Views: 0 Author: Site Editor Publish Time: 2026-11-22 Origin: Site
Qualcomm AR1: $55, with no price drop in three years. Yet, its relative weight within a pair of smart glasses has steadily declined—from one-third to one-fifth, and now down to 15.7%. This figure represents the share of the entire Qualcomm "suite" of components, not just a single chip.
The chip’s price hasn’t changed. What has changed is the glasses themselves.
Wellsenn XR has performed a teardown of the Ray-Ban Meta smart glasses. As a non-display AI-powered wearable, the device has a Bill of Materials (BOM) of approximately $149; when manufacturing costs are factored in, the total hardware cost reaches $164. The most expensive component is the AR1 chip—costing $55 and accounting for one-third of the BOM—followed by the frames and lenses, which cost about $37. In this context, the glasses function essentially as a smartphone worn on the face, with the bulk of the value concentrated in the silicon.
The picture changes entirely when looking at the Rokid Glasses, which feature a built-in display.
The BOM here is $285. The display and optical system account for $120—or 42.1%—of the total. The mainboard and its chips represent 36.8%, with the AR1 chip again priced at $55. The Chinese supply chain contributes $172 in value (60.4%), while the U.S. share stands at 34%, driven primarily by the chip.
Meta’s own Ray-Ban "Display" model takes this even further. Third-party estimates place the cost of the complete hardware package (including an EMG wristband) in the $550 range. A teardown by TechInsights reveals that display-related components—the largest cost item—account for 50.8% of the total, whereas the entire suite of Qualcomm components makes up only 15.7%. The retail price is $799.
From 164 to 285, and then to 550.
Almost all of the additional cost went toward a single purpose: letting light into the eye.
What does $120 buy you?
A Micro-LED light engine, plus two waveguides. Light emerges from the engine, is collimated, and coupled into the glass; it then travels through a series of total internal reflections before being coupled out and directed into the eye. As the light bounces repeatedly within the waveguide, a significant portion is lost—dissipated during coupling or leaking out.
Consequently, the light engine must be exceptionally bright. Rokid employs a "one-to-two" configuration: a single light engine is positioned in the center of the frame, splitting the light into two paths—one for each eye—to achieve a brightness exceeding 1,500 nits. This design eliminates the need for a second light engine, an additional micro-display, and an extra set of coupling processes. The trade-off, however, is that the initial brightness must be high enough to ensure a clear image for each eye after the light is split.
Brightness requires power. Power requires a battery. A battery adds weight.
That $120 figure in the Bill of Materials (BOM) represents only half the story—the part that’s actually visible.
Xiamen, China. At 2:00 AM, the Micro-LED production line at Tianma is still running.
Xi Kerui stands beside the inspection equipment. The chips are finer than a human hair—invisible to the naked eye. Incoming components undergo AI-driven sorting, laser lift-off, flipping, and transfer. This sequence is known as "mass transfer"—the make-or-break process for Micro-LED technology.
He once cited a figure: a single machine can transfer 40 million chips per hour, with a yield exceeding 99.99%. Each additional "9" reduces the number of defective pixels by an order of magnitude.
But 99.99% is not enough. A single display contains tens of millions of individual light-emitting elements; a failure rate of just one in 10,000 would result in hundreds or even thousands of defective pixels. Xi Kerui’s next goal is 99.9999%, which would bring the number of defects down to single digits. Two more "9s" to go.
What appears as a single line item for a "micro-display" on a bill of materials translates, on the production floor, into countless nights just like this one.
Once the light source challenge is met, the optical path presents an even greater hurdle.
A waveguide is essentially a piece of glass etched with a nanograting. Parameters such as the tooth profile, groove depth, duty cycle, and the alignment precision of the lithography layers must all be controlled within 100 nanometers—a tiny fraction of a human hair, which is roughly 70,000 nanometers in diameter.
Even the slightest deviation results in uneven brightness across the output, rendering the entire piece useless.
In this industry, the average yield for mass production hovers around 50%. AAC Technologies has achieved a yield of over 80% for single-layer full-color surface-relief gratings (SRG), while Zhige Technology reports an 85% "pass-through" yield, reaching 90% for projects with key clients.
Using the same production lines and equipment, one company might produce 800 functional units out of every 1,000, while another produces only 500.
That gap of 300 units needs no explanation on a price quote.
Scaling up the mass production of next-generation waveguides cannot be achieved simply by purchasing standardized equipment. Buying the same nanoimprint or lithography machines merely secures your entry ticket. If a tiny defect on the master template isn't caught, it gets replicated with every single imprint—a localized flaw that ruins an entire batch.
Silicon carbide (SiC) is even more demanding. Harder yet more brittle than glass, it presents lower yields at every stage—cutting, grinding, and polishing. Even at the industry’s cutting edge, edge chipping can only be contained within 10 micrometers.
Materials, processes, and inspection—these three pillars are inextricably linked; a change in one ripples through the entire system.
Consequently, you cannot simply "fix" your way to a higher yield. Yield is determined at the design stage and forged through hundreds of rounds of trial production.
In June of this year, in Shanghai’s Lingang area, Goertek Optics launched a 12-inch wafer DUV (deep ultraviolet) lithography and etching production line—the first of its kind in China. A CCTV reporter visited the facility: where lenses were once ground individually, they are now produced like newspapers—printed in bulk across a single wafer.
Less than six months after the line went into operation, the resulting lenses were on display at the China International Import Expo (CIIE). On the exhibition stand sat two items: a silicon carbide waveguide lens and a full-color Micro-LED optical engine. According to CCTV: "Previously, high-end optical waveguide lenses like these could not be mass-produced domestically; we had to pay a premium to import them from abroad. Now, however, we are capable of manufacturing them ourselves."
Industry sources indicate that the price for a single imported diffractive waveguide lens is nearly 1,000 yuan. Guangzhou-based company Guangna Siwei (Guangna 4D), located in the Huangpu District, has managed to cut that price by more than half—all while pursuing the most challenging technical path in the industry: single-element full-color waveguides. This wasn't achieved through mere negotiation; the company has brought the entire stack—materials, equipment, software, and packaging—under its own control.
Shi Rui is searching the globe for talent. Those who truly understand the EDA (Electronic Design Automation) algorithms for diffractive optical waveguides can be counted on one’s fingers—and that’s assuming they are also willing to relocate to Guangzhou, China, and dedicate themselves to the etching of silicon carbide.
A headhunter in Shenzhen once crunched the numbers: a single pair of AR glasses involves 47 distinct specialized roles, more than half of which are entirely new. Last year, when trying to recruit a simulation engineer for diffractive waveguides, the offer jumped from 450,000 to 800,000 RMB plus stock options—yet, even after six months, a second candidate could not be found.
Production capacity can be built in six months. But these people? You can’t find them in six months.
This is precisely why capital is willing to pour heavy assets into optical companies. Goertek entered the XR optics space in 2012; by 2020, it had established its first 12-inch nanoimprint mass-production line, with a DUV (Deep Ultraviolet) line slated for production in 2026. Fourteen years—all to refine a single yield curve. New entrants might be able to purchase the same equipment, but they cannot buy the technical parameters forged through hundreds of trial-production runs—nor can they acquire the people who know exactly where those critical parameters are hidden.
The bet isn’t on production capacity. It’s on a time advantage that latecomers simply cannot bridge.
Things are quiet, by contrast, on the chip front.
The AR1 is, in effect, a "tax." In glasses without a display, it’s an unavoidable $55 cost. In glasses with a display, while its relative share of the total cost is squeezed by the optics, the underlying architecture becomes significantly more complex. Rokid, for instance, employs a dual-chip setup—a Qualcomm AR1 paired with an NXP chip—where the primary processor handles heavy lifting, while the co-processor remains "always-on" to listen for wake words. According to a teardown by Wellsenn, the total bill of materials (BOM) for Alibaba’s Quark S1 reaches nearly 3,000 RMB.
This isn't just a display of technical prowess; it’s because the challenges of on-device computing power and power consumption have yet to be solved by a single, unified chip.
However, there is a clear roadmap for this trajectory.
The BES2800 from Bestechnic (BES) has already been integrated into products from Xiaomi, Quark, RayNeo, and Rokid. Its successor, the 6nm BES6100, is designed to consolidate dual-chip functions into a single die; it is slated for tape-out this July. Public reports suggest a unit price of under $40—a significant reduction from the roughly $78 total cost of current dual-chip solutions.
If successful, this could nearly halve the cost of the "chip" line item in the BOM.
Manufacturing processes, architecture, tape-out, validation, and mass production—each of these milestones can be mapped out on a calendar.
The same cannot be said for optics.
There is a rough consensus in the industry: only when the cost of a full-color optical engine—for a single display—is driven below $149 can manufacturers realistically target a "breakthrough" retail price point of under $300. We are still some distance away from that. Rokid’s $120 optical solution, for instance, relies on a "one-to-two" hybrid architecture; a true full-color solution is even costlier—the full-color version of the Rokid Color starts at $749.
Consequently, companies are betting big on capital-intensive infrastructure. Saphlux is aiming to push the cost of a monolithic full-color display below $45, backed by a $150 million investment in Wuxi. JBD has transitioned from 4-inch to 12-inch wafers, achieving a yield rate of over 98% on its pilot production line. RayNeo has built its own optical factory, painstakingly refining yields—managing to shrink the optical engine for the X3 Pro to 0.36 cubic centimeters, a 40% reduction from the previous generation. Meta has opted for LCoS combined with arrayed waveguides rather than Micro-LED; this isn't because they are unaware of Micro-LED’s superiority, but because the cost curve for consumer-grade, full-color Micro-LED simply isn't ready yet.
None of this follows the typical "internet-style" (software-first/asset-light) playbook.
If these bets pay off, by 2027, you might be able to purchase a full-color AR device for under $300. If they fail, we will likely remain stuck for another two years in a market segment where such devices are relegated to "geek gadgets," priced anywhere from $449 to $749.
Unpacking a pair of returned AR glasses—used for just three days.
No scratches on the temples, no oily sheen on the nose pads, the box kept in pristine order.
The reason for the return? "Wearing them feels like living in the future; using them feels like being a beta tester."
The hardware has arrived, but the use cases are still waiting in line.
For chips, there’s a roadmap for cost reduction. For optics, there’s only the yield curve.
Wafers in Shanghai’s Lingang, recruitment emails in Guangzhou, 2:00 AM in Xiamen...
With every percentage point the yield climbs, the line inches forward.
Source: VR Jiajia