GPU Architecture Patents: Who Leads, Where the Gaps Are 2026
- Filings have cooled since a 2018 peak of 146, with the 2022 midpoint at 73 — a signal that core claim space is filling in rather than expanding.
- Image data processing (G06T) touches 1,024 of 1,304 records, far outweighing the 145 that also carry an AI-computing IPC — the two disciplines still file mostly apart.
- Momentum is fading even among the largest filers, with the most active assignee down 33% year-on-year and several established names at zero in the latest year.
Filing growth compares 2021 (123 records) with 2024 (76) — a three-year span. 2024 is the most recent year we treat as complete: publication lags filing by roughly 18 months, so 2025 onwards are still filling in and any growth rate that ends there would understate the field. Top-5 share is the combined record count of the five largest assignees divided by all 1,304 records in scope (CR5), not by the ranked leaders only.
What this landscape covers
This landscape tracks 1,304 patent families filed against GPU architecture and parallel compute — spanning warp scheduling, memory bandwidth management, shader core design, cache optimization, and workload scheduling — classified under G06F15, G06F9, and G06T1. The window runs from 2015 through the 2026 data cut-off, though the most recent one to two years are understated because publication typically lags filing by around 18 months.
Filing activity rose through the mid-2010s, peaked in 2018, and has since declined toward the level seen at the start of the window. That pattern does not mean the field is finished — it means the foundational claim space around scheduling and memory access has largely been staked out, and new filings increasingly have to work around it rather than into open ground.
Filing trends and technology composition
Two views of the same corpus: the year-by-year filing count, and the IPC subclasses that carry the claims.
Filings rose to a 2018 peak, then declined
Annual filings moved from 94 in 2017 to a peak of 146 in 2018, back down to 73 by the 2022 midpoint and 6 in the most recent (partial) year. Read the tail end cautiously — publication lag means the last one to two years will fill in further as pending applications publish.
G06T and G06F dominate; AI-computing overlap is still thin
Image data processing (G06T, 1,024 records) and general digital data processing (G06F, 817) anchor the corpus, with display control (G09G, 97), video/TV pictorial communication (H04N, 48), and coding/conversion (H03M, 29) forming a smaller supporting band. Only 145 records also sit in G06N, the AI-computing class — most GPU architecture claims are still written independently of machine-learning-specific framing.
Shares are the percentage of the 1,304 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on GPU Architecture and Parallel Compute with Eureka
This page is one run against one query. Ask Eureka your own question about gpu architecture and parallel compute and every answer comes back with the patent numbers behind it.
Try EurekaThe most-cited records in this corpus
Graphics Processing Unit Performance Analysis Tool — Apple Inc.
Systems, methods, and computer readable media to analyze and improve the performance of applications utilizing graphics hardware. The disclosure covers monitoring run-time performance of shader programs from multiple concurrently executing applications on a GPU and presenting a visualization to the user, with profiling built from sampling multiple hardware performance counters and shader programs during execution.Filed by Apple Inc., published 2020-12-03 as US20200379864A1.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20090002379A1 | Video decoding implementations for a graphics processing unit | 466 |
| 2 | US20150002508A1 | Unique primitive identifier generation | 160 |
| 3 | US20190205746A1 | Machine learning sparse computation mechanism for arbitrary neural networks, arithmetic compute microarchitec… | 158 |
| 4 | US9177413B2 | Unique primitive identifier generation | 139 |
| 5 | US20180322607A1 | Dynamic precision management for integer deep learning primitives | 132 |
| 6 | US6452595B1 | Integrated graphics processing unit with antialiasing | 122 |
| 7 | US20120320070A1 | Memory sharing in graphics processing unit | 117 |
| 8 | US20180315399A1 | Instructions and logic to perform floating-point and integer operations for machine learning | 111 |
| 9 | US20180315159A1 | Compute optimizations for low precision machine learning operations | 104 |
| 10 | US20140267259A1 | Tile-based rendering | 101 |
Citation counts favour older filings that have had more time to accumulate references — read them as a measure of influence within this searched corpus, not as a ranking of current technical importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the numbers mean for strategy
Four read-outs from the filing trend, IPC spread, and citation data — each pointing to a different practical decision.
Growth has flattened, not accelerated
The 2018 peak of 146 filings has not been matched since; the 2022 midpoint of 73 and a thin latest-year count show a corpus past its filing crest. New entrants are more likely to be working around existing claims than filing into open space.
Image processing dominates over AI-specific framing
Image data processing claims outnumber AI-computing overlap by close to seven to one. Architecture work here is still largely graphics-first; teams building ML-specific GPU claims are filing into a comparatively less crowded intersection.
Influence is concentrated in a handful of early filings
The most-cited record predates the rest of the top group by a wide margin, and citation counts drop off quickly after the top few entries. That concentration reflects age and searched-corpus visibility as much as ongoing relevance.
Even the most active filer is pulling back
The assignee with the most recent-year filings is still down a third year-on-year, and several other major names show no filings at all in the latest year. That pattern is consistent with a maturing claim landscape rather than a collapsing one.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to gpu architecture and parallel compute, with the prior art for and against each one.
Who is filing, and where the gaps sit
Assignee activity is uneven: a small group of large filers dominate cumulative counts, but recent-year momentum has slowed across nearly all of them, leaving room to ask where the next wave of claims might land.
NVIDIA Corporation (NVIDIA) still files the most recently, but less than before
NVIDIA remains the most active filer in the latest year tracked, yet its own filing count is down a third year-on-year — a pullback from a much larger prior base rather than a sign of retreat from the field.
Qualcomm Incorporated (Qualcomm) shows a sharper drop
Qualcomm's latest-year count has fallen further than NVIDIA's on a percentage basis, consistent with a broader slowdown across the largest historical filers rather than an isolated event.
Several long-standing filers show no latest-year activity
Intel, Samsung, and ARM each register zero filings in the most recent year, each down 100% year-on-year. That does not mean these programmes have stopped — it more likely reflects filings still working through the publication pipeline.
Co-filing is rare and concentrated in legacy pairings
Only ten co-assignee pairs appear in the corpus, and the strongest link reflects AMD's historical ATI acquisition rather than a new joint-venture pattern. Cross-company collaboration on GPU architecture claims is the exception, not the norm.
| Assignee | Recent year | YoY |
|---|---|---|
| NVIDIA Corporation | 2 | -33% |
| Qualcomm Incorporated | 1 | -80% |
| Intel Corporation | 0 | -100% |
| Advanced Micro Devices, Inc. (AMD) | 0 | — |
| Samsung Electronics Co., Ltd. (South Korea) | 0 | -100% |
| Arm Limited | 0 | -100% |
| Imagination Technologies Limited | 0 | -100% |
| Microsoft Corporation | 0 | — |
Where to take this analysis
The dataset points to specific next steps depending on whether the goal is freedom-to-operate, competitive tracking, or claim drafting.
Run a freedom-to-operate check on the busiest classes
With 1,024 records in G06T and 817 in G06F, any new shader- or memory-scheduling claim needs a targeted clearance search against that dense overlap before drafting.
Explore GPU architecture prior art in EurekaWatch the assignees still filing into the latest year
NVIDIA and Qualcomm remain active in the most recent year even as counts fall; tracking their published applications as they clear the publication lag will surface the next round of contested claims early.
Set up assignee monitoring in EurekaTest claim language against the under-claimed sub-areas
The gate chips above mark branches with thinner density than the core scheduling art — a fast way to check whether a draft claim actually sits in open space or just looks like it does.
Draft and stress-test claims in EurekaCommon questions about GPU architecture patents
Within this dataset, NVIDIA (NVIDIA Corporation) is the most active filer in the most recent tracked year, followed by Qualcomm (Qualcomm Incorporated), with Intel, Samsung, and ARM also present among the historically large filers. Cumulative leadership and recent-year momentum diverge, though: several of the largest historical filers show no publications in the latest year, which is more likely a publication-lag effect than an exit from the field. Anyone building a competitive map should look at both total family counts and recent-year activity separately, since they tell different stories.
No — filings peaked at 146 in 2018 and have declined since, with the 2022 midpoint at 73 and only 6 recorded in the most recent partial year. This decline partly reflects publication lag, since patents filed in the last one to two years typically have not published yet, but the multi-year downward trend from the 2018 peak predates that effect. The pattern is more consistent with a maturing claim landscape than with a technology in decline.
The core classes in this landscape are G06F15 and G06F9 for digital data processing and G06T1 for image data processing, which together anchor most of the corpus. G06T alone touches 1,024 of the 1,304 records, with G06F close behind at 817, while smaller supporting classes include G09G for display control, H04N for pictorial communication, and H03M for coding and conversion. Only 145 records overlap with G06N, the AI-computing class, indicating that GPU architecture claims are still mostly drafted independently of machine-learning-specific language.
The thinnest filing density relative to the dominant scheduling and memory-bandwidth claims sits in areas like cross-core cache coherence protocols, dynamic voltage-frequency shader scaling, and non-neural sparsity-aware scheduling. These are adjacent to heavily claimed territory but not directly overlapping with it, which gives a first claim there more room to stand on its own. That said, thinner density in a search corpus is not proof of true openness — a freedom-to-operate search against the full G06T/G06F overlap is still the necessary next step before relying on any gap.
Citation counts in this corpus are heavily concentrated at the top: the most-cited record, US20090002379A1 on GPU video decoding, carries 466 citations, well ahead of the next tier, which sits in the 130–160 range. That concentration reflects the age advantage older filings have in accumulating citations within a searched corpus, not necessarily their current technical relevance. Citation rank is useful for identifying foundational prior art but should not be read as a measure of which patents matter most today.
Research GPU Architecture and Parallel Compute in depth with Eureka
Go past this page: query the whole gpu architecture and parallel compute corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.