Shenzhen’s robotics industry recorded RMB 242.6 billion in total output value in 2025, up 20.56% — a record. Industrial robot output reached 194,900 units, up 43.1%, equal to 25.2% of China’s national total; service robots reached 7.9665 million units, 42.9% of the national total. [1]
The line item that matters most for manufacturing is the newest one. Output in Shenzhen’s humanoid robot whole-machine category rose 83.1% year on year in 2025 — the steepest growth rate anywhere in the white paper. [2]
Growth at that rate changes the engineering question from “can we build one” to “can we build them repeatably, at weight, on schedule.” UBTech, the first listed Chinese humanoid robot maker, shows what that transition looks like inside a single set of accounts: its full-size embodied-intelligence humanoid robot products and solutions generated RMB 821 million in 2025, a 22-fold increase that lifted the segment from 2.7% to 41.0% of total revenue, on 1,079 units delivered. Its Walker S2 line is quoted at an annualised capacity above 6,000 units. [3]
This is the point at which metal additive manufacturing stops being a prototyping convenience for robot developers and starts being a production input. This article covers which parts are actually being printed, what constrains them, and where in the Greater Bay Area the supply chain for those parts now sits.
Four properties of the application line up unusually well with what powder bed fusion does.
Mass is the binding constraint. Every gram in a limb costs actuator torque and battery life, and the cost compounds — a heavier forearm requires a stronger elbow actuator, which requires a stronger shoulder mount, which requires more battery. Topology-optimised titanium and high-strength aluminium parts built for humanoid applications have been reported at up to 60% lighter than conventionally manufactured equivalents, with improved fatigue life alongside the weight saving. [4]
Part consolidation removes assembly, not just weight. Bright Laser Technologies has shown shoulder brackets, sternum structures, forearms, thighs and finger joints produced as single SLM parts using topology optimisation and multi-scale structural design, replacing what were previously multi-component assemblies. Fewer parts means fewer fasteners, fewer tolerance stack-ups and fewer failure modes. [4]
Iteration speed makes tooling uneconomic. Humanoid platforms are currently iterating on cycles shorter than a year. Amortising injection or casting tooling over a design generation that short rarely pencils out, particularly at the volumes any single structural variant actually reaches.
Some geometries simply cannot be machined. Internal cooling channels, routed cable paths and lattice-filled load paths have no tool access. If the design requires them, additive is not the cheaper option — it is the only option.
The pattern across published examples is consistent: structural, sensing and thermal parts, not cosmetic ones.
Structural. A full-scale demonstrator shown at the World Robot Conference stands 1.65 m and integrates twelve metal AM components — thorax, arms, thighs and shins among them — printed in TC4 titanium using topology optimisation and hollow-section design. [5]
Force sensing. This is the less obvious and arguably more interesting application. The elastomer bodies inside six-axis force sensors — the parts that convert applied load into measurable strain — are being printed as single pieces at diameters below 10 mm. Fingertip, wrist and ankle sensor structures in production humanoid platforms now use this route, with reported weight reductions of 20–30% and measurement range improvements of 50–250% against machined equivalents. [4]
Thermal. The newest category, and the one that connects this supply chain to a much larger one. More on it below.
Notably, printing is not confined to one alloy family. Titanium (TC4) dominates where stiffness-to-weight governs; high-strength aluminium appears where cost and build rate matter more; and copper is entering for thermal parts.
Titanium against aluminium is the first real decision. Titanium wins on specific stiffness and fatigue life, which matters for limb structures under cyclic gait loading. Aluminium wins on powder cost and build rate, which matters when the part count per robot is high and the load case is modest. Most published bills of materials use both.
Powder economics decide whether the case survives contact with volume. Per-part cost in powder bed fusion is governed less by the price per kilogram than by how many reuse cycles a powder holds specification through — a variable that changes effective material cost by a multiple, not a percentage. Robot programmes moving from tens of units to thousands feel this immediately.
Post-processing is where the schedule goes. As-built surfaces are unacceptable on bearing races, joint interfaces and sealing faces. Every printed structural part in a humanoid assembly carries a machining and finishing operation behind it, and that operation, not the print, usually sets the lead time.
The qualification gap is the real barrier. Robot makers need part-level validation — this specific geometry, in this specific alloy, on this specific machine, repeatably. Machine-level certification does not deliver that. A supplier that can show batch data across production lots is in a different category from one that can show a single impressive sample, and the distinction is worth pressing on during any supplier conversation.
As humanoid robots move to continuous operation — UBTech’s Walker S2 is specified for 7×24 duty with three-minute autonomous battery swaps [3] — heat stops being a peak-load problem and becomes a duty-cycle problem. Actuators, drive electronics and compute all need to shed it, in a package with no room for a conventional heatsink.
That requirement is identical in form to the one driving AI data centre cooling, and it has produced a shared manufacturing bottleneck around one material: copper.
Copper conducts heat at roughly 400 W/(m·K), well ahead of the aluminium alloys normally printed, which makes it the natural choice for cold plates and heat exchangers. It has also been close to unprintable. Conventional metal AM uses infrared lasers, and copper reflects most of that energy while conducting away what it absorbs — producing unstable melt pools, porosity and densities too low for pressure-tight parts. This is why aluminium, titanium and steel found production applications years ago while complex copper thermal structures did not. [6]
The unlock is wavelength. At 532 nm, green lasers are absorbed by copper far more efficiently than infrared, stabilising the melt pool. The resulting process window supports densities above 99.9%, wall thicknesses down to 0.15 mm, micro-channels of 0.3–0.8 mm equivalent diameter, and fin aspect ratios beyond 20:1 — geometries where conventional machining risks tool breakage or channel deformation. [6]
Shenzhen Addireen (Hall 15 / E35) is the clearest local example of this capability. The company develops its own 532 nm green laser sources and a machine range from M100G to M660G, with printing services across pure copper, CuCrNb, CuCrZr and tin bronze. In August 2026 it published an application case addressing thermal management for humanoid robots specifically — the same process line it applies to AI liquid cooling plates and optical module heat management. [7] Shenzhen Coolingstyle Technology (Hall 15 / H36) exhibits in the same category.
The economics are worth stating honestly. For AI chips above 800 W TDP, a green-laser printed copper cold plate currently costs two to three times a conventionally milled and brazed equivalent per unit. The case rests on yield, leak-free reliability and system-level energy savings rather than piece price — and the gap is reported to be closing at roughly 20% per year as scan speeds rise and domestic copper powder supply scales. [6]
For a robotics buyer, the practical consequence is that the supplier who can print your actuator cooling manifold is likely the same supplier serving data centre cold plate demand — which is both a capability signal and a capacity risk worth asking about.
The structural advantage in the Greater Bay Area is not any single technology. It is that robot OEMs, AM service providers, powder producers and post-processing shops sit inside the same commuting radius.
Shenzhen counted 4,676 companies holding robotics-related patents in 2025, up 19.44%, and completed 167 robotics financing events — over 22% of the national total. [1] The Nanshan robotics valley alone clusters more than 200 supply-chain companies. [2] Industry participants describe the practical effect as a “half-hour supply circle”: a design change can have suppliers assembled the same day and an agreement in place the same evening. [2]
For a technology iterating on sub-annual cycles, that response time is worth more than a unit-cost advantage.
The genuine bottleneck is not capability — the machines, powders and design tools all exist. It is batch repeatability: producing the two-hundredth part to the same specification as the first, with documentation that survives an audit. That is a process maturity problem, and it is the question worth taking to every stand.
Metal systems
Powders
Thermal
Three questions to carry between stands: what does your batch-to-batch data look like across production lots; how many reuse cycles does this powder hold specification through; and is your certification held by the equipment or by the part production process.
The conference programme includes a Thermal Management and Electronics Manufacturing Forum covering high-conductivity copper alloys, which sits directly on the convergence described above. [8]
TCT Shenzhen 2026 runs 14–16 October 2026 at Hall 15, Shenzhen World Exhibition & Convention Center (Bao’an) — 20,000 m², 250+ exhibitors, 25,000+ expected trade visitors, 80+ conference sessions and 25+ overseas buyer delegations. [8] Metro Lines 12 and 20 both serve Guozhan North station; exits C1 and C2 lead directly into the North Entrance Hall. [9]
Free visitor pre-registration closes at 12:00 Beijing time on 13 October 2026. After that, admission is RMB 30 / US$4.50 per person.
[1] Shenzhen Robotics Industry Development White Paper (2025), released April 2026 at FAIR plus 2026 — total output value RMB 242.6bn (+20.56%); industrial robots 194,900 units (+43.1%, 25.2% of national output); service robots 7.9665m units (42.9% of national); 4,676 companies with robotics patents (+19.44%); 167 financing events (>22% of national). Via 21st Century Business Herald: https://www.21jingji.com/article/20260423/herald/f1abff1c7e6b68bab77ac44eb9ba9498.html
[2] Shenzhen Development and Reform Commission / Science and Technology Daily — humanoid robot whole-machine category output up 83.1% year on year in 2025; Nanshan robotics valley 200+ supply chain companies; “half-hour supply circle”: https://www.stdaily.com/web/gdxw/2026-07/21/content_550790.html
[3] UBTech 2025 annual results (published 31 March 2026) — total revenue RMB 2.001bn (+53.3%); full-size embodied-intelligence humanoid robot products and solutions RMB 821m, up 22x, 41.0% of revenue; 1,079 units delivered; Walker S2 three-minute battery swap, 7×24 operation, annualised capacity above 6,000 units. Via TechWeb / Sina Technology: https://finance.sina.com.cn/tech/roll/2026-04-02/doc-inhtavkc4470886.shtml
[4] Metal AM in humanoid robotics — Bright Laser Technologies SLM part consolidation (shoulder brackets, sternum, forearm, thigh, finger joints), shown at the 2024 World Robot Conference; topology-optimised titanium and high-strength aluminium parts reported at up to 60% lighter; six-axis force sensor elastomers under 10 mm diameter with 20–30% weight reduction and 50–250% measurement range improvement. Via China 3D Print: https://www.china3dprint.com/archives/5457.html and East Money: https://caifuhao.eastmoney.com/news/20251230195104459917420
[5] World Robot Conference exhibit — 1.65 m humanoid demonstrator integrating 12 metal AM components (thorax, arms, thighs, shins) in TC4 titanium with topology optimisation and hollow sections: https://www.worldrobotconference.com/expo/product/704.html
[6] Green-laser copper additive manufacturing — copper thermal conductivity ~400 W/(m·K); infrared reflectivity problem; 532 nm absorption advantage; density above 99.9%; 0.15 mm wall thickness; 0.3–0.8 mm micro-channels; fin aspect ratio beyond 20:1; unit cost 2–3x conventional for 800W+ TDP applications, with the gap narrowing ~20% per year. Analysis citing Huafu Securities, via ITES Shenzhen Industrial Exhibition: https://www.iteschina.com/zh-cn/news/details/2911
[7] Shenzhen Addireen Technologies official site — 532 nm green laser sources developed in-house; M100G to M660G machine range; printing services in pure copper, CuCrNb, CuCrZr and tin bronze; application case on humanoid robot thermal management published 15 August 2026: https://www.addireen.com/blog.html
[8] TCT Shenzhen 2026 official website — dates, venue, 20,000 m², 250+ exhibitors, 25,000+ visitors, 80+ side events, 25+ overseas delegations, conference programme: https://en.tctasia.cn/
[9] Shenzhen World Exhibition & Convention Center official transport information — Metro Lines 12 and 20, Guozhan North station exits C1/C2 to North Entrance Hall: https://www.shenzhen-world.com/overview/traffic/index.html
TCT Asia
