A Revenue Mix That Says More Than a Headline
On 28 August 2026 a major Chinese precision manufacturing group published its first half-year report after listing in Hong Kong. Revenue for the first half reached 25.149 billion yuan, up 6.45 percent year on year, with gross margin at 16.73 percent, an improvement of 1.65 percentage points. Within that total, AI hardware related business reached 20.464 billion yuan, or 81.37 percent of revenue. Thermal management contributed 2.920 billion yuan, up 43.46 percent, and XR wearable products contributed 1.794 billion yuan. Research spending reached 1.442 billion yuan, directed at foldable displays, vapor chamber cooling, server liquid cooling, power supplies and embodied robots.
Read carefully, the report also separates two effects that are often confused. The gross margin improvement came from product mix rather than from volume, because higher-value assemblies carry better margin than the parts they replaced. Meanwhile the research spending line names the technologies the company expects to matter next, and almost all of them are interface problems between mechanics, thermal hardware and electronics rather than purely mechanical problems.
The interesting number is not that AI appears in most of the revenue. It is that the same manufacturing capability now serves consumer electronics, extended reality headsets, servers and robots at once. Those are different electronic systems with different board requirements, and serving all of them from one process base is a statement about how far precision manufacturing has migrated.
From Mechanical Precision to Electronic Precision
Companies in this category built their reputation on mechanical parts: housings, frames, hinges and thermal hardware, produced to tolerances measured in tens of microns. That capability base has been moving steadily toward electronics because the products themselves demand it. A foldable phone hinge, a vapor chamber, a liquid cooling cold plate and a high-density printed circuit board all depend on the same underlying disciplines: dimensional control, surface quality, material handling and process repeatability.
The transfer is not automatic. Metalworking tolerances and laminate tolerances behave differently. Copper expands, resin flows, and a laminated stack changes dimension in ways that a machined part does not. A supplier entering electronics from a mechanical base has to build new metrology, new process windows and new acceptance criteria, and the organizations that did this successfully treated it as a genuine capability program rather than as an extension of existing capacity.
Once a manufacturer can hold those tolerances in metal, extending the same control to laminates, solder joints and flexible circuits is a natural progression rather than a leap. The teams that made that transition early are now positioned to supply subsystems rather than individual parts, which changes both the revenue per unit and the engineering relationship with the customer.
Why Thermal Management Sits at the Center
Thermal business growing at more than forty percent is not a coincidence. Nearly every product in this portfolio is thermally limited. A thin phone cannot dissipate heat through surface area alone, a headset runs a display and a processor in a sealed shell, and a server rack concentrates tens of kilowatts into a volume that must remain serviceable. Each of those problems is solved by combining a material, a structure and a board design.
This is where the disciplines converge. A vapor chamber spreads heat from a processor, but the rate at which heat reaches it is set by the copper structures and thermal vias in the board stackup. A liquid cooling plate removes heat from a rack, but the interface between the plate and the board assembly determines the actual thermal resistance. Thermal management becomes a system property, and the organizations that own several parts of that system have an advantage over those that own one.
A second thread runs through the thermal portfolio: interface materials. A vapor chamber, a heat pipe and a cold plate all depend on the layer between them and the heat source, and that layer is where most thermal designs lose performance in production. Controlling its thickness, coverage and cure across a production run is a materials discipline that sits closer to electronics assembly than to metal fabrication.
XR Devices: The Hardest Assembly Problem in Consumer Electronics
Extended reality hardware compresses more function into less volume than almost any other consumer product. Cameras, inertial sensors, displays, eye tracking, radios, batteries and processing all occupy a space that must remain comfortable on a face or a head. The interconnect is usually a rigid-flex construction with multiple rigid islands, and the tolerances are tight because an optical path and an antenna share the same few millimetres.
Producing these assemblies requires more than a capable placement machine. It requires handling of thin and flexible substrates, controlled adhesive processes, precise connector seating and inspection that can verify optically critical parameters. These are electronic assemblies that sit at the intersection of mechanics, optics and electronics, and most of the yield is lost at the intersections rather than inside any one discipline.
Server Liquid Cooling Changes the Board, Not Just the Rack
Liquid cooling in data centers is often described as a facility decision, but it reaches the board. When a cold plate sits close to the package, the board must survive mechanical loads from the mounting hardware, and the assembly must maintain flatness so that the thermal interface does not develop voids. Warpage during reflow and warpage during operation both affect contact pressure, and both are influenced by the layer structure and copper balance of the board itself.
There is a manufacturing consequence as well. Liquid-cooled assemblies are often built, tested and shipped as complete units, which means the factory has to handle fluid interfaces, leak testing and packaging for a heavier, more fragile product than a bare board. This is a different logistics and test discipline, and it is normally found in organizations accustomed to building boxes rather than boards.
Connector and manifold placement also becomes a layout question. Tubing and manifolds occupy space that would otherwise be used for routing, and service access has to remain possible. Suppliers who can advise on these interactions, rather than only building the board to a drawing, shorten the customer’s development cycle.
Power Supplies and Robots Complete the Picture
Power supply work appears in the same research portfolio for the same reason: high-performance systems need efficient conversion at high current density. A server power stage and a robot joint driver share design requirements in loop layout, thermal spreading and current sensing, even though the products look unrelated. Building capability in one transfers directly to the other.
For board and assembly manufacturing this convergence creates predictable demand patterns. Compute-heavy boards need high layer counts, fine lines and controlled impedance. Power-heavy boards need thick copper, thermal vias and careful high-current assembly. A supplier serving both ends of the portfolio has to maintain two quite different process regimes under one quality system, which is exactly the challenge the industry reports show customers are increasingly asking for.
Building Subsystems Instead of Parts
The strategic move visible in this kind of revenue mix is a shift from components to subsystems. A customer who previously bought a thermal part and separately sourced a board now prefers an integrated assembly that arrives tested and accountable. That preference has a practical basis: interface problems between two suppliers are expensive to diagnose, and responsibility for a system-level failure is difficult to assign.
The commercial logic behind it is straightforward. A subsystem supplier can absorb design changes late in a program because it controls more of the interfaces, and it can accept responsibility for a functional outcome rather than for dimensional conformance. Customers increasingly prefer that arrangement because their own engineering teams are stretched across more platforms than they can support in detail.
Supplying subsystems requires manufacturing partners to hold capabilities that used to be separate. Board fabrication, component procurement, dense assembly, thermal interface integration and functional test all have to work under one process discipline, with quality management covering interfaces rather than only individual steps.
That shift also changes the qualification burden. A subsystem has to be validated for mechanical shock, thermal cycling and electrical function as one unit, and a failure in any dimension stops the shipment. Manufacturers that move into this role typically build test infrastructure before they build sales capacity, because the customer’s acceptance criteria now cover the whole assembly.
What This Means for Procurement Teams
The practical lesson for anyone sourcing electronics in 2026 is to evaluate suppliers on breadth of process control rather than on a single capability. Ask which product families the supplier serves, how process knowledge transfers between them, and what evidence exists that the transfer actually happened. A manufacturer moving from mechanical precision into electronics has to demonstrate measurement discipline on the new materials, not assume it carries over.
The revenue mix discussed here is one company’s result, but the pattern is broader. As AI hardware spreads into wearables, vehicles, servers and robots, the boundary between mechanical and electronic manufacturing keeps thinning, and the suppliers that treat it as a single integrated problem, and invest accordingly, will keep gaining share.
