Federated Inference at the Edge Patents: Who Leads, Where Gaps Are 2026
- Concentrated but not locked up. The ranked leader holds 6 records and the top 5 assignees together account for 21 of 49 records in scope (42.9%) — enough to signal direction, not enough to close the field.
- Filing is still rising, not falling. 2021 to 2024 filings grew 25% (4 to 5), and 2023 is the peak year so far at 13; 2025-2026 figures are still filling in due to the usual ~18-month publication lag.
- AI-model claims dominate the transport layer. G06N (AI models) touches 57.1% of the 49 records, well ahead of H04L transmission claims at 34.7% — the contest is being fought in model orchestration, not network protocol.
Filing growth compares 2021 (4 records) with 2024 (5) — 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 49 records in scope (CR5), not by the ranked leaders only.
What this landscape covers
Federated inference at the edge sits where on-device AI meets multi-access edge computing: workloads are split or coordinated across mobile devices, edge nodes and cloud without moving raw training data off the device. This landscape pulls 49 published records filed or published between 2015 and mid-2026 that combine federated or distributed inference terminology with edge computing, MEC network or edge cloud language in the title, abstract, claims or description.
The set is small enough that individual filings move the picture, and large enough to show where claim density is building. Filings arrive from India, the United States, the WIPO PCT route, China and the EPO, with India and the US the two largest single receiving offices.
Let an AI agent run this analysis on your own technology
Pick a task. Every answer cites the patents behind it.
Filing trend and technology composition
Two views of the same 49 records: how filing activity has moved year over year, and which IPC subclasses carry the claims.
Filing trend, 2017-2026
Filings were at zero in 2017 and climbed to a peak of 13 in 2023. The 2021-2024 window shows genuine growth (+25%, 4 to 5), and 2026's count of 12 is a partial year rather than a plateau — later years understate real filing activity because publication trails filing by roughly 18 months.
IPC subclass composition
G06N (AI model computation) appears in 57.1% of the 49 records, ahead of H04L digital transmission (34.7%) and G06F general data processing (28.6%). Smaller shares in G06Q, G05D, G06V, H04B and H04W (each 8.2-10.2%) mark adjacent claim areas — business-process orchestration, non-electric control, vision, and wireless transport — that are present but far from saturated. Records can carry multiple classes, so these figures sum past 100%.
Shares are the percentage of the 49 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Edge Computing — Federated Inference at the Edge Patent Landscape with Eureka
This page is one run against one query. Ask Eureka your own question about edge computing — federated inference at the edge patent landscape and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative and most-cited filings
Optimized orchestration in federated inference across mobile devices
Rule-based decision-making is augmented by adaptively partitioning AI workloads across a federated inference infrastructure based on real-time assessment of device capabilities, network conditions and workload requirements, distributing inference demands across heterogeneous mobile devices.Filed by IBM, published 2026-03-12 — a recent illustration of orchestration claims layered on top of adaptive workload partitioning.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20210406782A1 | System and method for decentralized federated learning | 86 |
| 2 | CN116579417A | 边缘计算网络中的分层个性化联邦学习方法、装置及介质 | 30 |
| 3 | CN114997737A | 基于分层联邦学习的无人机小基站集群RAN切片方法 | 20 |
| 4 | WO2024054628A2 | Integrated unmanned and manned UAV network | 19 |
| 5 | CN114327889A | 一种面向分层联邦边缘学习的模型训练节点选择方法 | 17 |
| 6 | CN114745383A | 一种移动边缘计算辅助多层联邦学习方法 | 10 |
| 7 | US20260004150A1 | Temporal dynamics simulation in matmul-free neural architectures | 6 |
| 8 | US20240346327A1 | Online optimization for joint computation and communication in edge learning | 3 |
| 9 | WO2023007461A1 | Online optimization for joint computation and communication in edge learning | 3 |
| 10 | WO2022005937A1 | System and method for decentralized federated learning | 3 |
Citation counts favour older filings simply because they have had longer to accumulate references — treat this as a measure of influence within the corpus, not of present-day importance.
Patent titles are shown in the language they were filed in, not translated, so that each record stays verifiable against the original filing — a translated title will not match in Eureka or in any national register. 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 a filing decision
Three read-outs from the ranking, the trend and the classification data, aimed at where to file next rather than at describing the dataset.
The lead is real but shallow
The leading assignee holds 6 records; fifth place holds 3. That gap is not steep enough to call this a closed field — a strong second-mover claim in model partitioning or resource scheduling still has room.
Growth continues past the peak year
2023 is the peak year on record at 13 filings, but the 2021-2024 growth figure confirms this is not a one-year spike. 2025-2026 counts will rise as publication catches up with filing.
Model logic outruns network protocol claims
G06N appears in over half of records (57.1%) versus roughly a third for H04L transmission claims — the heavier claim traffic is in how inference is split and coordinated, not in the radio or network layer carrying it.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to edge computing — federated inference at the edge patent landscape, with the prior art for and against each one.
Who is filing, and where the gaps sit
39 companies and institutions make up the entire ranked list returned for this search — this is not a top-50 or top-100 cut, it is the whole field as the dataset defines it.
A single-digit lead, not a monopoly
The top-ranked assignee holds 6 of the 49 records in scope. That is enough to set direction on orchestration claims but leaves the majority of the field open to challengers.
Most of the ranking is single- or double-filers
By tenth place, holdings drop to 2 records. Combined with 39 ranked companies total, this points to a field where academic labs, telecom vendors and smaller AI firms are still staking early claims rather than consolidating.
Co-filing is limited and concentrated
Only 7 co-assignee pairs appear across the corpus, and the strongest pairing links two US university research bodies. Corporate-academic joint filings involving telecom vendors are present but thin.
| Assignee | Recent year | YoY |
|---|---|---|
| Beyond AGI LLC | 2 | 0% |
| FISCHIONE CONSULTING AB | 0 | — |
| George Washington University | 0 | — |
| THE RES FOUNDATION FOR THE STATE UNIV OF NEW YORK | 0 | — |
| Telefonaktiebolaget LM Ericsson (publ) | 0 | — |
| Nokia Technologies Oy | 0 | — |
| Intel Corporation | 0 | — |
| International Business Machines Corporation (IBM) | 0 | -100% |
Where to take this next
The dataset points to open questions rather than a finished picture — these are the natural next steps for a team deciding where to file or who to watch.
Model-partitioning claim scope
G06N carries the heaviest claim density in this field. A freedom-to-operate check on adaptive workload-partitioning language, specifically, is worth running before drafting a new claim in this space.
Explore claim scope in EurekaWatch the mid-tier filers
With holdings dropping to 2 records by tenth place, momentum could shift quickly. Tracking new publications from mid-ranked assignees is more informative here than watching the leader alone.
Set up assignee tracking in EurekaRecheck 2025-2026 once data settles
The most recent two years are still filling in behind the roughly 18-month publication lag. Revisit the trend once 2025 filings have largely published before drawing conclusions about slowing activity.
Revisit this landscape in EurekaCommon questions on this landscape
This landscape identifies 49 published records matching federated or distributed edge inference terminology combined with edge computing, MEC network or edge cloud language, covering filings from 2015 through mid-2026. That is a search-defined count based on a specific keyword string, not an exhaustive count of every related filing worldwide, since adjacent phrasing or classification-only filings may sit outside the search terms used. Treat it as a representative slice of the field rather than a total census.
The ranking covers 39 companies and research institutions, with the leader holding 6 of the 49 records in scope and the top 5 combined accounting for 21 records (42.9%). That concentration is moderate: it shows clear early movers but leaves most of the field, including positions from sixth place down through single-filing entrants, still open. No single assignee currently holds a dominant enough position to be described as controlling the space.
Yes, based on the years that can be treated as complete. Filing volume rose from 4 in 2021 to 5 in 2024, a 25% increase over that span, with 2023 standing as the peak year on record at 13 filings. Figures for 2025 and 2026 will look lower for now simply because publication typically lags filing by around 18 months, so those years should not be read as a slowdown.
The dominant classification is G06N, covering AI-model computation, which appears in 57.1% of the 49 records — reflecting that most claims focus on how inference workloads are modeled, partitioned or coordinated rather than on the underlying network. H04L digital transmission (34.7%) and G06F general data processing (28.6%) follow. Smaller shares in G06Q, G05D, G06V, H04B and H04W mark adjacent areas such as business-process integration, non-electric control, vision inference and wireless transport, each present in under 10% of records and comparatively under-claimed.
The classification data points to several under-claimed branches relative to the dominant AI-model and transmission classes: non-electric variable control tied to edge orchestration, business-process layers built on top of federated inference, vision-specific inference splitting, and wireless resource allocation tuned specifically for federated learning traffic. Each of these subclasses sits at 8.2-10.2% of the 49 records, well below the 57.1% share held by core AI-model claims. That gap suggests room for narrowly drafted claims in these adjacent areas rather than in the already dense orchestration space.
Research Edge Computing — Federated Inference at the Edge Patent Landscape in depth with Eureka
Go past this page: query the whole edge computing — federated inference at the edge patent landscape 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.