optimus’s precision and durability requirements is itself a significant risk factor — if the in-house program slips, the external supply constraint becomes binding.

This bottleneck affects every humanoid maker simultaneously: Figure AI, Agility Robotics, Boston Dynamics, Unitree, and 1X all draw from the same global actuator supply. There is no incumbent with captive supply except Tesla if its in-house program succeeds.

Bottleneck 3: TSMC Advanced Node Concentration

Both Tesla’s HW4 FSD chip and NVIDIA’s Drive Orin and Drive Thor are fabricated at TSMC on advanced process nodes (4nm and 3nm). This creates a shared structural dependency:

Geographic concentration: TSMC’s advanced node fabs are concentrated in Taiwan. Seismic risk and geopolitical risk — in particular, the cross-strait situation — are neither theoretical nor remote. Any disruption to TSMC’s Taiwan operations affects every company that relies on advanced-node fabrication.

Capacity competition: Tesla, NVIDIA, Apple, AMD, and Qualcomm compete for the same TSMC capacity at 3nm and 4nm. In periods of high demand, automotive customers have historically been lower priority than consumer electronics. If AI server chip demand from hyperscalers continues to grow, automotive allocation at advanced nodes could face pressure.

Lead time: Design-to-volume-production timelines at advanced nodes run 18–24 months. A new chip design cannot be taped out and in production within a single product cycle. This means supply chain errors have long tails — a capacity shortfall identified today cannot be remedied within a year.

Section 3 — Tesla’s Vertical Integration Advantage

Tesla’s supply chain strategy is a deliberate program of dependency reduction. Across every major hardware category, Tesla’s approach is to move from external sourcing toward in-house design or manufacture. The strategic logic is consistent: external suppliers create pricing leverage, capacity constraints, and competitive intelligence exposure. Vertical integration eliminates all three.

The camera-only approach to AV sensing is worth highlighting separately. By eliminating LiDAR entirely, Tesla removed a $1,000–7,000+ per-unit component (est.) from the bill of materials — and simultaneously eliminated exposure to LiDAR supply constraints. Whether camera-only is sufficient for full autonomy is a technical debate. As a supply chain strategy, it is unambiguously cleaner.

The 4680 cell program serves the same supply chain logic as HW4. Tesla is not developing in-house battery cells because it can manufacture them more cheaply than CATL in the near term. It is developing them to reduce dependence on a single dominant supplier whose capacity decisions, pricing, and geopolitical exposure are outside Tesla’s control.

If Tesla’s in-house actuator program for Optimus succeeds on timeline, Tesla will have vertically integrated the single most constrained component in humanoid manufacturing — creating a supply chain moat that no competitor can replicate quickly.

Section 4 — Supply Chain Implications for the 2027 Ramp

The table below maps the most likely binding constraint for each fleet or production scale milestone, based on the supply chain analysis in sections 1–3. All targets and lead time estimates are approximate.

What this table implies:

At the Waymo 5,000-vehicle scale, the binding constraint is mundane: factory production rate and logistics. At 10,000 vehicles, geopolitics becomes the binding constraint — and geopolitics does not respond to capital or engineering effort.

At Optimus 10,000 units, the external actuator supply chain is binding. This is solvable with capital — Harmonic Drive and Maxon can expand capacity given sufficient lead time and purchase commitments. At Optimus 100,000 units, only Tesla’s own in-house actuator program resolves the constraint. No external supplier can ramp from thousands to millions of precision actuators per year within a two-year window.

The TSMC concentration risk sits across the entire industry. It is the least likely near-term disruption (Taiwan risk is a known geopolitical concern with active hedging by TSMC through Arizona capacity expansion) — but it is also the hardest to resolve if it materializes, because there is no alternative foundry for 3nm-class automotive SoCs.

Section 5 — Who Controls the Commanding Heights

Supply chain analysis ultimately asks: who has leverage over whom? The answer in physical AI is becoming clearer as the industry scales.

TSMC has leverage over every AV compute chip in the industry. Its pricing, allocation, and roadmap decisions affect Tesla and NVIDIA simultaneously. TSMC’s Arizona expansion reduces but does not eliminate this leverage.

CATL has leverage over EV battery supply for most non-Tesla AV fleet operators, including Waymo’s Zeekr vehicles. CATL’s pricing and export decisions are also subject to Chinese government policy.

Harmonic Drive Systems and Harmonic Drive AG have leverage over every humanoid robot maker that does not have an in-house actuator program. This leverage is currently underpriced by the market because humanoid production volumes are still too small to stress the supply chain.

Tesla is systematically reducing its exposure to all three. HW4 reduces TSMC leverage (still TSMC-fabbed, but custom design removes NVIDIA pricing leverage). The 4680 program reduces CATL leverage. The in-house actuator program aims to remove Harmonic Drive leverage.

Waymo is moving in the opposite direction — deepening its dependency on Zeekr rather than reducing it. This is partly a strategic choice (focus on software and operations; outsource hardware) and partly a resource constraint. Alphabet has not committed to building domestic AV manufacturing capacity.

The supply chain structure of physical AI in 2026 reflects a broader strategic divergence: Tesla integrates vertically to control its supply chain; Waymo specializes in software and operations and accepts supply chain dependencies. Both strategies can win — but they fail for different reasons. Tesla’s failure mode is internal execution risk. Waymo’s failure mode is external disruption risk.

Section 6 — About This Series

This is article 32 in the Physical AI Benchmark Series. Previous articles in this series have covered the ramp index, the humanoid race, unit economics, global competition, HD mapping, fleet operations, software and OTA, insurance and liability, consumer demand, partnerships, competitive moats, Cybercab versus Model Y, safety data, Waymo Gen 6, Optimus manufacturing, scorecard snapshots, the 2030 forecast scenarios, the investor framework, Waymo’s city expansion pipeline, Tesla’s state approval map, AV weather and climate constraints, the talent war, the regulatory calendar, robotaxi fare pricing, the AV data flywheel comparison, and the humanoid deployment tracker.

This article adds the supply chain dimension: the hardware map beneath the software race, the three high-priority bottlenecks, and the structural divergence between Tesla’s vertical integration strategy and Waymo’s dependency-accepting specialization. The supply chain does not determine who wins. But it determines the failure modes, the ceiling on scaling speed, and the points of vulnerability that competitors and geopolitics can exploit. As physical AI scales from thousands to millions of units, the hardware behind the ramp will matter as much as the software running on it.

Sources

• Waymo Gen 6 vehicle — Zeekr partnership — Waymo blog ↗
• Tesla HW4 FSD chip — Tesla AI ↗
• NVIDIA Drive Thor automotive SoC — NVIDIA ↗
• Harmonic Drive actuator supply — Harmonic Drive Systems ↗

Tags