AI Glasses FPC and Micro HDI: Manufacturing at Ten Million Units

Through September 2026, the AI glasses industry has visibly accelerated in both capitalization and scale. Industry information indicates that more than 65 new AI glasses products were launched globally in the first half of the year, XREAL has advanced its Hong Kong listing process, Rokid completed its shareholding reform, and RayNeo Innovation raised over 1 billion RMB in funding. IDC data shows global smart glasses shipments reached 3.566 million units in the first quarter of 2026, growth of 130.1 percent year over year, with 610,000 units shipped in China during the same period. More notably, display-equipped products have gained momentum, as AI glasses extend from voice and capture functions toward visual interaction and more complex terminal forms.

Adding a Display First Squeezes Internal Space

Early AI glasses were relatively simple devices, integrating Bluetooth, microphones, speakers, a camera and basic processing. Adding display capability complicates the hardware structure quickly. Beyond the main controller and wireless communication, the design must accommodate an optical engine, display driver, camera, power management and additional sensors.

The problem is that the glasses do not gain more internal space as a result. If anything, terminal manufacturers must continue reducing weight, shrinking temple volume, and controlling heat generation and battery life simultaneously.

IDC data shows that approximately 2.25 million non-display smart glasses shipped globally in the first quarter of 2026, growth of about 167 percent year over year, while the AR and VR market that contains display glasses also maintained rapid growth. Both product paths are expanding at once, which means both the simpler and the more complex architectures are scaling.

This structural change directly raises the importance of flexible circuits and micro HDI. The main control area must fit a system on chip, memory and power management into a smaller footprint, so routing density continues to rise. The connections between the temples, camera, speakers and battery are better served by flexible circuits, and some designs adopt rigid-flex construction to combine component carrying capacity in rigid regions with space efficiency in flexible ones.

The Flex Challenge Is Stable Bending, Not Bending

Glasses are a high frequency wear product. Opening and closing the temples every day, torsion during wear and internal temperature rise all act continuously on the flexible circuits. The value of a flex circuit here cannot be reduced to replacing a wire harness.

As traces become finer and space tighter, bend region copper thickness, coverlay structure, corner geometry and stress distribution become increasingly sensitive. The area near the camera and antenna must satisfy high speed signal and radio frequency performance at the same time, and if impedance varies too much across the flexible connection region, data transfer or wireless performance can be affected directly.

Because of this, once AI glasses enter the million to ten million unit range, the manufacturing focus shifts from whether a sample can be produced to whether the bend performance, dimensional accuracy and electrical parameters of several hundred thousand flex circuits can be held consistent.

That distinction between prototype and volume is where most of the difficulty resides. A prototype flex that passes bend testing demonstrates that the design works. A production run of 300,000 units requires that the laminate lot, coverlay alignment, plating thickness and bend geometry stay within a window across every panel. Programs building flex PCB assembly capability at this scale need process data that describes the distribution, not just the best case.

Micro HDI Carries More Compute and Power Pressure

As AI capability migrates to the device, the tasks handled by the glasses’ main control board have increased. Image capture, voice processing, wireless communication and some local AI inference run simultaneously, placing conflicting demands on board area, power delivery and heat dissipation. Computing capability must rise, but the PCB cannot grow noticeably larger, and functionality must increase without adding significant weight.

High density interconnect is the practical answer. Smaller laser microvias, via-in-pad structures and tighter BGA fanout reduce the routing area consumed by conventional through holes, freeing the limited space for high speed signal, power and ground networks. For camera modules in particular, fine line capability and high density SMT directly determine final module dimensions.

Manufacturing then has to focus more closely on laser drilling, plating and via filling, layer-to-layer registration and board warp control. Each of these becomes harder as the board gets smaller and the layer count rises, because the same absolute tolerance consumes a larger share of a smaller feature. A registration error that would be acceptable on a large control board can exceed the capture pad on a micro HDI module.

At the assembly stage, devices continue to shrink and pad pitch continues to narrow, which raises the importance of solder paste inspection, automated optical inspection and X-ray. Supporting SMT PCB assembly for this class of product means being able to verify joints beneath packages that cannot be inspected optically, and to do so at a rate that supports production volume rather than only engineering builds.

Supply Chain Competition Shifts to Yield and Cost Curves

The early AI glasses market allowed hardware manufacturers to cover relatively high manufacturing costs through high unit prices, but the mass consumer market does not offer the same tolerance.

IDC estimates that non-display smart glasses alone may reach approximately 13.6 million units globally in 2026. Meta held 69.2 percent of the relevant market share in the first quarter of 2026, showing that leading brands have already established meaningful scale advantages. When order volumes reach that level, PCB suppliers face a different set of metrics: material utilization, panel layout efficiency, SMT yield, test takt time and cross-batch consistency all enter the cost calculation.

A manufacturer able to complete several dozen engineering prototypes is not necessarily able to deliver several hundred thousand units at the same precision. The gap between those two capabilities is the gap between process capability and process control, and it is where supplier selection decisions are actually won or lost at scale.

For supply chain managers, the practical implication is to evaluate suppliers on yield data and process capability indices rather than on prototype quality. For manufacturers, it means that investment in metrology and documentation has a direct commercial return once volumes reach the point where consistency determines cost. A quality management system that tracks these metrics at the lot level makes the difference visible before a customer discovers it.

Where the Value Concentrates

The increment AI glasses bring to the PCB industry is not simply more flexible circuits sold. What matters more is that product structure is concentrating value into micro HDI, rigid-flex construction, high density flex and precision assembly.

This has a specific consequence for capability planning. A supplier whose strength is standard flexible circuit production will find AI glasses volume less accessible than expected, because the requirements combine miniaturization, high reliability, rapid iteration and volume yield in a way that standard flex processes do not automatically satisfy. Conversely, a supplier able to combine PCB fabrication across flex, rigid-flex and HDI with assembly and inspection under one process chain can address the whole requirement rather than a fragment of it.

The iteration requirement should not be underestimated either. AI glasses form factors are still evolving rapidly, and a program that stabilizes its design after two revisions is unusual. Manufacturing partners supporting IoT PCBA development work in this environment need to build multiple board types at small quantities quickly while keeping measurement methodology constant, because otherwise each revision’s results cannot be compared with the previous one.

Display Modules Raise Alignment and Coating Requirements

Adding a display changes more than the component count. It introduces an optical alignment requirement that conventional board manufacturing does not normally carry.

The display module, the optical engine and the waveguide must be positioned relative to each other within tolerances measured in fractions of a millimeter, and the flexible circuit that drives the display often also serves as the mechanical reference that constrains that alignment. If the flex is manufactured with dimensional variation between lots, the alignment shifts even though the optical design was correct.

This makes registration accuracy and dimensional stability on the flex itself a functional requirement rather than a process quality metric. It also constrains how the flex can be handled during assembly, because a flexible circuit that deforms during placement will position the display differently than one that does not.

Coating and stiffening choices interact with the same requirement. A stiffener added for mechanical protection changes the flex’s local stiffness, which affects how it sits in the assembled device. Where the flex also carries an antenna or a camera link, coating coverage affects high frequency behavior, so the coating selection has to be evaluated electrically as well as mechanically.

For programs developing display-equipped wearables, these considerations favor sourcing the flex and the rigid board from a process chain that can evaluate alignment, stiffness and coating together. Treating them as separate purchases typically produces a design that meets each individual specification and still fails optical alignment at final assembly.

The Real Competitive Question

As AI glasses move toward the ten million unit market, the factor that determines a supplier’s position is increasingly not who can make the board smaller. It is who can make the small board consistently, over a long production period, at a sustainable cost.

That is a harder problem than miniaturization because it cannot be solved by a single process improvement. It requires material control, registration accuracy, plating uniformity, assembly precision and test coverage working together, with data showing that each of them remains inside its window over time.

For hardware companies planning AI glasses programs, the practical recommendation is to define the qualification criteria in terms of consistency from the beginning: measured bend performance from production flex, impedance verification on production boards, and yield data across lots. For manufacturers, the recommendation is to build the measurement infrastructure that makes those answers available as a matter of routine rather than as a special project.

The market is expanding quickly enough that both ends of the supply chain will be tested on whether their processes scale. The companies that treat scale as a process discipline rather than a volume target will be the ones positioned to serve it.