Usually, when people say that a technology can improve over time, they make an automatic assumption that products will naturally get better and better. Punters wave their hands and go, “don’t worry, it’ll be resolved.” It's less of a case here because companies are quite literally fighting against the laws of physics. Waveguides are incredibly difficult to make and are supply-restricted, and the chips and cameras draw so much power that they drain the battery and generate heat. Try wearing glasses that can bake the sides of your face.

This is why this is such a fascinating new product category. For everyday users, this could be a really interesting new type of hardware; a hands-free device that gives users contextual information or records activities while they sort other items. Plus the specifications are remarkable, and there are components which are either difficult to make or rare; it mimics the AI race for RAM in some cases.

There are global elements to this too. China has a real ecosystem when it comes to its own manufacturing processes, as well as innovation through state funding and cutting-edge technologies. The West has its own strengths: US firms such as Meta, Qualcomm and Google own the platforms, chips and AI, while Europe and the UK lead in precision optics through companies like EssilorLuxottica, SCHOTT and WaveOptics (owned by Snap).

It is hard to tie a neat bow around this, as manufacturing explores lenses, AI, compute, chains, countries, and many other items besides. Plus supply chain information can be as interesting as reading the manual for a Volkswagen Polo. But I will try. I will break down the key elements of the manufacturing process and then provide my own views on design considerations. Let’s dive in.

There are effectively six different types of glasses to be explored, and I'll be ranking them from the most expensive and most complicated down to the cheapest and most straightforward to make. And just for fun, I will try to compare each one to car brands (or at least try to).

Think of this as the most technically proficient option, though it may lead to you re-mortgaging your home. These specs are the most expensive as they require wide waveguides, several tracking cameras and serious computing power. These are the Ferrari of the specs world; the absolute best quality possible, and (quite literally) hand-made with each unit (at least for now in Meta’s case).

As examples, Snap's Specs cost $2,195, weigh 132 to 136 g and last about four hours, using two Snapdragon chips, one of them dedicated to computer vision. Meta's Orion prototype (the far-horizon vision of the company’s goals with smart glasses) reportedly cost around $10,000 per unit to build, with the silicon carbide lenses the main culprit.

One step down, but still useful and technically impressive. Think of these as glasses which can still provide contextual information on your lenses, but without AR. We are now downgrading from a Ferrari to a Porsche; still astonishing and impressive, but it turns fewer heads at a luxury party.

These glasses are still expensive because full colour is far harder to make than a single colour. For example, the Meta Ray-Ban Display costs $799 and weighs 69 g, with a small projector reaching up to 5,000 nits and a 20° field of view. Display components make up about half of its bill of materials. Supply was so tight that Meta's waveguide orders were raised from 80,000 to 150,000 (at the time of writing).

These are the good enough ones, using a single colour to show contextual information such as notifications, directions and translations. Think of it as going from a Porsche to a Jaguar; it has a similar design and comparable capabilities, but with compromises to fit the vision.

Why one colour? Green is the easiest colour to make efficiently at tiny sizes, and the eye is most sensitive to it. For example, Even Realities' G2 weighs just 36 g, has no camera, and supports prescriptions from −12 to +12.

Perhaps the most mainstream, as all they require is a camera, microphones and AI. For example, the Ray-Ban Meta Gen 3 costs $449 (£409 in the UK), and in China Xiaomi's AI Glasses sell for about $280. I would say these are comparable to Audi; still a little luxury, but with a mainstream element to it (primarily because you are moving on from the expensive optics).

Later in this piece, I will argue that (design-wise), I suspect that this product will perform the best from a manufacturing context.

We’ve now hit Volkswagen Polo territory; still expensive as you’re still buying a car, but it’s a more mainstream item regardless. This form factor is the cheapest and simplest, dropping the camera entirely. Examples include Meta's Ray-Ban Meta Audio, which weighs 43 g and claims 12 hours of battery life for $349. Best used to get info when needed while sorting items inside and outside the home.

I’ll drop the car metaphor here as, honestly, I see it as a separate product category which feels more comfortable using at home. (Plus the car metaphor is wearing its welcome). Think of these as glasses tethered to either a puck or a phone. Brands such as XREAL, VITURE and RayNeo skip waveguides entirely and use "birdbath" optics, a curved mirror in front of a small screen. They are bright and colourful, but bulkier, and they dim the world around you, so they suit watching films or working rather than all-day wear.

And those are the form factors! Phew. Now bear these in mind as we will now go into the core engineering and manufacturing side of the piece, as this will be more relevant for some devices than others. But you can keep one key takeaway in mind: the engineering challenge is massive.

These areas are as wide and disparate as genres in a book shop, but there are several broad categories which are worth diving into. And for each of the sections I will label which types of glasses these areas are most relevant to, in case you wish to focus on the types you prefer.

This is a real challenge, and really difficult to make. Waveguides struggle to get even 1% of the projector's light to the eye, so the projectors have to be extraordinarily bright.

Why are they so hard to make? Well to put it briefly, a waveguide has to carry an image through a thin slice of glass while you still see the world clearly through it, which means patterning structures with precision at a tenth the thickness of a human hair, or stacking and bonding layers of tiny mirrors without a single flaw. Any imperfection shows up as blur, rainbows or dim patches, so yields are low, and the best materials, such as silicon carbide, are notoriously hard to manufacture. These are also expensive and supply-constrained; making these takes time, and they have not ramped up to a broader market.

Relevant to: Standalone AR glasses, full colour display AI glasses and monochrome HUD glasses. Remember, tethered media glasses use simpler birdbath optics instead of waveguides.

This is dominated by companies like EssilorLuxottica, which makes the Ray-Ban and Oakley Meta glasses and held over 70% of smart glasses manufacturing in 2025. This really shows the dominance of key players in the creation of glasses. Otherwise, Google and Samsung have partnered with Warby Parker and Gentle Monster.

Prescriptions are a real engineering problem too: the Meta Display supports only −4 to +4, while Even Realities bonds the prescription directly to the waveguide. The number of people who wear glasses is huge globally, so it may narrow the number of people who can use specs.

(I attempted to see if this is still within the bounds of common prescriptions, but I cannot find information. If it is, then this could be seen as a minor element. If not, then expanding prescriptions may be important to capture more people).

Relevant to: All glasses

The engine of the device. Qualcomm has a big role to play here, as its Snapdragon AR1 is the default chip across both Western and Chinese brands; and its successor, the AR1+, is 26% smaller and can run a 1 billion parameter AI model on the glasses themselves. Competitors like Apple would be making their own silicon.

But then there is the question of whether to process on device or beyond. You could do it on the device, but it could heat up and drain the battery more. You could pass it to the cloud or another device, but that means carrying another device with you. Meta's Orion, for example, relied on a wireless puck. This is one of the most important differentiators between devices as it leads to a genuine user impact. Would you like to carry a puck, or have glasses with (potentially) compromised power?

Relevant to: All glasses (varies on required power).

This is a really interesting one because, for the form factor, batteries have to be quite small. The first Ray-Ban Meta glasses had a battery of about 154 mAh, and the Meta Ray-Ban Display about 250 mAh, compared with 4,000 to 5,000 mAh in a typical phone. Heat can be quite high as well, so it needs to be regulated. And bear in mind there is no fan, and the hot parts sit against your temples.

Progress is striking, though. Snap's 2024 developer Spectacles weighed 226 g and lasted 45 minutes, while its 2026 Specs weigh 132 g and last about four hours. But still, you’re fighting against physics here, and you can’t exactly uppercut the fabric of space towards a cool pair of shades.

Relevant to: All glasses

A rather straightforward area (at least compared to other parts) with well-established supply chains. 12 MP cameras are now standard, with much of the challenge lying in processing the images within a tiny power budget. Meta uses arrays of six microphones and says the Audio model blocks 90% of street noise. Open-ear speakers have to be loud enough for the wearer without leaking to those around them. Recording lights are now effectively a design requirement, as explored in my piece on privacy.

Relevant to: All glasses (cameras only for standalone AR, full colour display AI and camera AI glasses)

It's important to be able to read the world, but that may mean a separate chip which drains the battery further. Mapping a room in real time needs several cameras running constantly, which is why Snap dedicates a whole chip to it.

Still, most AI glasses skip this entirely. They take a snapshot when you ask "what am I looking at?", send it to a cloud model, and anchor nothing in 3D space.

Relevant to: Standalone AR glasses (though camera AI and full colour display AI glasses use a simpler snapshot approach)

Oh look, the societal elephant has entered the room. Still, it’s important. Counterpoint found AI features in 88% of smart glasses shipments in H2 2025, and there's a reason why Meta is focusing on its own models. If you want to make good smart glasses, having a good enough AI model which is energy-efficient and can run on the device is absolutely vital for success.

For now, almost all the heavy reasoning still happens in the cloud, whether through Meta AI, Google's Gemini on Android XR or Alibaba's Qwen in China. I softly suspect that AI agents such as Muse, which Meta created and which can already shop with retailers like Walmart and Instacart, are a precursor to this, because the same efficiency on device could be applied to agentic use and other AI technologies. But we will see.

Relevant to: Standalone AR glasses, full colour display AI glasses, monochrome HUD glasses, camera AI glasses and audio AI glasses

A relatively straightforward area. Operating systems are consolidating into a few camps: Meta's own platform, Google's Android XR (now opened to XREAL and Vivo), Snap OS, which Snap says it built partly to control how heat is spread, and Chinese ecosystems tied to super apps, such as Xiaomi's HyperOS and Alibaba's Quark.

Relevant to: All glasses

This is really difficult because so many pieces are very fiddly. The waveguide is so delicate that when it breaks, you effectively have to replace all of it. Components are glued and sealed in, and if you want the glasses to be sweat-resistant and water-resistant, everything has to sit in an enclosure, which by its own nature makes it harder to repair. iFixit found that reaching the Meta Ray-Ban Display's battery means splitting a glued seam on the arm, which sacrifices its splash resistance, and Meta sells no spare batteries.

This also clashes with regulation (to an extent), as the European Union's batteries rules push for users to be able to swap out their own batteries from 2027, which may impact design.

Relevant to: All glasses

And breathe! That’s all of them, grotesquely summarised. But it’s just enough to help you get the area of the market, and know where to dig deeper. But now I want to go a little deeper and see where all this may lead.

If it is such a drastic engineering challenge, what type of device would win out from a design perspective? Ironically, for me, I feel like it's the lower end, at least in 2026. You don't necessarily need to climb up the ladder to glasses with waveguides or displays to fulfil user needs. So my top contender over the next few years will be the Camera AI glasses, WITHOUT the display or waveguides.

Why? Let me explain.

From a design perspective, you have to consider the core user needs and what people actually care about. What qualities do people want with their smart glasses?*

The camera itself needs to provide context, whether it’s tourists exploring the world or seeing a problem first-hand. This is why I feel that you need a camera to see the world, as well as AI to provide further information. That's very clear-cut, though it's worth noting that privacy concerns are pushing some firms the other way, with Meta launching camera-free audio glasses and most HUD glasses skipping the camera entirely.

This blurs with the differential category, but in the sense that it’s the inherent qualities of the product as opposed to how different it is from other types of work or play.

If you're going to be using these day to day, they should at the very least be fashionable, or people shouldn't be embarrassed to wear them. This is a key problem at the moment because, for lack of a better word, they are relatively thick and not so slick (rhyme unintended). Fashion matters here, in the same way it did for the Apple Watch, which Apple launched as a fashion object before it found its footing in fitness and health. Fashion got it onto wrists, and usefulness kept it there. The same with smart glasses, too.

I am also aware that this is a very difficult design challenge, because you have to fit all the components in somehow. I'm merely pointing out that it just needs to look good, which I know might be tricky. Glasses will be used every day and battered around, potentially used for sports and more, so they need to withstand wear and tear. This directly clashes with repairability, as the glued seams that keep sweat and rain out are exactly what make batteries so hard to replace.

As explored elsewhere in the Immersive Wire, privacy is a big concern for a lot of people, and it has interesting consequences for design. If you want to be private, you want the information to stay on the device and be processed there. I'm willing to bet that Apple will make that argument with its N50 smart glasses. But the problem is that on-device processing can drain the battery faster and generate more heat, and it limits how capable the model can be. Privacy has an energy cost.

Blurs with the useful category, but to a point. We already have mobile phones in our pockets, which can take photos, let us talk to our friends and give us information about the world. What are the core reasons for using glasses instead? Why use specs, anyway?

In my view, you can wear them while doing things, so you have both hands free for certain tasks. It could be if you're parenting a child and need contextual information, if you're baking some cakes and want to show off what you're making, or if you're doing sports and want to focus on the sport while recording it. And all of these qualities must be baked into the design of the product. (All of this tells me that there needs to be some element of durability, too).

*All of these qualities are based on my own assessment and qualitative discussions, but I haven’t surveyed several thousand people to really pin down user needs etc. If a large survey was done, then more accurate user needs could be spotted and then designed for.

Of all the product categories that best serve users, the closest would be the camera AI glasses. Display glasses will be more expensive devices, applying to more niche but still relevant areas of people's lives. There may be an argument for waveguides providing that in the future, with pop-up information, displays or text. But considering that AI can already provide the same information by voice, perhaps it's the case that cost will win out for the mass market, with people choosing camera and AI glasses rather than those with a waveguide.

Once waveguides cost tens of dollars rather than hundreds, maybe displays will become the differentiator. But for now (or in 2026), the cheap end is where the volume is. In a market which seems to be price elastic, I strongly suspect the cheaper glasses with on-device AI will win out, provided the battery issues are addressed.

Now that we've established all this, let's take those insights across to actual manufacturing and consider the wider market context. After all, what would help a device win out in the future?

In the meantime, if you want to be updated, please subscribe to the Immersive Wire at www.immersivewire.com.

Note: The Immersive Wire is run by Tom Ffiske, who also works at Accenture. The contents of the website and newsletter should not be regarded as Accenture’s views.