Robot 101 · Chapter 09

The robotics market landscape: who builds what, and where

Five segments, five different games

"The robotics market" is a misleading singular. It is more useful to think of roughly five overlapping segments, each at a different maturity stage with different economics. Industrial arms — six-axis manipulators bolted to a factory floor — are the oldest and most mature segment: decades of installed base, established service networks, and slow-moving competitive dynamics. Collaborative robots (cobots) are a lighter, safety-rated variant designed to share a workspace with people rather than operate behind a cage. Mobile robots — automated guided vehicles and autonomous mobile robots that move materials around a warehouse or hospital — are a faster-growing, more fragmented segment. Service robots are built for a specific non-industrial task, such as delivery, cleaning, or inspection. And humanoid robots are the newest and least commercially proven category: general-purpose, bipedal, still mostly in pilot and demonstration phases rather than volume production.

Many industry estimates suggest the overall market is growing at a strong compound rate through the rest of the decade, but that headline figure obscures more than it reveals. The industrial-arm segment tends to grow more slowly and steadily; mobile and service robotics are generally reported to grow faster; and the humanoid segment, while commercially unproven, has attracted a large share of investor capital relative to its current revenue. Understanding which segment a given company or product sits in matters far more than any aggregate number, because the competitive dynamics, capital requirements, and time-to-revenue are different in each.

The humanoid investment wave

The mid-2020s have seen an unusually concentrated wave of investment directed at humanoid robotics — general-purpose, human-scale machines intended eventually to perform a broad range of physical tasks. Industry observers generally attribute this wave to the simultaneous maturing of several underlying technologies covered elsewhere in this handbook: higher-density batteries, more precise and backdrivable actuators, and transformer-based AI models capable of turning a camera feed and a natural-language instruction into robot actions or control targets. Several labs have published demonstrations of a robot completing a non-trivial manipulation task in an unstructured environment without task-specific pre-programming, and those demonstrations have done much of the work of attracting further capital.

It is worth being precise about what this wave does and does not mean. It reflects a bet — widely but not universally shared among investors and researchers — that a general-purpose dexterous robot capable of a large share of routine human physical labour is a multi-year engineering problem rather than a multi-decade one. It is not a guarantee that any particular company will reach that outcome, and it says nothing about near-term profitability for the companies involved, most of which generate little to no revenue today. This handbook treats the wave as an industry phenomenon worth understanding, not as a signal to act on.

Established players: the industrial-arm incumbents

The industrial-arm segment is dominated by a small number of long-established manufacturers — companies such as FANUC, Yaskawa, ABB, KUKA, and Mitsubishi Electric — that between them account for the overwhelming majority of arms shipped into factories worldwide each year. Their advantage is not primarily technical: it is decades of installed base, service and spare-parts networks in essentially every industrial region, and integration relationships with factory system integrators that took decades to build. This is a difficult segment for a new entrant to compete in directly, and most new robotics companies — including the humanoid and mobile-robot players described below — build in adjacent segments rather than attempt to displace these incumbents on their own ground.

The Chinese humanoid and component ecosystem

A cluster of Chinese companies has emerged around humanoid and legged robots at price points substantially below many Western peers, alongside a deep domestic supply base for the components those robots use. Unitree Robotics, based in Shenzhen, has become known for quadruped and humanoid platforms priced well below comparable Western machines, with a partially open software stack that has attracted an active developer community. Fourier Intelligence combines a rehabilitation-exoskeleton business — which has commercial rehabilitation products already in clinical use — with humanoid development, giving it a more compliant, clinically oriented joint design than most peers. AgiBot and UBTECH are both pursuing manufacturing and logistics deployments, with UBTECH in particular citing extended real-world operating hours in factory and utility settings rather than demonstration videos alone. Collectively, these companies sit inside — and depend on — the same Pearl River Delta component base described below.

Western AI-robotics labs and the software-first bet

A separate cluster of Western companies is betting that the hard problem in robotics is now software rather than hardware, and is pursuing correspondingly different business models. Figure AI builds its own humanoid hardware alongside a large-language-model-based perception and language-understanding stack, and has pursued manufacturing and logistics partnerships. Physical Intelligence builds no robot of its own; it trains a general manipulation policy intended to be licensed across other companies' hardware, in roughly the way a chip or operating-system vendor licenses its product to many device makers. Skild AI is pursuing a similar general-purpose "robot brain" strategy. Boston Dynamics, a longer-established player, and Agility Robotics have each demonstrated humanoid or near-humanoid platforms performing structured tasks in real commercial environments such as warehouses, which is a meaningfully different proof point than a research demonstration.

Why the China–West supply chain is splitting

The robotics industry is increasingly organised around two only partially overlapping technology stacks. A Western-centred stack is generally built on ROS 2 for robot software, PyTorch for AI model training, and NVIDIA hardware for inference. A China-centred stack draws more on domestic frameworks and domestic AI-accelerator hardware, alongside a supply chain that runs from battery-cell chemistry through to final robot assembly largely within China. Export controls on advanced AI training chips have accelerated this separation by pushing some Chinese developers toward downgraded or domestic hardware alternatives. This is a factual, structural description of how the industry is organising itself, not a prediction about who will benefit — both stacks continue to develop, and many companies deliberately work across both rather than choosing one exclusively.

Regardless of which stack a given robot's software runs on, its physical components increasingly originate from the same place: the Pearl River Delta, anchored by Shenzhen. That region has built manufacturing density for motors, battery packs, sensors, and precision mechanical parts that few other regions can match on cost and lead time, and in many categories buyers report a narrowing gap between component quality from Chinese suppliers and longer-established Japanese and Western suppliers over the past decade. A robot company anywhere in the world — building on any software stack — is increasingly likely to source at least some of its hardware from this ecosystem, even if final assembly and branding happen elsewhere.

China's national robotics policy signal

China's Ministry of Industry and Information Technology (MIIT) published guidance in late 2023 setting out national targets for humanoid robotics, including domestic development of a humanoid operating system and a target for domestic self-sufficiency in key components such as actuators, sensors, and AI chips. Provincial and municipal governments, including Shenzhen's, have followed with their own supporting measures — grants for embodied-AI hardware development, procurement subsidies aimed at public-sector robot adoption, and designated industrial parks for robotics manufacturers. Described factually, this is a coordinated, multi-level policy effort to build a self-sufficient domestic robotics supply chain and to accelerate public-sector adoption of care, industrial, and logistics robots — one input, among several, into why component availability and pricing out of the region have moved the way they have.