Introduction

Last month at MicroLED and ARVR Connect, I gave a “Master Class” on the conference’s first day. I tried to cram about 2 hours of content into a 1-hour presentation. I presented on the trends I saw in optical see-through (OST) AR in 2026. With the conference’s permission to use the video, I decided to turn my presentation into a series of articles. I started with the slides and transcript from my presentation, then slightly reordered them and added more information and images (basically to fill out what would have been a 2-hour-plus presentation). In cases where the slide pretty much says everything, I won’t add too much in the text.

This first part sets up the rest of the presentation by covering my background, the high-level trends I saw in 2026, and the barriers to consumer adoption.

My Background

I originally worked in semiconductors. I spent 20 years designing graphics processors, video processors, and other pixel-generating devices in including the TMS9918, the first sprite chip, and the TMS34010, the first fully programmable graphics processor. I also led the design of at least five CPUs (including graphics and imaging processors), so I was really a CPU architect. I arrived here by a circuitous path after 20 years at TI, where I became the company’s youngest Fellow. Along the way, I worked on Video DRAM (also known as Multi-port DRAM and VRAM) and early synchronous DRAM (essentially a new variation on Video DRAM), which most of you probably use. After leaving TI, I moved from graphics processors that generated pixels to LCOS display devices that displayed pixels.

In 1998, after leaving TI, I began working on an LCOS microdisplay for an AR-like headset (see patent application below, filed in 1999) as CTO of the startup Silicon Display. In 2004, I co-founded the LCOS startup Syndiant. Toward the end of my time at Syndiant, I worked on laser-illuminated LCOS.

Upon leaving Syndiant, I started this blog, KGOnTech. In 2013, I co-founded Navdy and served as CTO, working on an aftermarket automotive HUD that used a DLP projector. From 2018 to 2022, I served as chief science officer of RAVN, a startup working on a military headset.

So while I have worked on display devices and systems since 1998, my formal education and training are in CPU architecture, graphics and image processing, electronics, and integrated circuits. My optical training comes from on-the-job experience, plus exposure to diverse optical systems from this blog’s coverage and the research I do to write about these optical designs.

AR Market Categories

I distinguish portable VR and video glasses from optical see-through products by how they are used. They fall into consumer applications and what I call low-end enterprise applications. For example, a UPS driver receiving package directions may need characteristics similar to those of a consumer product.

High-end enterprise is different. You can add simultaneous localization and mapping, or SLAM, a large field of view, full color, and many other features, but the result starts to become a helmet. I sometimes compare a 30-degree field of view to the speed of sound for AR glasses. You can go beyond it, but it comes at the expense of size, weight, and cost, and it quickly spirals away from anything that even approximates “normal eyeglasses.”

A wider field of view needs more light and therefore more power. It needs more pixels, processing, and communication, each adding power. Soon the device no longer resembles glasses. Extra cameras make this worse: cameras can consume as much power as the display, or more.

Adding SLAM cameras and other features increases power further and moves the design away from a glasses-sized solution. There are also high-end military products.

I don’t see yet where Snap Specs fits. They seem to target consumers while sharing many characteristics with higher-end systems, yet still maintaining the fiction that they are “glasses.” I appreciate that Snap is trying to make a real product, not just a lab prototype. At just over $2,000, the price isn’t out of line in technology history if it can really meet an early adopter’s needs. The question is whether enough people will find uses that justify the price. They are far too big and bulky to be all-day wearable glasses, which means they can’t really fill the role of AI smart glasses. Snap Spec better meets the functionality needed for enterprise applications, but the “glasses-like” form factor is ergonomically bad, and they would be better off with a headband/helmet form factor. But clearly the enterprise space does not have the unit volume Snap is hoping for.

Major Trends in OST AR in 2026

This presentation focuses on trends I have seen over the past year, with an emphasis on what differs from previous years. Originally, optical AR seemed to be working down from VR. People wanted VR’s wide field of view, perhaps 90 degrees, and tried to build complex headsets like Magic Leap and HoloLens that could do everything.

In the last few years, the AR industry’s direction has shifted. Companies are now building up from audio-only AI glasses (such as Meta Ray-Ban Wayfarers) rather than down from VR headsets, as they used to. Meta has had some success with audio-only AI glasses. I say “some” because shipments are reaching roughly 10 million units a year, which seems to be a critical inflection point. We will see whether they fully cross that barrier and break through into the “mass market.” VR headsets hit the 10MU mark in 2021, then fell back to about 3.5MU to 6MU (depending on the analyst) in 2026. Moving beyond enthusiasts and early adopters into the broad consumer market remains a real barrier to VR.

Full-color microLEDs remain slow to emerge. I hear rumors that someone will go into production this year, but even those discussions involve fairly low resolution, regardless of the approach. MicroLED resolution is also improving slowly. We typically see 640 × 480, VGA-class displays, a resolution PCs reached in the 1980s.

Then there is the camera controversy. Can AI glasses work really well without cameras? That remains a major question. AI is also a double-edged sword for component supply. It has absorbed the industry’s spare manufacturing capacity and more, putting increasing pressure on companies trying to build products. These products need semiconductors, and AI’s consumption of components is affecting the whole industry.

Barriers to consumer adoption

I think major barriers remain to AR becoming a consumer product. These are real problems based on my experience in the electronics industry.

One greatly underappreciated issue is the friction prescription glasses create. I have spent the past year trying to use these products as a glasses wearer. I tried contact lenses, but then I needed reading glasses, too. It becomes a mess. The prescription problem is worse than it may first appear. It’s not just about finding an approach; it’s about seeing how that approach fits into today’s retail prescription-glasses infrastructure. Non-display AI glasses simply cut a conventional prescription lens to fit the frame, often locally, but display glasses typically require gluing prescription lenses onto waveguides at the factory, creating a one-per-customer SKU and taking weeks. Furthermore, AR glasses prescriptions are often limited in their ability to correct astigmatism and variable/bi-focal vision, and they don’t work as well for many technical reasons. Aftermarket clip-in prescriptions are bulky and may work in enterprise applications and VR, but not as “glasses.”

Cameras bring their own trade-offs. Size, weight, and styling matter too: people want attractive, small glasses. Solving all these problems simultaneously is difficult. Cameras can also consume considerable power for operation, plus processing, storage, and image recognition of the resultant images and video.

Full-color microLEDs remain slow to emerge. I hear rumors that someone will go into production this year, but even those discussions involve fairly low resolution, regardless of the approach. Resolution is also improving slowly. We typically see 640 × 480, VGA-class displays, a resolution PCs reached in the 1980s.

Then there is the camera controversy. Can AI glasses work really well without cameras? That remains a major question. AI is also a double-edged sword for component supply. It has absorbed the industry’s spare manufacturing capacity and more, putting increasing pressure on companies trying to build products. These products need semiconductors, and AI’s consumption of components is affecting the whole industry.

List of Major Technical Challenges

For many years, I showed a chart of what I see as the major technical challenges to OST AR (right); the count of key issues varies depending on how I group them. Many people focus on cost to drive consumer adoption, but I put it much farther down the list. First, we need to make something useful that early adopters can afford.

Jason Hartlove of Meta has been presenting “$100 per eye for 100 Million Units” (what he dubs 100/4/100) as what is needed to make AR glasses reach 100 million units. I think this is putting the horse in front of the cart. Adjusted for inflation, an Apple II would cost about $6,000 today, and the original Macintosh about $7,000. Putting price ahead of functionality makes it impossible to build anything significant. Adding substantial capability, which I think is required to prove usefulness, while keeping the price very low either won’t happen or will take decades. Lower cost follows from making something desirable, increasing volume, and letting the “democratization of technology” take over.

The chart’s bottom line is that no single barrier matters; you must solve the entire set of barriers simultaneously. Solve one, and roughly 17 others remain. Worse, improving one issue often hurts multiple others. Even if you hypothesize, say, a bright, power-efficient, high-resolution, full-color, and inexpensive MicroLED, you still have many major issues to resolve. This is “too few factor analysis,” where people focus on a single technology’s virtues while ignoring the system’s other requirements.

This slide collects examples ranging from transparency problems to rainbow effects, eye glow, and artifacts from slanted gratings. People are beginning to address some of them.

A slide I took from Meta illustrates a similar point: solving one problem is insufficient. A product has to address all of them.

Other Important OST AR Hardware

The chart on the right comes from work with my Display Skeptics partner, Radu Reit. It covers cameras, audio, and tint, among other requirements. We tried to roughly rate the issues associated with each component.

Dimming/tint is especially important for all-day glasses used outdoors. The display cannot become invisible as soon as the user steps outside. No matter how bright you make the display at the eye, outdoor readability without dimming is a problem.

Eye tracking and retinal sensing could be a major market enabler

Radu Reit likes to emphasize eye tracking and, more importantly, the ability to look into the eye. There are already known medical diagnostic benefits, such as detecting glaucoma and other conditions in real time. As with watches, medical benefits could be a major enabler for AI glasses with or without a display.

AR Optics Basics

Almost anything that combines a display image with the real world degrades both images. It may dim the real-world view, introduce reflections or grating artifacts, or compromise the virtual image. Almost anything placed between the eye and the world affects vision.

For all-day AR glasses, I consider the most important characteristic to be how they work when switched off. Do they function like normal glasses? Poor prescription correction or distortion is unacceptable. If users have to remove the glasses and put them in a large protective case, the product is already moving away from the everyday consumer use people imagine.

We also need realistic expectations for image quality. I frequently hear that people will watch movies on optical see-through glasses. Even the best optical see-through systems I’ve seen, with more than 60% transparency, are unsuitable for watching a movie.

I put birdbath optics, such as XREAL-type products, and similar designs in a different category. At around 20 percent see-through, they are effectively about 80 percent VR. They prioritize the virtual image. Every design makes a tradeoff between the real-world view and the virtual image. If you block much of the real world to improve the virtual image, the result is closer to portable VR. Many applications for those more VR-like products are movie watching and similar activities.

For what I consider true optical see-through, transparency should be at least 80 percent. That requires so many compromises in routing the image to the eye, including diffraction and other effects, that the resulting image generally isn’t suitable for watching movies, even with a dimmer. We need to be realistic about what users will see.

Monocular, Binocular, and Full Color

We see different strategies, from monocular to binocular displays, and different choices about image placement. Binocular designs generally place the image near the center of the view. A monocular design can put it in a corner, as with the Meta Ray-Ban Display. Green-only versus full-color operation is another choice, and these decisions are interconnected.

With a monocular image in a corner, a little smudginess or an imperfect view through the optics may be tolerable. With a binocular image in the center, any loss, diffraction, or other artifact also sits in the center of the user’s vision and becomes much more objectionable.

Most consumers will expect binocular glasses with full-color displays, but there are reasons, and even technical advantages, for the alternatives beyond cost.

A monocular display keeps the image out of the center of the field of view. It keeps the virtual information from covering the real world, and any artifacts caused by the waveguide or other combiner technology wouldn’t affect central vision. Long-time AR expert and advocate Thad Starner is a major proponent of a monocular display below and outside the center of the FOV (see my 2019 FOV Obsession and 2024 AWE 2024 Panel: The Current State and Future Direction of AR Glasses articles).

While binocular is what most people expect, it comes with several issues. It typically results in the virtual image covering the center of the field of view. It’s also much harder to get right, requiring display positioning based not only on the person’s IPD, but also on each eye, since many people’s eyes aren’t perfectly symmetrical horizontally or aligned vertically.

Most OST AR glasses today use green-only LED microdisplays. This works reasonably well for basic content with text and simple symbols. But consumers, by and large, expect full-color displays, and the lack of color limits the content and range of suitable applications.

Brightness and Contrast Ratio

The contrast ratio of a largely transparent system depends on the brightness of what you see after dimming versus the brightness of the virtual image. A 2:1 contrast ratio is barely readable and takes longer to comprehend. A 2:1 contrast ratio also offers very little color discrimination. Some, including Meta, aim for 5:1. This is why outdoor dimming matters. Sunlit concrete can be around 10,000 nits. Delivering something like 50,000 nits to the eye would be very difficult and uncomfortable. Thus, almost any AR OST used outside in daylight will need some form of dimming. While photochromic (light-based) dimming is the easiest to implement, electronically controlled dimming is preferable to achieve the effect the user needs, even though it will add complexity to the system.

Optical efficiency and Étendue and MicroLEDs

I will not go deeply into étendue, except to say it is a complex but important parameter that limits the optical efficiency of a light source. Like the 2nd law of thermodynamics, where energy becomes more random with every interaction/transformation, the product of the area and randomness of light rays can only get worse. When optically expanding a light source, the light rays can be more collimated; but then reducing the same light down to fit through a small aperture increases the angular spread of the rays, and the product of the area and their randomness can only get worse. You trade angular spread against the ability to fit light through an aperture; the product of area and randomness can only get worse.

Etendue is a significant issue for any microdisplay projector systems and a particular problem for MicroLED displays when coupling them to the small entrance area of a typical waveguide. It is a major issue for microLEDs, which often have a near-Lambertian (fairly random) emission profile.

Near-eye displays route light through a waveguide’s very small entrance pupil. The waveguide can direct only the less angularly diverse light rays to the eye. Any light rays outside the angles the waveguide can use must be blocked, or they will cause ghost and contrast issues. MicroLEDs’ broad, approximately Lambertian emission makes that coupling inefficient, and this has been a continuing problem.

Bernard Kress produced a diagram comparing LCOS efficiency with MicroLEDs, DLP, and Laser Beam Scanning at more or less the system level. With LCOS, the illuminating LED sets the étendue, and that LED in AR glasses is typically much smaller than the LCOS display. With an emissive display, such as a MicroLED, the display’s size sets the étendue. As a result, for a full-white image, LCOS can be roughly ten times more efficient than MicroLEDs with typical waveguides. As shown in Kress’s diagram, crossover between LCOS and MicroLED occurs when the average pixel value (aka average pixel lit) is about 10-12 percent.

MicroLEDs have the advantage that you can switch individual pixels off, and power is roughly proportional to the average pixel value. In a typical text image, perhaps 5 percent or less of the pixels are on, and the system can look very efficient. The problem arises when a headset must show a camera viewfinder, a white image, internet content, or other material whose content cannot be controlled. The design must handle the worst case, including thermal dissipation. Inefficient coupling from the microLED into the waveguide becomes a power problem, which becomes a heat-dissipation problem. I like to say amateurs worry about battery life, while professionals worry about thermal management.

Next Time – 2026 Trends in Waveguides

I’m planning the next part of this series, based on my MicroLED and AR/VR Connect presentation, to cover the trends I saw in Reflective and Diffractive Waveguides in 2026.