Artificial intelligence (AI) and robotics are converging at speed, turning once-specialist machines into increasingly adaptable, perception-rich physical agents. From factory floors to fulfilment centres, from construction sites to hospitals and homes, the new wave of intelligent robots aims to move, see, grasp, plan and learn—often through the same foundation-model approach that transformed language and vision. This report examines the latest technical breakthroughs, maps near-term applications in manufacturing, logistics, autonomous drones and personal assistance, weighs the societal implications of ever-smarter robots, and assesses the intensifying geopolitical race—particularly between the United States and China. We close with a balanced debate on whether “robot taxes” on humanoids and other embodied AI could or should help cushion labour disruption, and the trade-offs such policies entail.
Breakthroughs at the AI–Robotics Frontier: Foundation Models for Machines
In March 2024, NVIDIA unveiled Project GR00T, a general-purpose foundation model for humanoid robots, alongside new hardware (Jetson Thor) and upgrades to its Isaac robotics stack—aimed squarely at teaching bipedal robots dexterous, generalisable skills via simulation, video and teleoperation data [1]. In March 2025, the company announced Isaac GR00T N1, an open humanoid foundation model and simulation frameworks to speed robot development; the release sharpened the industry’s focus on “generalist” manipulation and locomotion across platforms [2]. Complementing this, Boston Dynamics retired its hydraulic Atlas and introduced a fully-electric Atlas designed for “real-world applications”—a notable pivot from parkour demos to industrially relevant manipulation and mobility, supported by new control and hardware advances [3][4].
Start-ups are racing, too. U.S. humanoid firm Figure has raised substantial new capital and reoriented its software strategy toward in-house models, signalling the strategic primacy of vertically integrated “embodied AI” stacks as robots graduate from staged demos to pilot deployments [5][6]. Apptronik, backed this year with a $350 million raise and a Mercedes-Benz pilot, underscores how automakers are testing humanoids for part-handling, quality checks and line-side assistance—focusing on repetitive, ergonomically risky tasks where labour is tight and task variation is frequent [7][8]. Tesla continues to publicise timelines for Optimus—in-factory use first, with external sales to follow—though cost, data scale and reliable autonomy in cluttered, dynamic spaces remain central hurdles for the entire field [9][10].
Autonomous Drones: From Pilots to Scaled Services
Parallel progress is unfolding in aerial robotics. U.S. regulators have steadily expanded approvals for “beyond visual line of sight” (BVLOS) operations—crucial for scaled logistics—backed by the FAA’s certification pathways and growing operational data. The FAA details how package-delivery operators must certify under Part 135 and secure appropriate waivers, while an Inspector General report shows BVLOS approvals rising from 1,229 (2020) to 26,870 (2023)—a regulatory and operational inflection that enables routine city-scale services [11][12]. Alphabet’s Wing and other operators have urged durable BVLOS rules to cement reliability and safety at scale, aligning industry roadmaps with national airspace integration [13]. Zipline, meanwhile, has expanded medical and retail delivery with FAA-approved operations and city partnerships, illustrating how autonomy, routing and lightweight airframes can reduce delivery emissions and congestion while meeting tight service-level windows [14]. On the enterprise side, Skydio’s X10 platform exemplifies high-autonomy inspection drones for first responders and infrastructure owners, pairing upgraded compute and sensors with robust obstacle-avoidance to reduce mission risk and pilot workload [15][16].
Manufacturing & Logistics: Where AI-Robots Go to Work First
Manufacturing and warehousing are the near-term beachheads for humanoids and other embodied AI because facilities are labour-constrained, task variety is high, and safety/ergonomics matter. Amazon has showcased new robotics systems and trials of Agility Robotics’ bipedal Digit for tote handling and back-of-house moves—jobs that are arduous for people yet simple enough to pilot with constraints and human oversight [17]. Industry trackers note Digit’s early commercial availability and pilots across multiple logistics sites, while more capable generations target grasping, placement and mobile manipulation in mixed human–robot workflows [18]. On the plant floor, pilots with Apptronik and others emphasise “co-present” operations: robots manage repetitive fetch-and-place or inspection loops; humans focus on exception handling, quality judgement, changeovers, and continuous improvement [8].
Personal Assistance & Service Use-Cases: From Homes to Hospitals
Service robots and eventual home assistants remain a major prize. Fully general household robots remain years away, but the trajectory is clear: foundation-model reasoning, video-learned policies, and compliant actuation are converging to enable object search, tidying, item retrieval and assisted mobility. Policymakers and analysts foresee humanoids and legged robots increasingly present in public spaces, healthcare and hospitality during the 2030s, provided safety envelopes, force limiting, and trustworthy behaviour are proven beyond staged demos [19].
The Global Race: Who’s Ahead—United States or China?
On industrial adoption and supply chains, China currently leads. According to the International Federation of Robotics’ 2025 World Robotics release, factories installed 542,000 robots in 2024—74% in Asia—while China alone accounted for roughly 295,000 new units (54% of worldwide deployments) and surpassed the 2 million mark in installed base; domestic suppliers also captured a majority of China’s home market for the first time [20][21]. Policy support is explicit: China’s Ministry of Industry and Information Technology (MIIT) outlined a 2024 humanoid roadmap and mass-production ambitions by mid-decade, with provincial pilots and full-stack incentives [22]. At the product level, Chinese vendors are pushing price points down: Unitree’s 2025 R1 humanoid launched at a steep discount to its 2024 model, signalling fast cost curves and potential for scaled deployments in education, R&D and light-duty work [23]. Reporting and commentary also highlight the state-directed capital and export-led logic behind China’s surge into embodied AI, with a clear geopolitical framing around technological self-reliance and labour-force demographics [24][25].
The United States, by contrast, retains a strong edge in core AI platforms (foundation models, silicon, toolchains) and high-end mechatronics via firms such as NVIDIA, Boston Dynamics, and a dense start-up ecosystem (Figure, Apptronik, 1X and others). With NVIDIA’s GR00T/Isaac releases, many U.S. and allied companies can train “generalist” robot policies in simulation and transfer them to hardware faster, potentially narrowing any mechatronics gap with data and software leverage [2]. Hardware availability, unit economics and access to large-scale, diverse human-demonstration datasets will determine how quickly pilots turn into meaningful deployments. On balance: China is ahead today in factory robot installations, manufacturing scale and cost; the U.S. is competitive in embodied AI software, chips and platform middleware—advantages that could compound if open(ish) robotics stacks standardise around U.S. technologies. The “winner” is likely to be domain-specific over the next five years: China in factory automation volume; the U.S. in general-purpose embodied AI platforms and high-complexity pilots.
Who Else Is in the Race?
Beyond the U.S. and China, Europe, Japan and Korea remain pivotal. Europe brings safety standards, collaborative robot expertise and advanced industrial niches; Japan and Korea remain leaders in robot density and mechatronics, with extensive automotive and electronics ecosystems. Multinationals like Mercedes-Benz piloting U.S. humanoids in EU plants, and Japanese and Korean vendors partnering with U.S. AI stacks, suggest the future will be a networked—rather than purely national—supply chain, albeit under tightening export controls and standards regimes [8].
Societal Implications: Safety, Jobs, and Public Trust
As robots gain mobility and judgment, safety cases move from the lab to the regulator and insurer. Force-limiting, geofencing, compliant actuation and operational design domains (ODDs) will be mandated for many public deployments; aerial systems must satisfy evolving BVLOS rules, detect-and-avoid requirements and community expectations on noise and privacy [11]. In labour markets, the likely near-term pattern is “task remix” rather than full job elimination: robots handle repetitive lifts, inspections and fetch-moves; humans escalate exceptions, supervise fleets, manage changeovers and deliver judgment-heavy work. This shift can improve safety and productivity, but it also risks displacing certain roles (especially entry-level material handling) unless reskilling and internal mobility keep pace. Analysts anticipate broader social adoption in the 2030s if early deployments prove reliable, affordable and visibly beneficial to workers and customers [19].
Applications Snapshot: What’s Working Now—and Next
Manufacturing. Repetitive part moves, machine tending, in-line inspection and kitting are early targets for humanoids and mobile manipulators, especially where legacy cells cannot be economically retooled. Automakers are piloting in low-rate, low-risk zones before scaling [8]. Logistics. Bulk tote transfer, returns handling, trailer unloading and depalletising are promising beachheads, with Amazon’s robotics updates and Digit pilots emblematic of a phased approach to mobile manipulation in people-designed spaces [17][18]. Aerial logistics. BVLOS approvals and city partnerships are turning drones into practical delivery and inspection tools, reducing road congestion and improving time-to-X for urgent goods [12][13]. Personal assistance. Early roles will be supervised: fetch-and-carry, tidying, mobility support, and telepresence—progress that depends on reliable perception, safe contact, and robust failure handling [19].
Why “Generalist Robotics” Matters
Platform-level progress—foundation models trained on simulation plus human demonstration—could let robots perform thousands of tasks with little per-task programming. NVIDIA’s open Isaac GR00T N1 and associated sim/data pipelines are intended to catalyse an ecosystem in which policies generalise across arms, hands and bodies, shrinking costs and time-to-deployment. If successful, that favours countries and firms with AI compute, data assets and developer communities—capabilities in which the U.S. currently leads [2].
Who’s Likely to “Win” Over the Next Five Years?
On present evidence: Manufacturing volume and cost—advantage China, thanks to scale, policy, and fast component supply chains [20][21]. Generalist AI and developer platform gravity—advantage U.S., with chips, models and middleware that may become the “Android layer” for embodied intelligence [1][4]. End-user value will accrue to regions that convert pilots into sustained productivity growth with worker reskilling and safety proof points; that is as much policy execution as engineering.
Should We Tax Humanoids? The Robot-Tax Debate
As humanoids and other embodied AI scale, some propose taxing robots (or the firms deploying them) to slow displacement and fund reskilling. Bill Gates popularised the idea: direct levies on automation could finance care work, education and transition support, aligning fiscal receipts with the shift from human to machine labour [26]. The European Parliament weighed a robot tax during its 2016–17 robotics deliberations but rejected the measure, citing definition challenges and innovation risks; instead it emphasised liability, ethics and transparency frameworks—an early signal that line-item robot levies are politically fraught [27]. Legal scholarship continues to explore designs (e.g., payroll-tax parity, accelerated depreciation caps, gain-sharing on productivity) but stresses that clear definitions of a “robotic substitution event” and avoidance of innovation chilling are preconditions to workable rules [28].
Pros. A robot tax could (i) cushion displacement by funding reskilling and mobility; (ii) preserve public-finance stability as wage-based tax bases shrink; and (iii) slow harmful, low-road automation where externalities (injury, inequality) are not priced. Cons. It risks (i) deterring productivity-enhancing investment, dulling competitiveness; (ii) pushing automation—and jobs—offshore to lower-tax jurisdictions; (iii) creating administrative complexity and loopholes (what counts as a “robot”?). Several countries have instead pursued indirect approaches: for example, South Korea limited certain tax credits for automation rather than imposing explicit robot levies, an approach sometimes (informally) dubbed a “robot tax” because it narrows the pro-automation subsidy rather than creating a new tax head [29]. Broader OECD work on tax and digitalisation, while not endorsing robot-specific taxes, documents how administrations are digitising and using AI to stabilise revenues through compliance improvements—arguably a more practical lever in the near term [30][31].
Policy Priorities: Compete, Deploy, Protect
Governments weighing national advantage and social stability face a common playbook: (1) invest in embodied-AI R&D, compute and standards; (2) derisk first wave deployments through safety certification, sandbox regimes and insurance frameworks; (3) scale reskilling and job-matching at pace, with special focus on entry-level logistics and manufacturing roles most exposed to mobile manipulation; (4) modernise competition and trade tools to ensure resilient supply chains for actuators, sensors and robot-grade compute; (5) measure impacts with task-level data to separate disruption myths from realities. Countries that get this sequencing right—combining platform leadership with worker transitions—will capture the upside of embodied AI while keeping publics onside.
Our View
“AI inside” is making robots more useful, sooner, in more places. Over the next five years, the most visible wins will be in logistics and manufacturing tasks that are dull, dirty, dangerous or simply repetitive; aerial logistics will continue to scale under BVLOS frameworks; home assistance will progress, but carefully, under tight safety envelopes. China’s adoption-and-manufacturing machine gives it an early lead in volumes and cost; the U.S. enjoys deep moats in chips, models and developer ecosystems that may define the software “operating systems” of embodied AI. Rather than a binary “winner,” expect a divided map: China leads in volume; the U.S. leads in platforms and high-complexity deployments; allies plug critical capability gaps. On taxation, we see little case for blunt robot levies today: better to (a) remove distortionary subsidies that favour low-road automation, (b) expand worker transition finance through broad-base instruments, and (c) double down on reskilling and mobility so augmentation > displacement in practice. If, in later years, humanoids substitute wage labour at scale, narrowly tailored fiscal instruments may be revisited—anchored in evidence, clear definitions and international coordination.
Summary: Foundation models, better actuation and richer data are pushing AI-robotics from demos to deployments. Near-term value clusters in factories, warehouses and BVLOS aerial logistics; personal assistance will follow with strong safeguards. China leads in industrial adoption and cost; the U.S. leads in embodied-AI platforms. Societies should prioritise safety, standards and worker transitions over blunt “robot taxes” for now—while keeping the fiscal toolbox open if substitution accelerates. The countries that align platform leadership with human capital will set the pace in the age of intelligent machines.
[1] NVIDIA — Foundation model & Isaac robotics platform — link
[2] NVIDIA — Isaac GR00T N1 open humanoid foundation model — link
[3] Boston Dynamics — Electric Atlas: a new era — link
[4] Boston Dynamics — Atlas overview — link
[5] Reuters — Figure valued at $3.9B funding round — link
[6] TechCrunch — Figure drops OpenAI for in-house models — link
[7] Axios — Apptronik raises $350M; humanoid robots — link
[8] Reuters — Mercedes-Benz takes stake in Apptronik; factory tests — link
[9] Reuters — Tesla could sell Optimus robots by end of next year — link
[10] Reuters — Tesla to use humanoid robots internally next year — link
[11] FAA — Package delivery drones & advanced operations (Part 135) — link
[12] U.S. DOT OIG — BVLOS Drone Operations Final Report (2025) — link
[13] Wing (Alphabet) — Ensuring FAA enables BVLOS progress — link
[14] Zipline — City partnerships & operations — link
[15] Skydio — X10 platform — link
[16] Skydio — Introducing X10 — link
[17] Amazon — New robotics solutions — link
[18] Contrary Research — Agility Robotics company profile — link
[19] World Economic Forum — Humanoid robots: disruption & promise — link
[20] IFR — Global robot demand in factories doubles over 10 years — link
[21] IFR — China robotics press release (2025-09-25) — link
[22] McKinsey — Humanoid robots: crossing the chasm — link
[23] Reuters — China’s Unitree prices new humanoid at deep discount — link
[24] South China Morning Post — How China is supercharging humanoid robots — link
[25] ISEAS — AI-robotics revolution, China–US rivalry & SE Asia — link
[26] World Economic Forum — Bill Gates on taxing robots — link
[27] Reuters — European Parliament rejects robot tax, calls for robot law — link
[28] IBFD — Robot taxation: a normative tax-policy analysis — link
[29] Ius Laboris — Is it time to tax the robots? — link
[30] OECD — AI in tax administration (report) — link
[31] OECD — Tech-enabled tax administration of the future (blog) — link
beFirstComment