V
Force-torque, tactile & vision sensing
Manipulation lives or dies on force and touch sensing; the merchant suppliers are few and mostly buried inside diversified names.
Manipulation is the gate every humanoid program must pass, and it does not open without force, torque, and touch. A leg can balance on inertial sensing and joint encoders, but a hand that closes on an egg, threads a connector, or hands you a cup needs to feel — six-axis force/torque at the wrist, joint-torque feedback in every actuator, and tactile sensing in the fingertips. This is the least-commoditized layer of the robot bill-of-materials and, perversely for public-market investors, the hardest to own cleanly.
Who owns it
The purest US-listed merchant exposure is buried inside a diversified name. ATI Industrial Automation — the reference six-axis force/torque sensor brand in robotics, with families from the Nano17 to the Omega331 — is a unit of Novanta (NOVT), acquired in 2021 for $172M when ATI was doing a bit over $70M of revenue. It now sits in Novanta's Automation Enabling Technologies segment alongside laser steering and end-of-arm tooling. TE Connectivity (TEL) sells a merchant menu of torque, force, and position sensors for cobot joints, including a ring-shaped safety torque sensor rated to ISO 13849 Category 4. On the proprioception side, Allegro MicroSystems (ALGM) supplies the magnetic/inductive position sensors and motor drivers that close the loop inside each actuator. Vision — the other half of perception — is dominated by Cognex (CGNX), and edge sensing/MEMS by STMicroelectronics (STM), which has built a "Physical AI" sensing narrative and is absorbing NXP's MEMS business.
What breaks it
Three structural facts make this a genuine chokepoint. First, the merchant base is thin and fragmented — outside ATI and TE, the recognized force/torque names (Bota Systems, Robotous, FUTEK) are private and several are non-US. Second, fingertip tactile sensing has no mature merchant standard; it is a cluster of startups and in-house programs with no durability track record. Third, the strongest demand nodes are integrating the component away.
What forces the reprice
The re-rate is volume. If humanoid unit counts inflect, the wrist sensor, the joint-torque rings, and the position ICs scale with degrees of freedom — and DoF is climbing (Tesla's Optimus Gen 3 hand reportedly carries 50 actuators across both hands). Allegro has put numbers on the slope: it sees semiconductor content rising from ~$5 in a home robot to ~$55 in a cobot to ~$150 in a humanoid. The catalysts are program ramps, supplier disclosures at investor days, and any merchant design-win announcement into a named humanoid line — which, today, essentially do not exist for US-listed names.
Be honest about size and earliness
This is the part the narrative skips. None of these companies has disclosed material humanoid revenue. ATI is real but small inside a Novanta that earns most of its money in surgical robotics and industrial precision motion. TE Connectivity is a $17.3B company whose growth is being driven by AI datacenter connectors, not robots; its headline cobot torque sensor was first announced in 2020. Cognex's 2025 revenue (~$994M) grew on logistics and consumer-electronics factory automation — its earnings commentary does not break out humanoid at all, making it the clearest robot-washing risk in this set. Allegro's $150/humanoid figure is an addressable-content estimate, and its content is position/current sensing, not force/torque or tactile.
Worse for the merchant thesis: the two leading Western humanoid programs are vertically integrating the very components this chokepoint describes. Tesla designs Optimus's actuators, torque sensors, and fingertip tactile sensors in-house. Figure AI, after finding merchant tactile sensors couldn't survive real-world use, built its own — Figure 03 fingertips now resolve forces down to 3 grams. When the highest-volume customers make the part themselves, the merchant force/torque/tactile story is capped to the long tail of programs that don't, plus the broad industrial/cobot base that was always there. Own this chokepoint for the optionality and the proprioception ICs, not for a humanoid revenue line that any of these names has actually printed.
Who owns the choke
Allegro MicroSystems, Inc.
Allegro markets a dedicated humanoid-joint portfolio (48V BLDC drivers, TMR/inductive position sensors, current sensors) and estimates semiconductor content rising from ~$5 in a home robot to ~$55 in a cobot to ~$150 in a humanoid; FY2026 revenue was ~$890M, ~70% automotive.
- Bull
- Allegro's dedicated humanoid-joint portfolio rides rising semiconductor content (~$5 home robot to ~$150 humanoid) as degrees of freedom and joint counts climb.
- Bear
- FY2026 revenue was ~$890M and ~70% automotive, so humanoid is a content estimate, not a printed line, and the auto cycle drives the model.
Cognex Corporation
Cognex's In-Sight/VisionPro/DataMan vision systems guide industrial and logistics robots; 2025 revenue was ~$994M (Q3 2025 +18% YoY) driven by logistics and consumer-electronics factory automation, with no humanoid revenue broken out in earnings commentary.
- Bull
- Cognex's In-Sight/VisionPro/DataMan systems dominate vision-guided robotics, scaling with logistics and electronics factory automation as the perception layer every robot fleet needs.
- Bear
- 2025 revenue (~$994M) grew on logistics and consumer-electronics automation with no humanoid broken out, making it the clearest robot-washing risk in sensing.
Novanta Inc.
Novanta acquired ATI Industrial Automation in 2021 for $172M (ATI then >$70M revenue); ATI's Nano17/Omega331-class six-axis force/torque sensors now sit in Novanta's Automation Enabling Technologies segment and are widely used in industrial, collaborative, and surgical robots.
- Bull
- Novanta owns ATI Industrial Automation, the reference six-axis force/torque sensor brand, scaling with humanoid degrees of freedom as the purest US-listed merchant force/torque exposure.
- Bear
- ATI is small inside a Novanta earning most of its money in surgical robotics and industrial motion, with no material humanoid revenue disclosed.
STMicroelectronics N.V.
STMicroelectronics hosted an 'Intelligent Sensing Enabling the Physical AI' investor call on 16 March 2026 led by its APMS group, and in February 2026 closed the acquisition of NXP's MEMS sensor business (up to $950M) to expand its sensing portfolio across industrial and consumer applications.
- Bull
- STMicroelectronics built a 'Physical AI' sensing narrative and closed the NXP MEMS acquisition (up to $950M), expanding the edge-sensing and MEMS portfolio that robots increasingly need.
- Bear
- The physical-AI sensing story is an investor-day narrative spanning industrial and consumer with no disclosed humanoid revenue, leaving robotics as speculative exposure.
TE Connectivity plc
TE Connectivity sells a merchant range of torque/force/position sensors for robots and cobots, including a ring-shaped safety torque sensor integrated at each joint and rated to ISO 13849 Category 4 PL e; FY2025 revenue was $17.26B with growth led by AI/datacenter, not robotics.
- Bull
- TE Connectivity sells a merchant range of torque/force/position sensors for cobot joints, including an ISO 13849 Category 4 safety torque ring, scaling with robot degrees of freedom.
- Bear
- It is a $17.3B company whose growth is AI/datacenter, not robotics; its headline cobot torque sensor was first announced back in 2020.
Tesla, Inc.
Tesla designs Optimus actuators in-house — removing the largest single buyer from the merchant frameless-motor order book — while outsourcing manufacturing to Chinese suppliers (Sanhua Tier-1 actuator assembly, Tuopu rotary actuators, Green Harmonic reducers). V3 Optimus production is targeted for summer 2026, with actuators ~56% of BOM cost.
- Bull
- Tesla designs Optimus actuators in-house with V3 production targeted for summer 2026, making it the highest-volume Western humanoid program and the bar suppliers price against.
- Bear
- It outsources manufacturing to Chinese suppliers, faces an unproven ramp, and humanoid mapping spans only actuators and sensing within a vast auto business.
Catalyst calendar
- 2026-07-29TE Connectivity fiscal Q3 2026 earnings (sales guided to ~$5B)mediumTests whether TE's torque/force/position robotics-sensor line is more than a rounding error against AI-datacenter-driven growth, and whether any robotics design win is disclosed.
- 2026-07-30Allegro MicroSystems fiscal Q1 2027 earnings (quarter ended ~June 26, 2026)mediumFirst print after Allegro framed ~$150 of semiconductor content per humanoid; watch for updated robotics design-win and content commentary that sizes the proprioceptive-sensing slope.
- 2026-08-05Cognex Q2 2026 earningsmediumWhether management quantifies any robotics/humanoid vision exposure (vs. logistics and factory automation) is the test of whether the humanoid tie is real revenue or robot-washing.
- 2026-10-19RoboBusiness 2026, Santa Clara (Oct 19-21), humanoid-for-industry tracklowPrimary venue where merchant force/torque/tactile suppliers and humanoid OEMs disclose design wins, making it the most likely source of a verifiable supply-position catalyst for this chokepoint.