A guide for founders and their attorneys, written with genuine concern for your financial mortality.
Here is what nobody tells you at demo day:
The device on your face — the one that renders photons into a spatial illusion so convincing that your vestibular system files a formal grievance — is, from a legal standpoint, not really a device at all. It is a territory. And the question of who gets to build things in that territory, and who gets sued for trespassing, and who gets to charge rent for the privilege of rendering a virtual cube that hovers convincingly above a conference table — this question is being decided right now, in patent offices and federal courtrooms, by people who may or may not understand what a waveguide does, and the answer will shape your company's future more decisively than your next funding round.
The global AR/VR market is on pace to exceed $100 billion by 2026. Apple's Vision Pro validated spatial computing as a consumer product category in 2024, which had the effect of a starting pistol fired in a room full of patent attorneys. NPE litigation targeting AR/VR companies jumped 47% year-over-year in Q1 2025. If you are building anything that overlays digital information onto the physical world, or replaces the physical world entirely, or does something weird and in-between that you're calling "mixed reality" because your marketing team liked the sound of it — you need a patent strategy. Not eventually. Now.
This post is that strategy.
I. The Battlefield, Which Is Already Quite Full of Other People's Flags
Before you can decide what to patent, you need to understand — with the kind of granular, slightly nauseating specificity that makes this stuff actually useful — who already owns what.
The short answer is: enormous companies own enormous amounts.
Magic Leap holds roughly 8,555 total patents and applications (5,303 granted) across 1,062 patent families — which is remarkable for a company whose consumer product was, charitably, a commercial disappointment.[^1] Microsoft holds over 10,000 AR/VR patents and leads in forward citations and classification breadth, meaning its patents are not merely numerous but influential, even after it discontinued HoloLens and handed the $22 billion U.S. Army AR program to Anduril in 2025. Samsung — the overall USPTO leader with 10,427 granted patents in 2024 — dominates display technology and filed over 4,094 VR/AR applications in 2020–2021 alone. Sony holds approximately 3,612 published VR/AR patents.
[^1]: Magic Leap's filing volume peaked in 2018, which was also the year they shipped the Magic Leap One to reviews that ranged from "underwhelming" to "I paid $2,295 for this?" The patents, however, are doing just fine. Patents don't need product-market fit. This is one of the many things about the patent system that is either brilliant or insane depending on your relationship to it.
Meta presents a fascinating case study in the imperfect correlation between money and intellectual property. The company invested $43 billion in R&D in 2024 and maintains over 650 active AR-related filings, yet according to IEEE Spectrum, it ranks "surprisingly low" in patent power relative to its spending. Mark Zuckerberg has described Meta's neural wristband — built on the $1 billion CTRL-Labs acquisition — as shipping "in the next few years," which is the kind of timeline that means the patent filings are already in the system even if the product isn't on your wrist. Fenwick & West manages the portfolio. The Orion AR glasses, meanwhile, remain unreleased, their patent filings visible but the product itself a kind of legal ghost — real enough to sue over, too expensive to manufacture.
And then there's Apple. Apple's entry with Vision Pro accelerated its VR publication growth rate to 19.1% annually — more than triple the sector average of 5.8%. The company now ranks in the top 15 VR hardware patent holders, covering mixed-reality displays, eye and gaze tracking, iris-based biometric authentication (they call it Optic ID, which is the kind of name that sounds like it was chosen by the same committee that decided "spatial computing" was preferable to "VR headset"), spatial audio, and hand gesture recognition.
Here is something important about geography: China dominates in raw filing volume, with 1,213 AR/VR patents issued per quarter from CNIPA, driven by Tencent (4,085 applications), Baidu (3,094), and ByteDance/PICO. The United States accounts for over 55% of global AR/VR patent applications. Asia now represents 68.7% of all global patent filings and 56.3% of PCT applications — up from 40.6% a decade ago. South Korea ranks third (Samsung and LG), Japan fourth (Sony), Europe fifth.
And here is something slightly weird: VR patent grants at the USPTO actually declined from 57,252 in 2023 to 42,559 in 2024 — a 26% drop. But AI patent grants rose to 54,022 in the same period. The innovation didn't slow down. It migrated. AR/VR inventions with AI components are increasingly classified under AI patent codes, because the two fields have become so entangled that the classification system itself can't quite keep up. Which tells you something about where the next wave of filings needs to go.
II. What to Patent: A Layer Cake of Increasingly Abstract Things
The most defensible AR/VR patent portfolios provide coverage across multiple technology layers — hardware, software, interaction, networking, applications — such that attacking any one patent is like pulling a thread that's woven into the entire fabric. Filing across only one layer is like building a house with only walls and no roof: technically a structure, but not one you'd want to be inside when it rains litigation.
Hardware: Where Patents Are Simplest and Strongest
Hardware patents face the fewest Section 101 challenges, which — for those of you who haven't yet had the pleasure — is the section of U.S. patent law that asks whether your invention is actually an invention or merely an "abstract idea" that a computer happens to execute.[^2] Hardware creates physical barriers to entry. Hardware is tangible. Patent examiners understand tangible.
[^2]: We will discuss Section 101 at length later. It is the part of patent law that keeps AR/VR patent attorneys up at night, staring at the ceiling, wondering whether the Federal Circuit will, next month, issue a decision that retroactively invalidates half their portfolio.
Optics remains the most heavily patented hardware subsystem, and for good reason — the optics are what make the magic possible. Diffractive waveguides (used by HoloLens and DigiLens) and reflective waveguides (Lumus's Z-Lens, which claims 5–10x better optical efficiency) are both active battlegrounds. Metasurfaces and freeform optics represent emerging frontiers: ACS Applied Optical Materials published 2024 research on broadband achromatic metalenses and dynamic metasurface spatial light modulators. These are new claim territories. If you are doing novel work in optics, this is where a smart provisional application could establish a priority date in terrain that isn't yet dense with prior art.
In micro-displays, micro-LED is emerging as the winner for AR because AR demands brightness that OLED cannot deliver. JBD (Jade Bird Display) leads with 10,160 PPI displays at 480 Hz. Samsung Display has confirmed micro-LEDs will likely prevail over OLED for AR applications. Micro-OLED holds advantages in contrast for VR headsets. Patent claims covering specific display architectures, pixel structures, and brightness-enhancement methods are, as a general matter, quite defensible, because they describe specific physical structures rather than abstract computational processes that an examiner can characterize as "something a person could do with a pencil."
Sensors — LiDAR, depth sensors, IMUs, cameras, eye-tracking hardware — are the quiet workhorses. Microsoft's 2025 patent for AR glasses features a novel thermal and depth sensor array optimized for low-light environments. Meta holds patents on light-field camera-based eye tracking. Sensor patents serve a dual function: they're valuable on their own, and they anchor the software claims that depend on them. Remember that.
Software and Algorithms: The Beautiful, Treacherous Heart of It
Software patents are where the most innovation occurs in AR/VR. They are also where the most patents die.
The key areas: foveated rendering (rendering high resolution only where the user's gaze falls, which saves enormous computational resources and is the kind of insight that seems obvious in retrospect but wasn't), SLAM algorithms (simultaneous localization and mapping — how a device understands where it is in physical space), computer vision and scene understanding, gesture recognition, and eye-tracking algorithms. Meta was granted US 11,215,837 in January 2025, covering eye tracking integrated with beam-steering optics that dynamically redirect light based on gaze — a patent that links a software process to a specific physical mechanism, which is exactly the kind of claim structure that survives eligibility challenges.
We'll deal with Section 101 strategy in its own section below, because it deserves the space. For now, the essential principle: every AR/VR software claim must be tethered to specific hardware, must recite concrete technical improvements with measurable outcomes (reduced latency, improved accuracy, lower power consumption), and must describe specific data processing steps beyond generic computer functions. Vague, functional language is death. The Federal Circuit said so explicitly.
Interaction, Networking, and the Farther Reaches
Haptics represent a $16 billion market by 2032, and the patentable frontiers include dielectric elastomer actuator arrays, EMG-based wrist interfaces, and multimodal haptic rings. Skin-attached haptic patches and haptic-feedback rings are the subjects of active academic research that will become commercial products within years. Claims combining multiple input modalities — gaze plus gesture plus voice with specific interaction logic — create the kind of multi-layered protection that is genuinely difficult to design around.
Networking patents may be the most underappreciated category. Split rendering architectures — methods for dividing rendering between device and edge/cloud — are essential for lightweight AR glasses that offload compute. Adaptive foveated rendering offloaded to edge servers via deep reinforcement learning is an active research area. Multi-user AR coordination platforms using edge cloud architectures are generating patentable innovations in session management, latency compensation, and spatial anchor synchronization.
Digital twins — virtual replicas of physical systems — have seen filings exceed 1,600 publications in 2023. But digital twin patents face unique Section 101 risks because they rely heavily on modeling and data analysis. The survival strategy: claim the feedback loop implementation and specific physical-world behavioral changes, not abstract data processing. If your digital twin patent reads like a math paper, you're going to have a bad time at the USPTO.
III. Timing, Money, and the Art of Filing Patents When You Can't Afford to File Patents
Here is the part where I tell you how much things cost, because nobody else seems to want to.
Empirical data reveals that 56% of startups that eventually patent file their first application before raising a seed or Series A round. In patent-intensive industries — and spatial computing is nothing if not patent-intensive — that figure exceeds 60%. The relationship between patents and startup survival depends less on portfolio size than on having at least one strong, well-timed patent aligned with core technology.
Provisional patent applications cost $65–$325 in USPTO fees (micro to large entity) plus $4,000–$7,000 with attorney assistance. They establish a priority date. They give you 12 months to refine the technology. They let you say "patent pending" to investors, which, if we're being honest about the sociology of venture capital, is worth more than its legal weight. Non-provisional applications typically run $10,000–$20,000 for complex AR/VR inventions, with another $3,000–$6,000 per office action response.
Continuation applications are the thing that separates sophisticated patent strategy from expensive paper collection. They keep patent families "open," allowing claims to adapt to competitive threats, business pivots, and changes in case law without losing the original priority date. A strategy that works: file an initial application with modest claims to get quick issuance (useful for fundraising), then pursue broader claims through continuations as budget allows. Note: the USPTO introduced surcharges in January 2025 for continuations with priority claims more than six or nine years old. The message is: file your continuations in a timely fashion, or pay extra for the privilege of procrastination.
For international protection, file a PCT application within 12 months of the provisional. The PCT provides up to 30 months from the initial filing date to decide where to enter national phase — which is the international patent system's way of saying "we understand you're a startup and you don't know if you'll have money in 30 months." Focus on three to five jurisdictions: the United States, China (key manufacturing hub, largest raw filing volume), the EPO (the Unitary Patent system now offers streamlined multi-country European coverage), South Korea, and Japan. A practical heuristic: file in markets that will capture at least two-thirds of future product revenues.
What to spend, and when
Pre-seed and seed: $20,000–$25,000 for year one. One to three provisionals covering core technology plus reserves for office action responses. Even one well-drafted patent can meaningfully change fundraising dynamics, and this is not aspirational language — it's empirical.
Series A: Three to five filings (mix of provisionals and non-provisionals), a PCT application, and the beginning of a continuation strategy.
Series B and beyond: Five to fifteen-plus patents through continuations, divisionals, and international filings. Consider defensive publications for non-core innovations — you don't want to patent everything, but you definitely don't want anyone else patenting it either.
Leverage entity fee discounts aggressively. Small entities pay 40% of standard fees; micro entities pay 20%. Total lifecycle cost per U.S. patent runs approximately $20,000–$25,000 including maintenance fees at 3.5, 7.5, and 11.5 years. In the context of a technology company's operating budget, this is inexpensive insurance against existential risk.
IV. The Litigation Landscape, or, Everyone Is Suing Everyone
Listen. I don't want to alarm you. But the AR/VR patent litigation environment has become something that, if it were a weather system, meteorologists would give a name to.
Patent case filings jumped approximately 22% in 2024, with over $4.3 billion in damages awarded — the highest annual total in history. NPE filings climbed 21.6%, and NPEs added 608 new defendants in Q1 2025 alone. If you are selling an AR/VR product, someone is thinking about suing you. If you are not selling an AR/VR product but you've filed patents in the space, someone is thinking about buying your patents and then suing someone else.
IngenioSpec LLC — formerly SmartIGlasses LLC, a name change that tells you something about the theater of patent enforcement — has emerged as the most aggressive NPE in spatial computing. It wields a 72-patent portfolio issued between 2019 and 2024. Its first ITC investigation (November 2023) targeted 13 respondents including Meta, Magic Leap, Vuzix, and XREAL over "electronic sunglasses." It settled with Meta; the investigation wound down in September 2024. A second ITC investigation (October 2024) targeted ByteDance/PICO, HTC, Meta, and Valve over AR/VR headsets. Meta responded with a declaratory judgment action in Northern District of California. IngenioSpec simultaneously filed district court suits against Bose, LG, Samsung, and Sony in Eastern District of Texas, which — and this is important — has reclaimed its position as the top patent venue nationally, with over 1,000 new patent suits filed in 2024.
Immersion Corporation, holding over 1,200 haptics patents and previously the beneficiary of a $150 million judgment against Sony over DualShock controllers, settled with Meta in February 2024 after asserting eleven haptics patents against Quest 2 and Quest 3. Meta received a non-exclusive, worldwide, perpetual license covering all current and future products — which is legal language for "we paid enough money to make this problem go away permanently." Immersion's ongoing suit against Valve (targeting Steam Deck and Valve Index) underscores a simple truth: if your VR device vibrates, someone probably holds a patent on the specific way it vibrates.
Among operating companies, XREAL filed against Viture in November 2025 over AR optical system patents, simultaneously in Eastern District of Texas and Munich, where the Munich Regional Court issued a preliminary injunction. This is a startup-versus-startup fight with parallel U.S. and European proceedings — which tells you that patent warfare in AR/VR is no longer the exclusive province of billion-dollar companies. Apple sued a former Vision Pro engineer in June 2025 for allegedly taking trade secrets to Snap, illustrating that the competitive intensity around AR talent and proprietary technology has reached the "personal lawsuit" tier.
Defense costs deserve their own paragraph of respectful silence. IPR proceedings run $200,000–$500,000. District court litigation costs $2–$10 million through trial. ITC investigations run $3–$5 million or more. A modest patent portfolio doesn't eliminate these risks, but it creates cross-licensing leverage and mutual deterrence that a company with no patents simply does not have. This is, if you want to be blunt about it, the main reason startups need patents: not to sue people, but to make it expensive for people to sue you.
V. Section 101: The Existential Question of Whether Your Software Invention Is an Invention
Under 35 U.S.C. § 101, a patent must claim patentable subject matter. Since the Supreme Court's 2014 Alice decision, software patents have faced a two-step inquiry: (1) Is the claim directed to an "abstract idea"? (2) If so, does it contain an "inventive concept" that transforms it into something patent-eligible? This framework has been described by patent attorneys as "workable," "unpredictable," "maddening," and, less diplomatically, by words I shouldn't reproduce here.
For AR/VR, three categories of abstract ideas are the recurring threats: mathematical concepts (fundamental to AR/VR data manipulation), mental processes (anything an examiner can characterize as "something a person could theoretically do in their head," which, for AR/VR, is almost always a stretch but often attempted), and methods of organizing human activity.
Here is the good news, and it is genuinely good:
The August 2025 USPTO memorandum from the Deputy Commissioner for Patents represents the most significant shift in eligibility guidance in years. It instructs examiners that "mental process" means mental process — if a claim limitation cannot practically be performed in the human mind, it shouldn't be rejected under that grouping. Most AR/VR computations — real-time rendering, SLAM, eye-tracking inference — obviously cannot be performed in anyone's mind, no matter how gifted. The memo also states that close calls are not rejections. Examiners should reject only when ineligibility is more likely than not.
The July 2024 AI Guidance Update introduced three detailed examples for AI/ML patent eligibility analysis. Generative AI applications in Tech Center 2100 are seeing a 79% allowance rate. Tech Centers 2400 and 2600 achieve 90–97% allowance rates. If your AR/VR claims land in these Tech Centers — rather than in Tech Center 3600 (business methods, 63% allowance) — your odds improve dramatically. This is not a matter of luck; it's a matter of claim drafting.
Key case law you need to understand: Contour IP v. GoPro (Fed. Cir. 2024) reversed summary judgment of ineligibility for patents on portable cameras generating video in two simultaneous formats — positive precedent for AR/VR claims tied to specific hardware capabilities. Conversely, Mobile Acuity v. Blippar (Fed. Cir. 2024) affirmed invalidation of AR-like image recognition patents whose claims consisted solely of result-oriented, functional language. The Federal Circuit's April 2025 decision in Recentive Analytics v. Fox Corp. now controls for AI/ML claims: applying established ML methods to a new data environment, without disclosing improvements to the ML models themselves, is not enough. Every AI/ML claim in your AR/VR portfolio must articulate the specific technical improvement to the model — not just its application to AR/VR data.
The pending Patent Eligibility Restoration Act (S. 1546, H.R. 3152) could fundamentally reshape this landscape by abrogating Alice. Passage remains uncertain. Draft your claims as if Alice will persist, and celebrate if it doesn't.
VI. Five Mistakes That Will Cost You Real Money and Possibly Your Company
One: Filing too late. The United States is a first-to-file system. This means exactly what it sounds like. Once you share technology with customers or partners or demo it at CES, competitors can start patenting similar innovations, and your "but we thought of it first" will be legally meaningless. File provisional applications as soon as core technical concepts are clear.
Two: Patenting only one layer of the stack. Your AR device involves optics, sensors, algorithms, interaction paradigms, and networking. Protecting only the algorithm is like locking the front door and leaving every window open.
Three: Ignoring freedom-to-operate. This is the mistake that can actually kill you. Filing your own patent does not grant you the right to practice your technology — it only grants you the right to stop others from practicing yours. The AR/VR landscape is dense enough that FTO searches are essential before product launch or major funding rounds. You might hold a beautiful patent on your rendering pipeline and still be infringing three of Microsoft's 10,000 patents every time you turn the device on.
Four: Closing patent families prematurely. If you fail to file continuations before a parent patent issues, you permanently forfeit the ability to pursue additional claims from the same priority date. This is the kind of procedural mistake that cannot be undone and that someone will eventually describe, in a board meeting, as "regrettable."
Five: Treating patents as legal footnotes rather than strategic assets. The best-positioned startups include jurisdiction maps, claim coverage summaries, and licensing potential in investor presentations. A well-constructed patent portfolio is not an expense line. It is a strategic asset that affects valuation, acquirability, and competitive positioning at every stage.
When budget is limited — and budget is always limited — prioritize patents on core differentiating technology that competitors cannot easily design around. Use a platform application strategy: draft a comprehensive initial patent covering core technology with room for future iterations, then branch off additional patents through continuations. File defensive publications for non-core innovations to prevent others from patenting them. And consider the trade secret alternative for backend algorithms, proprietary training data, and manufacturing processes that aren't reverse-engineerable from the product — but understand that anything a competitor could independently discover and patent must be patented, lest you find yourself locked out of your own technology.
VII. The Frontier: Where Smart Founders Are Filing Before Everyone Else
The next generation of high-value AR/VR patents is coalescing around convergence areas where the landscape is not yet impenetrable. If you are founding a company in these spaces, the priority-date clock matters more than usual.
Neural interfaces. Meta's EMG wristband, built on its $1 billion CTRL-Labs acquisition, demonstrated in a July 2025 Nature paper that wrist-based EMG can achieve handwriting input at 20.9 words per minute and gesture detection at 0.88 gestures per second. The ML models that translate EMG signals into precise commands are highly patentable, and the space remains relatively uncrowded compared to optical or display patents. If your company is doing novel work in neural input — EMG, EEG, or anything that reads bioelectric signals and translates them into spatial computing commands — this is a filing window, not a filing opportunity. Windows close.
AI-powered rendering. 3D Gaussian Splatting, introduced at SIGGRAPH 2023, achieved real-time rendering at over 100 fps at 1080p. VR-Splatting (ACM 2024/2025) combines neural point rendering with Gaussian Splatting in a hybrid foveated approach, hitting 90 Hz in VR at 2016×2240 per eye. Patentable innovations include novel scene compression methods, real-time editing of radiance fields in VR, text-to-3D generation pipelines, and mobile-optimized neural radiance field inference. The overlap between generative AI and real-time rendering for AR/VR is producing innovations faster than the prior art landscape can absorb them. This is good news for founders with something novel to claim.
Standards-essential patents. While OpenXR operates royalty-free (no traditional SEP licensing), MPEG-I (ISO/IEC 23090) is generating significant SEP potential across immersive video, point cloud compression, and immersive audio. Nokia is explicitly committed to licensing MPEG-I SEPs under RAND terms. AR/VR devices also face cumulative royalty exposure from 5G/Wi-Fi connectivity SEPs (87,000+ 5G patents declared essential to ETSI), video codec SEPs (HEVC, VVC pools), and display interface SEPs. If you're building a connected device, budget for these licensing costs in your device pricing model, or be surprised later. Nobody likes being surprised by patent royalties.
Active participation in standards bodies — Khronos Group (OpenXR), MPEG, 3GPP, IEEE — gives companies the ability to shape technical requirements that align with their innovations and to file patents on proposed contributions before standards finalize.
VIII. What to Do on Monday Morning
Three things are true simultaneously:
First, the empirical data is unambiguous. Startups with patents raise capital at 93% higher valuations at the angel stage, are twice as likely to be acquired, and more than double their probability of IPO. A well-constructed AR/VR patent portfolio's signal value — what it communicates to investors, acquirers, and competitors about the seriousness and defensibility of your technical position — exceeds its direct legal protection, though the direct legal protection is also quite real.
Second, the convergence of AI and AR/VR is simultaneously the biggest patent opportunity and the biggest eligibility risk. The August 2025 USPTO memo and Recentive Analytics decision create a clear framework: claims that articulate specific technical improvements to ML models survive scrutiny. Claims that apply generic ML to a new data environment do not. Draft accordingly.
Third, the litigation threat is real and growing. IngenioSpec's multi-front campaign against more than 15 companies demonstrates that no AR/VR company is too small to be named as a defendant. Building even a modest portfolio creates cross-licensing leverage and mutual deterrence that bare-handed companies simply do not have.
So: file provisionals covering your core technology layers. Maintain open continuation families. Preserve international options through PCT. Link every software claim to specific hardware and measurable technical improvements. Budget for FTO searches before you launch. And treat your patent portfolio as what it actually is: not a cost center, not a box-checking exercise, but the legal architecture of your competitive position in a market where the territory is imaginary and the stakes are not.