Robotics is becoming a strategic national capability while emerging manufacturers still need international customers, components and capital to make their machines commercially viable.
By Tim Harper. Published and researched to 9 November 2026.

Summary
US restrictions on new foreign-produced mobile robots and China’s prohibition on SoftBank acquiring ABB’s Chinese robotics business expose different forms of control over the industry. Neither is a universal ban on foreign automation. Together with industrial policy and prospective trade remedies, they make manufacturing location, ownership and operational data increasingly important to robotics economics.
The United States has restricted equipment authorisation for defined categories of foreign-produced robots. China has intervened in a Japanese investor’s acquisition of a Swiss industrial robotics business. These developments bring the geopolitics of robotics directly into factory investment decisions, extending political scrutiny from advanced chips and AI models to the machines that use them.
The timing matters. Established automation suppliers depend on international production and customers; emerging physical AI companies, developing machines that perceive and act in the real world, need those networks to reach commercial scale. Governments want dependable national capability, but dividing the market can make that capability more expensive to build.
What the US restrictions actually cover
On 28 July 2026, the Federal Communications Commission added foreign-produced advanced robotic devices to its Covered List. Public Notice DA 26-786 implements a 27 July interagency national-security determination under section 2 of the Secure and Trusted Communications Networks Act. Existing FCC rules prohibit new equipment authorisations for covered equipment, generally preventing new covered models from being imported, marketed or sold.
The definition covers ground-mobile machines operating at a distance from a human operator, weighing more than 4.4 pounds including any docking station, with environmental sensing, connectivity of at least 200 kilobits per second, and specified control software. It captures humanoids, quadrupeds and qualifying mobile industrial robots. Fixed industrial arms are excluded, alongside defined vehicles, aircraft, underwater vehicles and medical devices. This is not a prohibition on all foreign industrial automation.
The production test is also broader than Chinese ownership. A robot is foreign-produced if it fails the Buy American “domestic end product” standard. That normally requires US manufacture and the applicable domestic-component content, with exceptions including the component-test waiver for qualifying commercially available off-the-shelf products. Morrison Foerster’s analysis explains why American design or intellectual property alone does not settle eligibility.
Previously authorised models can continue to be imported, sold and used. The companion FCC waiver permits qualifying software and firmware updates through at least 1 January 2029; it is not blanket permission for new hardware or product generations. Conditional approvals provide another route: September’s ANSCER exemptions cover three named platforms, subject to continuing compliance with an onshoring plan and updated vetting. Exemption from the list does not replace normal equipment authorisation.
The government’s stated rationale includes supply-chain dependence, surveillance, data theft and remote interference with physical operations. These are plausible threat mechanisms, but country of manufacture does not establish that an individual robot contains a vulnerability.
A separate measure reaches further into industrial machinery. Proclamation 11077, signed on 28 October and published on 2 November, follows a completed Commerce Department investigation under section 232 of the Trade Expansion Act. It accepts the national-security finding and directs trade negotiations, with a progress update within 180 days. Tariffs, quotas, fees and production incentives are possible subsequent remedies. The proclamation itself imposes no new robotics tariff. Importers therefore face an authorisation barrier for certain products and prospective trade-cost exposure across a wider category, not two identical bans.
China’s intervention reaches established automation
ABB’s 8 October 2025 agreement valued its Robotics division at $5.375 billion on an enterprise-value basis. The sale to SoftBank replaced a planned stock-market spin-off and remained conditional on approvals. ABB’s separate Machine Automation division was outside the disposal. Robotics generated $2.3 billion of revenue in 2024: this is an established industrial business, not a speculative humanoid startup.
In its 6 November disclosure, SoftBank confirmed receiving notification on 4 November from China’s Office of the Working Mechanism for Foreign Investment Security Review prohibiting acquisition of ABB’s robotics business in China. It also confirmed continuing dialogue with ABB. The disclosed prohibition concerns Chinese operations; the statement does not announce termination of the entire transaction.
China is material to the deal’s industrial logic. ABB opened a $150 million Shanghai production and research facility in December 2022, alongside manufacturing in Sweden and Michigan. By July 2025, ABB said more than 90% of its robots for Chinese customers were manufactured locally. Local manufacturing has not removed the ownership question.
Reuters reported on 5 November, citing people familiar with the matter, concerns about sensitive data reaching the US, SoftBank’s relationships with Western companies including OpenAI, and Japanese control of technology. These are reported concerns, not reasons published in SoftBank’s disclosure or an accessible Chinese decision. SoftBank’s AI investments make the connection commercially intelligible; they do not prove the regulatory motive.
The wider acquisition risk is that a buyer may be acceptable in one jurisdiction but unable to acquire an essential operation elsewhere. Separating that operation could require agreements covering software, intellectual property, customers and research. Fewer feasible buyers and greater execution risk can weaken competitive bidding and valuations, although this transaction does not quantify an industry-wide discount. For venture-backed developers, a narrower acquisition market could also reduce exit options and make investors less willing to finance the next manufacturing expansion.
Hyundai and Boston Dynamics: several meanings of nationality
Hyundai’s acquisition illustrates the benefits of international integration. The transaction completed on 21 June 2021, valuing Boston Dynamics at $1.1 billion, with Hyundai Motor Group holding 80% and SoftBank retaining 20%. Those are the original ownership percentages, not a reliable description of 2026 holdings.
On 16 July 2026, Hyundai confirmed that shareholders were pursuing SoftBank’s entire remaining interest after it exercised its contractual put option, a right to require a purchase. Yonhap reported that the remaining interest was 9.65%. The primary announcement described a contemplated transaction; it did not confirm completion.
Hyundai’s January 2026 strategy envisages Atlas performing sequencing tasks at its Georgia manufacturing operation by 2028 and component assembly by 2030. Its target of capacity for 30,000 robots annually by 2028 is an announced manufacturing ambition, not current output or deployment.
Korean capital, American robotics engineering and automotive production in the US can reinforce each other. Corporate ownership, engineering location, manufacturing origin, legal ownership of intellectual property and control over factory data are nevertheless separate attributes. A shareholder does not automatically receive unrestricted rights to every customer’s data. Nothing in these announcements establishes an ABB-style prohibition against Hyundai or Boston Dynamics.

Who controls the robot’s data?
A robot’s sensors can reveal facility layouts, equipment geometry, workplace activity and production flows. Controller logs can expose cycle times, process parameters, faults and utilisation. In combination, those records may disclose how a factory achieves its yield or where its bottlenecks lie. The sensitivity depends on what is captured, retained and accessible. As the tactile-sensing guide explains, contact measurements help robots adjust physical actions; the same records can also reveal details of a customer’s production task.
Four operating arrangements need separating. A conventional arm may execute programmes on a local controller without exporting production data. A connected robot may send selected diagnostic records to a service provider. A cloud-dependent system may require external computing for particular functions. A learning system may collect images, trajectories and force measurements for subsequent model training. One product can combine these arrangements, but connectivity alone does not prove continuous overseas transmission.
Physical AI adds a valuable secondary use for operational information. Examples of successful and failed actions can improve models deployed elsewhere. The Open X-Embodiment research collaboration demonstrates the value of combining robot experience across platforms. Commercial deployment raises a further contractual question: does permission to service a machine also permit using its experience to train a supplier’s general model?
Customers need explicit answers about data ownership and licences, storage jurisdictions, retention, subcontractors and training rights. Remote administration and software updates deserve equal attention: an account authorised to maintain a robot may also provide a route to change its behaviour. Compromise can interrupt production or create physical hazards even without stealing data.
NIST’s operational-technology security guidance supports restricted remote access, authentication and monitoring. Local processing, network separation, controlled update deployment and auditable access can reduce exposure. Suppliers able to demonstrate these controls could win customers whose procurement teams need evidence of operational independence. Local hosting alone is insufficient if an overseas administrator retains unrestricted access.
Industrial policy meets an international supply chain
China combines policy support with a large production ecosystem. Its humanoid roadmap set goals for an innovation system by 2025 and a secure industrial supply chain by 2027; these were government targets. The International Federation of Robotics’ September 2026 figures measure realised industrial deployment: China installed 354,000 robots in 2025, 59% of the global total. Domestic suppliers held 55% of that market, leaving substantial business for foreign manufacturers.
Humanoid production claims need different treatment. A Shanghai government report in July relayed Agibot’s production milestone of 15,000 machines in June. Production does not establish paid utilisation, customer savings or profitable exports. International sales ambitions must pass those commercial tests as well as market-access rules. Foreign technology still participates: NVIDIA’s 2026 reference humanoid design combines a Unitree body with NVIDIA computing and models. Chinese hardware capability does not imply independence across the complete technology stack.
Policy can fund research and initial purchases; nearby motor, electronics and precision-component suppliers shorten engineering iterations. Those mechanisms reinforce each other, making China’s position difficult to attribute to subsidies alone. Replicating final assembly does not immediately reproduce the supplier relationships behind it.
Other regions are building overlapping frameworks. Japan’s June 2026 AI Robotics Strategy targets roughly ten million robots by 2040 and supports domestic AI models. South Korea’s 2024–2028 intelligent-robot plan includes 200 billion won for a national robot test field. These are capability-building commitments, not completed fleets.
Europe combines foreign-investment screening with product regulation. The Cyber Resilience Act began applying reporting obligations in September 2026; its main requirements apply from December 2027. The AI Act’s treatment of high-risk systems depends on intended use, with rules for relevant systems embedded in products scheduled for August 2028. These approaches impose different tests from a manufacturing-origin restriction, encouraging regional product configurations rather than a single global compliance assumption.
The commercial cost of political intervention
A domestic robotics supply chain requires more than a robot factory. Actuators combine motors, transmissions and feedback sensors; controllers depend on semiconductors and software; mobile machines add batteries and navigation hardware. Precision bearings, gearboxes and safety components require dependable production and qualification. Replacing a supplier can trigger mechanical redesign, control retuning and reliability testing. Export restrictions affecting chips or materials can propagate through those layers without directly naming robots.
The tension is already visible in semiconductor manufacturing. On 22 October the FCC granted Intel a one-year waiver for specified changes to particular automated guided vehicles. The order records Intel’s position that proven domestic alternatives were unavailable and replacement would take years. The relief is tightly limited, but the example shows how restrictions intended to strengthen national capability can complicate another strategic industry’s operations.
Regional production can reduce freight, improve service and protect access to customers. ABB’s existing regional factories demonstrate that localisation can be commercially sensible. However, duplicating tooling, inventories and engineering teams divides fixed expenditure across smaller production runs. Different approved components, software permissions and data arrangements for American, Chinese and European customers add continuing support costs.
For industrial customers, any additional tariff would affect the relevant imported value, not automatically the entire installed project. Integration, maintenance, energy and downtime remain part of the investment calculation. Regulatory delays can also postpone savings or leave a customer dependent on a model whose replacement has not been authorised. Investors must therefore assess access to markets and follow-on capital alongside technical performance.
There are opportunities on the other side. Procurement commitments, test facilities and investment incentives can support local component suppliers, integrators and secure software providers. A credible initial customer can make manufacturing finance possible. The policy test is whether that support produces reliable, competitive machines whose value survives beyond protected demand.
Humanoid economics acquire another variable
The existing analysis of humanoid robotics economics examines manufacturing cost, reliability, maintenance, energy, utilisation and integration through the cost of useful work. Geopolitical restrictions affect both sides of that calculation: compliance and duplicated supply chains can raise expenditure, while reduced market access limits the sales over which development costs are recovered. Industrial policy can supply capital and deployment opportunities, but a subsidised machine still has to perform its task.
This is the central tension in robotics commercialisation. Governments are recognising strategic value before much of the emerging industry has demonstrated durable commercial returns. Security requirements that reward demonstrable control of data and dependable local support can help the industry mature. Requirements that divide production and customers without addressing the actual vulnerability risk making useful automation harder to afford. For manufacturers and investors, commercialisation strategy now needs to account for both.
Principal references
Research cut-off: 9 November 2026. Additional company, technical and policy sources are linked at the relevant claims above. Production targets are distinguished from measured installations; economic consequences are the author’s analysis.
- FCC listing and national-security determination — 28 July 2026.
- FCC software and firmware waiver — 28 July 2026.
- FCC ANSCER conditional approvals — 18 September 2026.
- FCC Intel maintenance waiver — 22 October 2026.
- US Proclamation 11077 — 28 October 2026; published 2 November.
- ABB sale agreement — 8 October 2025.
- SoftBank transaction update — 6 November 2026.
- Reuters on China’s intervention — 5 November 2026.
- Hyundai statement on Boston Dynamics — 16 July 2026.
- IFR World Robotics 2026 results — 24 September 2026.

