Methodology

Twelve companies were analyzed: Meta, Apple, Microsoft, Google, Snap, Samsung (reported by legal entity — Electronics, Display, and other affiliates), EssilorLuxottica, Qualcomm, XREAL, Xiaomi, Vuzix, and Amazon.

To build a complete picture for each of them, patent activity was captured through two complementary searches — one anchored on smart-glasses and wearable-display hardware terminology, and one anchored on AI-capability terminology (assistants, speech, hand and eye tracking, gesture, spatial mapping, privacy, and memory/agent classifications). Results were combined at the patent-family level and validated for relevance through manual review of representative samples, producing the estimates used throughout this report.

Corporate variants and acquired entities (for example, Meta's Oculus and DAQRI holdings, or Google's North Inc. and Thalmic Labs) were normalized to their parent company, so the counts throughout this report reflect operating businesses, not legal shells.

One caveat applies across the board: because recent patent applications are not published until roughly 18 months after filing, 2025 figures likely understate true recent activity across every company.

A small number of estimates would benefit from further refinement — noted in the closing section — but none change the direction of any finding in this report.

The Smart-Glasses Patent Leaderboard

Start with the simplest possible question: who has filed the most patents in this category? The honest answer depends entirely on how the question is asked. Count everything remotely display-adjacent and one company appears to have lapped the entire field. Count only what's actually relevant to smart glasses, measured the same way for everyone, and a very different order emerges.

Figure 1. Validated smart-glasses-relevant patent families by company.

Two companies with no shipping AI smart glasses product already outrank most of the companies that do.

The top five is Meta, Apple, Microsoft, Google, Snap. Samsung Electronics is sixth on a like-for-like basis (~550), ahead of EssilorLuxottica (~530) and Qualcomm (~360), which in turn are both larger than XREAL, Xiaomi, or Vuzix.

Two companies with no shipping AI smart glasses product hold more relevant patent IP than five of the seven companies that actually sell glasses — a detail worth sitting with, because it's the thread the rest of this report pulls on. But before getting to why Apple and Microsoft show up this high at all, there's a more immediate puzzle to resolve: the single largest number on this chart may not mean what it appears to.

The Two Samsungs Hidden Inside One Patent Portfolio

Samsung's patent activity is often reported as one eye-popping number. A number that large invites an obvious question: is it measuring one thing, or several? Splitting it by legal entity — Samsung Electronics (814 families), Samsung Display (381), and other affiliates (12) — suggests it's the latter.

The classifications underneath each entity confirm it. Samsung Electronics carries the classic AR/XR glasses signature: HMD optics (G02B27, 921 families), input and interaction (G06F3, 1,172 families), and vision/AI classifications (G06V/G06T7, 242 families).

Samsung Display's portfolio tells an entirely different story — dominated by OLED-panel manufacturing codes (H10K59, 1,252 families), pixel-driving control (G09G3, 267), and OLED materials and manufacturing (roughly 460 more): upstream microdisplay IP, not glasses-system IP.

A closer relevance review confirms the split is real, not an artifact of the search: Samsung Electronics' portfolio is overwhelmingly core smart-glasses technology, while Samsung Display's is dominated by generic OLED-panel and foldable-phone-digitizer IP that surfaces only because of an incidental HMD or wearable-display keyword.

On a validated, like-for-like basis, Samsung Electronics' relevant portfolio is approximately 550 families, and Samsung Display's is approximately 60. The figure used for competitive rankings elsewhere in this report is the ~550-family Samsung Electronics number; Samsung Display is best read as a supply-chain sidebar, similar to Qualcomm or EssilorLuxottica, not a second glasses-system competitor.

That split cuts both ways. Samsung Electronics' AI-capability position turns out to be more substantial than the smaller, entity-specific number alone would suggest: approximately 119 families in AR rendering — comparable in scale to Snap (232) and Google (164) — plus real activity across every other AI capability measured, including 129 families in vision/AI, 112 in eye tracking, 105 in gesture, and 90 in wearable form factor.

So Samsung isn't the runaway leader its broadest possible count (1,525 families) implies, and it isn't the AI-absent hardware player a hasty read of last year's coverage maps might have suggested either. It's a legitimate top-six competitor with measurable AI activity across the board — which raises an uncomfortable follow-up. If a portfolio this large can be this misleading, what does it actually mean for a company to compete in the one category every single player here has already entered: optics and display?

Optics and Display Are Table Stakes

Every company in this landscape, whether it sells finished glasses or sells the component layer inside them, shows up in the optics and display classifications. That's not a competitive finding — it's the floor everyone has already cleared.

EssilorLuxottica's dominance of the spectacles-frame classification (G02C) makes the point concretely: 345 families, the highest of any company in the full battleground matrix later in this report, ahead of every glasses brand including Meta.

Frame-and-lens engineering is EssilorLuxottica's actual business, and it shows up here as the one battleground where a supplier, not a product company, leads outright.

Showing up in this classification is the entry fee, not the competitive edge. The real differentiation starts one layer up — in how each company actually got its hardware expertise, and what that choice reveals about how it operates.

Built vs. Bought — and What the Shopping List Actually Shows

Not every company in this set grew its hardware expertise the same way, and the assignee-history record inside each company's hardware-related patent families shows exactly how.

Figure 2. Share of each company's hardware-related patent families traceable to a specific acquired entity, based on assignee-history records.

Even the most aggressive acquirer here built less than 5% of its hardware portfolio through M&A — this is still overwhelmingly an organic-innovation race.

Snap's 2021 moves — acquiring Wave Optics (waveguide-display specialists) and a block of BAE Systems' waveguide IP — put roughly 25 patent families, 4.8% of its hardware-related portfolio, of specialized optics knowledge directly into its holdings in one motion.

BAE Systems is a UK defense contractor; a meaningful slice of the waveguide-optics IP now sitting inside a consumer social app's portfolio began life in military optics research. BAE Systems' filings are treated separately from Snap's own holdings here, since it remains unclear whether this IP is fully attributable to Snap or represents an external collaboration.

Google's acquisition trail is wider, but each individual bet is smaller: at least six distinct acquired specialists — North Inc., Thalmic Labs, Eyefluence, Eye Com Corp, Myvu Corp, MicroOptical Corp, and NVF Technology — contribute a confirmed 40 of 900 hardware-related families (4.4%).

Meta's inorganic share is smaller still: DAQRI, an AR company acquired out of bankruptcy in 2020, contributes just 14 families (0.8%). Oculus VR LLC is deliberately excluded from that count — it was Facebook's internal corporate name for its VR division for roughly a decade after the 2014 acquisition, so patents filed under that name mostly reflect ordinary Meta-directed R&D under a subsidiary's name, not inbound IP from the original startup.

Three different strategies, visible directly in the filing record: Google made the most numerous distinct acquisitions of any company in this set. Snap made the single most concentrated bet. Meta's growth remains overwhelmingly organic.

Acquiring scarce expertise can accelerate capability-building, while a predominantly in-house portfolio can offer greater continuity and control over the pace and direction of technical development — a reasonable inference from the filing pattern itself, not a claim about what either company's leadership actually decided or why.

But acquisitions inside a company's own portfolio are only half of the dependency story. The other half sits entirely outside every glasses maker's balance sheet — with two companies that don't make glasses at all.

The Chokepoint Nobody's Talking About

Two names in this landscape aren't glasses companies at all, and that's exactly why they matter. Qualcomm's validated core estimate is approximately 360 families; EssilorLuxottica's is approximately 530 — chipset/interface IP and optics/lens-integration IP, respectively. EssilorLuxottica is also, not incidentally, Meta's manufacturing partner for Ray-Ban Meta.

Neither supplier is coasting on hardware alone. EssilorLuxottica dominates the wearable/spectacles-form-factor category outright (345 families) and shows real, if modest, activity in perception (24), eye tracking (16), and memory/agent (16). Qualcomm shows meaningful activity across nearly every AI capability measured, led by perception (97) and AR rendering (56).

What isn't yet clear is how directly any of the glasses makers in this report depends on either supplier's specific patented technology, rather than on the general category of component each supplier is known for. What is clear is the scale: EssilorLuxottica and Qualcomm alone hold more combined relevant IP than XREAL, Xiaomi, and Vuzix put together.

If a single supplier with no glasses product of its own can out-patent three consumer brands put together, it raises an obvious next question: who else might be sitting on more relevant IP than their public position suggests — including companies most people wouldn't think to check at all?

The Competitors With No Smart Glasses Product

Ask which companies compete in AI smart glasses, and most lists stop at the seven that actually ship a product. The patent record disagrees. Apple, Amazon, and Microsoft don't sell AI smart glasses today, but their patent activity places two of them — Apple and Microsoft — squarely among the top tier of this entire category.

Apple is top-3 in every single AI battleground measured, and #2 in eye tracking (497 families, behind only Meta's 591) — a portfolio that reads as a fully AI-integrated glasses/headset stack, built on Vision Pro and a string of acquired specialists (Metaio, PrimeSense, Faceshift, SensoMotoric Instruments, Vrvana, and others).

Microsoft is essentially tied with Meta for the AR-rendering battleground (409 families versus Meta's 445) — the HoloLens fingerprint, still visible years later — but is the weakest of the top tier on voice (75 families, behind even Xiaomi's 125).

Amazon is the exception that proves the rule: with approximately 10 validated families, it's a genuinely peripheral player in this landscape. Echo Frames has not translated into meaningful patent depth in either hardware or AI-capability terms — a reminder that shipping a product and building a defensible IP position around it are two different accomplishments, and a company can do one without the other.

None of this means Apple or Microsoft are secretly building AI smart glasses — this report has no evidence either way about intent. It means the assumption that this category's competitive set is fixed at today's product-shipping companies doesn't survive contact with the patent record.

If either company chose to enter, or to license and partner rather than build, they would bring more AI-relevant IP to the table than most of today's actual glasses makers already have. Which leads to a different kind of question: if the competitive set is larger than the product shelf suggests, is everyone in it at least moving at the same pace?

2022 Was the Year the Filing Wave Peaked — Except for Microsoft's

They are not. Peak-filing-year data across all twelve companies surfaces a pattern that a simple headcount would never reveal: this category's activity didn't just happen at different scales — for most companies, it happened on the same clock.

Figure 3. Span from first recorded filing to 2026, with each company's peak filing year marked.

Every top-tier company's filing pace is still climbing or holding steady — except Microsoft's, which peaked years ago.

Meta, Google, Snap, Xiaomi, and XREAL all show their single highest filing year in 2022 — a genuine, shared wave, plausibly tied to the broader industry's AI-glasses moment rather than any one company's individual product cycle. Apple peaks a year later, in 2023, tracking the Vision Pro launch window. EssilorLuxottica and both Samsung entities peak later still, in 2024 — the Ray-Ban Meta and Android XR era.

Microsoft is the one real outlier, and it's worth naming directly, because it cuts against everything the earlier sections would predict: a company that ranks #3 overall and is nearly tied with Meta on AR rendering peaked in 2017 (135 families, the HoloLens 1/2 era) and has been cooling ever since.

Paired with Microsoft also carrying the highest granted-only rate in the entire landscape (see Portfolio Maturity below), the picture is consistent, not coincidental: this is a mature, largely-examined portfolio built on a product cycle that peaked years ago, not a company still accelerating.

Microsoft may be de-prioritizing consumer AR while Meta, Apple, and Samsung double down. A company can be large, technically strong, and still be a company whose best filing years are behind it — which raises one more question a scale-only view can't answer: does the reach of that protection match the reach of the products themselves?

The Geographic Gap Behind Global Products

Global product launches suggest global protection. Jurisdictional data says that assumption doesn't hold evenly across this landscape — it shows how far each company's patent protection actually reaches, not just where it was originally filed.

XREAL is the clearest case: of its 215 total families, 211 (98%) have their earliest priority filed in China. But the same 215 families also reach further than that headline suggests — 33 are also filed in the US, 23 in Europe, 6 in Japan, 3 in Korea, and 43 through the WO/PCT international system. The China-heavy profile is real. Reading it as an absence of international protection would be wrong.

A few other cross-company patterns stand out once the full table is in view. Google (1,196 WO/PCT filings, 57% of its portfolio) and Microsoft (938, 74%) file internationally more consistently than any other company here — a genuine Silicon Valley/West Coast R&D footprint extended broadly abroad.

Samsung Electronics is the only major Korea-priority filer (49% of its families originate there) despite reaching deep into the US and China as well. Xiaomi (97% China priority) and Samsung Display (51% China priority, reflecting Chinese OLED-panel R&D) round out the China-heavy group alongside XREAL.

EssilorLuxottica is the most Europe-weighted portfolio in the set (42% EP priority), as expected for a French/Italian group. Knowing where a company protects its inventions still doesn't answer what, specifically, it's protecting — which is the most direct question this data can answer, and the one the title of this report asks outright.

Who's Actually Building the AI: A Battleground-by-Battleground Look

Here is the most direct answer the data can give. Every company in this landscape has been measured against the same eight AI capability areas: AR rendering, voice, perception, gesture, eye tracking, privacy, memory/agent, and wearable form factor.

Figure 5. Patent family counts by company and AI battleground. Shading is relative within each column; darker means a stronger relative position in that specific battleground, not a larger company overall.

No company leads everywhere — the shape of each row is a strategy fingerprint.

No single company leads everywhere, and the shape of each portfolio tracks its public strategy closely enough to work as a strategic fingerprint, not just a patent count.

Meta leads outright in AR rendering, gesture, eye tracking, perception, and wearable form factor — a portfolio shape that lines up cleanly with the public Ray-Ban Meta and Orion strategy.

Google leads voice by a wide margin (898 families, more than double Apple's second-place 244) and memory/agent (271) — its Assistant and Gemini heritage showing up directly in the patent record — but is comparatively weaker in AR rendering. Microsoft is essentially co-leader with Meta in AR rendering (409 versus 445) but is the weakest top-tier player on voice.

The absences matter as much as the standouts. Vuzix shows zero families in perception and zero in memory/agent — the clearest evidence in this landscape that its portfolio remains enterprise-optics-focused rather than AI-forward.

Xiaomi is voice-heavy (125 families, third-highest in the set) but nearly absent from the wearable/spectacles-form-factor battleground (13 families) — a pattern that reads as a smart-audio and vehicle-AI company more than a dedicated smart-glasses program, whatever its product roadmap says.

A portfolio's shape reveals a company's actual bet more reliably than its announcements do — but even a well-shaped portfolio is only half the picture. The other half is how far along it actually is.

Portfolio Size Isn't Portfolio Maturity

A large patent count says nothing about how far along that portfolio actually is — how much of it has cleared examination, how much is still pending, and how much has already lapsed. Two companies with similar family counts can be in entirely different stages of the same race. Patent family-level legal-status records make the maturity picture concrete.

Figure 6. Legal status composition by company: granted-only, mixed (granted and pending members), pending-only, and inactive (lapsed, withdrawn, or expired), as a share of each company's total family count.

Microsoft's portfolio has already finished the race Samsung Display just entered.

Microsoft has the highest granted-only rate in the landscape (62% of 1,263 families) and the lowest pending-only rate (4%) — consistent with a mature portfolio dating largely from the HoloLens 1/2 era, and consistent with the 2017 peak-year finding above: this is a company whose AI-glasses-relevant IP has mostly finished moving through examination, not one still building out a pipeline.

Samsung Display sits at the opposite extreme: only 15% granted-only and 70% pending-only, the direct fingerprint of the concentrated 2024 filing surge that hasn't had time to clear examination yet.

Snap's portfolio is the freshest among product companies — only 2% inactive and the highest "mixed" share (61%) in the set, meaning it's actively converting pending applications into granted patents right now.

EssilorLuxottica carries the highest inactive rate (31%), the expected result of a portfolio with real filing activity dating back to 1976 — patents eventually expire.

Put the maturity data next to everything else in this report, and the same company keeps reappearing on the wrong side of the trend: Microsoft ranks #3 overall, holds AR-rendering IP on par with Meta, and is nonetheless the one major player whose portfolio looks finished rather than in progress. Scale, capability, and momentum turn out to be three separate questions — and a company can score high on the first two while quietly losing the third.

Areas for Further Analysis

The findings above are well-supported by the data collected. A small number of areas would benefit from additional analysis to sharpen specific estimates further — none of them change the direction of any finding in this report.

• EssilorLuxottica / Qualcomm — EssilorLuxottica's and Qualcomm's estimates (~530 and ~360) are based on assignee-pattern analysis; a more granular relevance review would tighten these figures further, though both companies' highly specialized patent portfolios suggest the current estimates are directionally reliable.
• Google — Google's AI-capability estimate (~750 families) includes some generic voice-assistant activity not specific to wearables; a more tightly scoped search focused on wearable-context AI would refine this figure further.
• XREAL — XREAL's total includes several Chinese entities (Matrixed Reality Technology, Shanyaoxianshi (Wuxi) Technology, and Beijing Unicorn Technology) whose corporate affiliation with XREAL would benefit from further confirmation; if any are unaffiliated, XREAL's total would be modestly lower.
• Meta + EssilorLuxottica — Additional analysis of joint or co-assigned filings between Meta and EssilorLuxottica would help quantify the depth of the market's most visible smart-glasses partnership at the patent level.
• Snap + BAE Systems — BAE Systems' waveguide IP is treated separately from Snap's own portfolio in this report; further analysis of current assignee records would clarify whether this technology should be attributed to Snap directly or remains an external collaboration.

What This Means: Questions Worth Asking About Any Company in This Category

So, who's actually building the AI in AI smart glasses? Not simply "Meta" — though Meta's lead is real and well-earned. The fuller answer is that the hardware race has a clear leader, the AI race now has real, comparable data behind it for every company measured, and two of the most AI-capable portfolios in the entire category belong to companies with no shipping product in it at all.

For a reader inside this industry, the practical value isn't the leaderboard itself — it's a short list of questions this analysis suggests are worth asking about any company in this space, including competitors and partners not covered here:

1. Is a large patent count real, or does it depend on how the count was assembled — which filing entities, which classifications, which time window?
2. Was a technical capability built in-house or acquired — and if acquired, how many distinct deals does that actually represent?
3. Does the patent protection actually extend to the markets the product ships into, or only to the market it's made in?
4. Is the portfolio concentrated around table-stakes hardware, or does it show measured AI-capability activity across multiple battlegrounds?
5. Where does the company depend on suppliers — chips, optics, manufacturing — and what does that mean for its negotiating leverage?
6. Is the portfolio still accelerating, or has its filing pace already peaked and cooled — and what does that suggest about current strategic priority?
7. Is a company outside today's competitive set — a Big Tech player with no shipping product — sitting on more relevant IP than the companies already in the race?

This report makes that comparison possible for every company in the category, including the two — Apple and Microsoft — that most competitors would not have thought to check. Someone who followed only product announcements and industry news would have missed both of them entirely, along with the chokepoint suppliers sitting underneath the whole category.

Today, the product shelf and the patent office are telling two different stories about this category. The more interesting question is which one is right in three years — whether Microsoft's cooling filing pace and Apple's silent buildup turn out to be a company sitting out the race or a company waiting for the right moment to join it, and whether the category's next real surge shows up on a product shelf before it shows up here, in the patent record, first.

The Signal Doesn't Stop When This Report Does

Everything in this report is a snapshot — patent activity as it stands today, across twelve companies, at one moment in time. The more useful read is what happens between snapshots.

A company's AR-rendering filings accelerating quarter over quarter, a supplier's patents suddenly reaching into a battleground it never touched before, a competitor's granted-rate climbing while its filing pace quietly cools — each of those is a strategic signal, and each one tends to show up in the patent record months or years before it shows up in a product announcement, an earnings call, or a press release.

That gap between what a company is patenting and what it's publicly saying is exactly where IP strategy lives.

Knowing where a competitor is filing defensively, where freedom-to-operate risk is building before a product ships into a new market, or where a supplier relationship is quietly creating more IP exposure than it appears to on the surface isn't incidental information — it's the kind of intelligence that shapes real decisions: where to file, where to license, where to watch, and where to move first.

It's also why sophisticated technology companies, investors, and corporate development teams already treat patent-landscape monitoring as an ongoing practice rather than a one-time report: a single read tells you where things stand, but only a continuous one tells you where they're heading.

The companies that watch this signal continuously are the ones who catch a competitor's pivot, a supplier's shift, or an emerging white space while there's still time to act on it. The ones that don't are left reading the product shelf for news the patent office already had — sometimes years earlier.

Portfolio Reference (Patent Families)

For reference — not the headline of the story, per the findings above.