Collective Communication Libraries Patents: Leaders & Trends 2026
- Concentrated but not locked up. the top 5 assignees account for 34.1% of all 138 records in scope, and the top 10 combined reach 55.8% — leaving a long tail of single- and few-filing entrants.
- Filings are still climbing. activity rose from 12 records in 2021 to 17 in 2024, a +42% increase over that three-year span, with 2025 recording 28 records so far as the field's peak year.
- Claims cluster in software, not silicon. G06F (electric digital data processing) covers 64.5% of the 138 records and H04L (digital transmission) 37.0%, showing the fight is over scheduling and data-processing logic more than physical interconnect hardware.
Filing growth compares 2021 (12 records) with 2024 (17) — 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 138 records in scope (CR5), not by the ranked leaders only.
What this landscape covers
Collective communication libraries coordinate how many compute nodes exchange and combine data — all-reduce, gradient aggregation, ring and hierarchical reduction schedules — during distributed training and high-performance computing jobs. This landscape tracks 138 published records matched on collective-communication operations paired with topology-awareness, scheduling and overlap terms, spanning filings from 2015 through the 2026-07-31 data cut-off. Because publication lags filing by roughly eighteen months, the most recent year understates real filing activity.
The scope sits at the intersection of distributed systems software and network hardware: expect scheduling algorithms, topology-aware routing logic and threshold-based message handling rather than physical switch or NIC design claims.
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
The record set spans eleven years of publication activity, concentrated in digital-processing and AI-adjacent IPC classes rather than pure networking hardware.
Filing trend, 2017–2026
Filings moved from zero in 2017 to a documented 2021→2024 rise of 12 to 17 records (+42%), with 2025 reaching 28 records as the peak year on record; 2026 shows 7 records so far but is still an incomplete, filling-in year under the usual publication lag.
IPC subclass composition
G06F (electric digital data processing) appears in 64.5% of the 138 records, H04L (digital information transmission) in 37.0%, and G06N (AI-model computing) in 26.8%; because records can carry multiple IPC classes these shares add to more than 100% of the record total, which is expected.
Shares are the percentage of the 138 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Collective Communication Libraries with Eureka
This page is one run against one query. Ask Eureka your own question about collective communication libraries and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative and most-cited records
Efficient all-to-all collective communication schedules for direct-connect topologies
The method scales a max concurrent multi-commodity flow (MCF) framework by decomposing and parallelising the MCF problem to compute fast link-based all-to-all schedules. It builds a time-stepped version of the MCF problem for host-based forwarding topologies, forms a direct-connect graph from it, and uses that graph to compute time-stepped MCF schedules for managing mixed topologies, including identification of a direct-connect topology for performing the all-to-all communication.Filed by University of Washington, published 2025-12-04 — a scheduling-algorithm claim rather than a hardware claim, typical of the software-heavy composition seen across this landscape.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20060069912A1 | Systems and methods for enhanced network security | 278 |
| 2 | US20250259085A1 | Convergent Intelligence Fabric for Multi-Domain Orchestration of Distributed Agents with Hierarchical Memory … | 208 |
| 3 | US6038604A | Method and apparatus for efficient communications using active messages | 65 |
| 4 | US20090037926A1 | Methods and systems for time-sharing parallel applications with performance isolation and control through per… | 61 |
| 5 | US20260046317A1 | AI-Optimized Memory Fabric for Large Contexts and Multimodal Workloads | 60 |
| 6 | US8065725B2 | Systems and methods for enhanced network security | 50 |
| 7 | US20220114475A1 | Methods and systems for decentralized federated learning | 41 |
| 8 | US20210374503A1 | Network-centric architecture and algorithms to accelerate distributed training of neural networks | 27 |
| 9 | US20110060891A1 | Parallel pipelined vector reduction in a data processing system | 20 |
| 10 | US20090276765A1 | Compiler driven mechanism for registration and deregistration of memory pages | 18 |
Citation counts favour older records inside any searched corpus — read them as a signal of influence on the field, not as a ranking of current 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 a filing decision
Three patterns matter more than the raw counts: where filings concentrate by class, how recent the growth is, and where citation weight sits.
Leadership is real but not exclusive
The top 5 assignees combine for 34.1% of all 138 records and the top 10 for 55.8%. That leaves close to half the field held by companies and institutions outside the ranked leaders, meaning freedom-to-operate analysis cannot stop at the visible leaders alone.
Growth is recent, not historical
Filings rose from 12 records in 2021 to 17 in 2024, and 2025 alone recorded 28 — the highest year in the dataset. Treat 2025 and 2026 counts as still filling in rather than as evidence of a plateau.
The contest is over scheduling logic
G06F carries 64.5% of the 138 records and H04L 37.0%, with G06N (AI-model computing) at 26.8% — a sign that claims are increasingly framed around AI training workloads rather than generic networking.
Older filings still anchor prior art searches
The most-cited records in this set include general network-security and active-messaging patents from the 1990s-2000s alongside recent AI-fabric filings, confirming that citation counts here reward age and breadth of applicability more than recency.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to collective communication libraries, with the prior art for and against each one.
Who is active and where the openings sit
Filing activity spans large systems vendors, networking specialists and university and institute research groups, with momentum data showing most named leaders flat or down year-on-year in the latest partial year.
Diversified computing vendors lead the count
The single leading assignee holds 12 records, ahead of a fifth-place holder at 7 and a tenth-place holder at 5 — a gradual taper rather than a sharp cliff, consistent with a field where scheduling and topology claims are filed alongside broader systems patent portfolios.
A long tail of narrower filers
Of the 47 companies in the full ranking, the 37 outside the top 10 hold the remainder of the 138 records between them — smaller networking and silicon specialists filing narrower, more defensible claims around specific topology or threshold mechanisms.
Momentum has cooled for named leaders in the latest partial year
Recent-year momentum figures show most of the named leading assignees, including two large systems vendors, at zero filings in the latest year, with two smaller filers down 100% year-on-year. Given the publication lag, this likely reflects filing-to-publication timing rather than an actual pullback.
| Assignee | Recent year | YoY |
|---|---|---|
| Hewlett Packard Enterprise Development LP | 1 | — |
| International Business Machines Corporation | 0 | — |
| UniFabrix Ltd | 0 | -100% |
| Huawei Technologies Co., Ltd. | 0 | — |
| Innovium Inc | 0 | -100% |
| Cambricon (Xi'an) Semiconductor Co., Ltd. | 0 | — |
| Samsung Electronics Co., Ltd. | 0 | -100% |
| Arista Networks Inc | 0 | -100% |
Where to take this next
The dataset points to specific next steps depending on whether the goal is freedom-to-operate, competitive tracking, or identifying open claim space.
Check freedom-to-operate against the long tail
With 55.8% of records held by ten assignees, nearly half the field sits with narrower filers whose claims are easy to miss in a leader-only search.
Run a full assignee sweepWatch 2025-2026 publications as they land
The peak year on record (2025, 28 records) and the partial 2026 count will keep filling in under the publication lag, so re-check the trend before finalising any filing decision.
Track ongoing filingsDraft around the under-claimed branches
Message-size threshold tuning and cross-vendor direct-connect scheduling show comparatively little claim density relative to the core scheduling classes.
Explore white space with EurekaCommon questions about this landscape
The leading assignee in this dataset holds 12 of the 138 records, with the fifth-ranked assignee at 7 and the tenth-ranked at 5. The top 5 assignees together account for 34.1% of all 138 records, and the top 10 combined reach 55.8%, which means close to half the field is held by companies and institutions outside the most visible leaders. A full ranking of 47 companies is available for anyone doing a competitive sweep rather than relying on the top names alone.
Yes, based on the complete years available: filings rose from 12 records in 2021 to 17 in 2024, a documented +42% increase over that span, and 2025 recorded 28 records, the highest year in the dataset. Because publication typically lags filing by around eighteen months, 2025 and especially the partial 2026 count should be read as still filling in rather than as a sign of a slowdown. The safest read is that growth accelerated into 2025 rather than plateaued.
The dataset is dominated by software and scheduling claims rather than physical interconnect hardware: G06F (electric digital data processing) appears in 64.5% of the 138 records and H04L (digital information transmission) in 37.0%. G06N, covering AI-model computing, appears in 26.8% of records, indicating a growing share of filings frame collective communication specifically around AI training workloads. Smaller shares touch image processing, coding conversion and memory architectures, but these are minor branches by comparison.
Message-size threshold tuning for mixed-precision transfers, hierarchical reduction across heterogeneous accelerator types, communication-computation overlap scheduling, and cross-vendor direct-connect graph scheduling all sit adjacent to the dense G06F/H04L core without matching its filing density. These are the branches worth checking first for a first claim, since the core scheduling and topology-awareness space is already fairly occupied by the ranked leaders. Confirming novelty still requires a targeted prior-art search rather than relying on aggregate class density alone.
The United States receives the largest share of filings among the offices tracked, with the European Patent Office, WIPO's PCT route, and Germany's national office following at smaller volumes. This distribution suggests most applicants are prioritising US protection first, then extending select filings through PCT or direct European and German routes. Anyone assessing regional freedom-to-operate should weight a US-first search accordingly.
Research Collective Communication Libraries in depth with Eureka
Go past this page: query the whole collective communication libraries 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.