What is an actuator

Frame

If actuators are the chokepoint of physical AI, then capability around actuators is a strategic asset. The Deloitte-format playbook below treats the asset as a system, not a part. Three lenses: capability model, operating model, risk register.

Capability model

A firm’s actuator capability is the sum of five sub-capabilities. Most firms are strong in two or three, weak in the others. Knowing which is which determines whether you build, buy, or partner.

Specification. Do you know what the load actually demands. Force, speed, duty cycle, environment, life. Most failures upstream of an actuator selection trace to bad specs.
Selection and sourcing. Can you map a spec to a family, a topology, and a supplier in under a week. The answer for most general contractors is no. The answer for Tesla and Figure is yes.
Integration. Mechanical mounting, thermal management, electrical interface, software handshake. This is where in-house engineering earns its salary.
Manufacturing. Tier-1 actuator production at the volumes a humanoid platform requires is its own discipline. The learning curve is two to four years from a cold start.
Service and aftermarket. Predictive maintenance, spares, end-of-life recycling. The cost of ownership tail is 30 to 50 percent of TCO over a 10-year life.

A weakness in any one sub-capability constrains the whole. A strength in specification covers a multitude of sourcing sins. A weakness in specification turns even the best supply chain into the wrong one.

Operating model

The operating model question is whether actuator capability lives inside the line organization, inside a center of excellence, or in a captive supply company. The answer depends on volume.

Below 1,000 units per year. Buy off the shelf. The bill is small, the engineering load is manageable, and customizing is more expensive than living with a 70 percent fit.
1,000 to 50,000 units per year. Build a center of excellence. Standardize on three or four families across the product line. Negotiate volume contracts with two suppliers per family for redundancy.
Above 50,000 units per year. Vertically integrate. The unit economics no longer support a margin stack with two suppliers and a distributor in the middle. Tesla, Figure, BYD, and Apptronik are all in this band or projecting to be.

The trap is vertically integrating before you know your spec is stable. Volume below threshold plus an unstable spec is the worst combination, you commit capex against a moving target.

Risk register

The five risks that should sit on every actuator program’s top sheet.

Single-source supply. One supplier for a critical SKU, especially harmonic drives or specialty motors out of Japan and Germany. Mitigation: qualify a second source on every SKU above the BOM 80th percentile by cost.
Geopolitical exposure. Chinese actuator supply remains the cheapest and the deepest. Tariff or export-control disruption resets unit economics by 20 to 40 percent overnight. Mitigation: scenario-plan a domestic-only stack at 1.5x cost.
Specification creep. A 10 percent change in load spec late in design can force a complete actuator family change. Mitigation: lock specs at gate 2, and burden in 25 percent margin on force and 40 percent margin on cycle life.
Reliability tail. Mean-time-between-failure on a brushless DC motor in factory air is 50,000 hours. In an outdoor humanoid working a construction site, it is 8,000 hours. Mitigation: instrument every actuator with current and temperature telemetry from day one. Build the predictive maintenance loop before you ship.
Talent concentration. The number of mechatronics engineers who can lead an actuator program at depth is small. Three-digit globally. Mitigation: build internal apprenticeship loops. The seniors will leave eventually. The system has to outlast them.

Build, buy, partner decision tree

The decision tree, simplified.

If volume is below 1,000 per year, buy.
If volume is 1,000 to 50,000, partner. Two suppliers per family. Co-engineer, but do not build.
If volume is above 50,000 and the spec is stable, build. Vertically integrate the top three SKUs by cost.
If volume is above 50,000 and the spec is unstable, buy aggressively for one more cycle, then revisit.
In all cases, internalize specification and integration. Never outsource those.

What this playbook is for