The assignee ranking for this dataset returns 7 companies, not a top-50 or top-100 list — it is the full ranked field the data endpoint produces for this search. One company leads with 8 records, while the fifth-ranked company holds only 2, so the field is dominated by one filer with a short tail of smaller portfolios behind it. This is a small enough group that tracking each ranked assignee individually is practical for competitive monitoring.

Filings peaked at 7 in 2021 and had fallen to 3 by 2024, a 57% decline across that three-year span, based on the most recent year that can be treated as complete. Counts for 2025 and 2026 look lower still, but publication lags filing by roughly 18 months, so those years are still filling in and should not be read as continued decline. The honest read is a filing burst around 2021 followed by fewer, more selective filings rather than a technology in retreat.

US12487844B2, assigned to International Business Machines Corporation and published 2025-12-02, covers a method for managing container images across runtimes by separating content data from management data and storing each selectively in a shared file system. The novelty sits in avoiding duplicate storage: management data is written on its own once the underlying content data already exists in the image-sharing file system. Anyone building image-sharing or deduplication logic for container runtimes should read the full claim set before assuming their approach is clear of it.

G06F, electric digital data processing, appears in 90.0% of the 20 records in scope, making it by far the dominant classification. G06V, image and video recognition, overlaps in 35.0% of records, pointing to a secondary cluster connecting container runtimes to vision or edge-inference use cases. G01C, G06K and G06N each appear in only one record apiece, marking those as largely unclaimed branches rather than established territory.

The classification data points to G01C, G06K and G06N as the thinnest branches, each holding only a single record among the 20 in scope, compared with 90.0% density in G06F. That imbalance suggests claims tying rootless runtime mechanisms to navigation, data recognition/presentation, or AI-model-based computing are largely open. A first claim drafted precisely in one of those branches would face far less crowded prior art than one filed into the core G06F space.