MLOps and AI/ML Infrastructure (LF AI & Data, MLPerf, Kubeflow)

Where this market publishes its structure

• LF AI & Data Landscape (lfai/landscape landscape.yml @95d361b8ab) machine-readable, 473 entries, 67 members, 40 hosting companies, 62 LF AI & Data projects; 69 providers linked
• MLCommons MLPerf results (mlcommons/*_results_v* repos: closed/ + open/ submitter directories) machine-readable, 54 round repos, 9 families (inference v6.1, training v6.0, storage v3.0, hpc v3.0, mobile, tiny, automotive, endpoints); 183 directories -> 118 organisations; 31 providers linked via Wikidata P856
• Kubeflow packaged distributions (kubeflow/website installing-kubeflow table @5a17330435) machine-readable, 6 rows; 1 linked (Red Hat); Canonical (ubuntu.com), prokube, QBO have no catalog holder; Azure/Nutanix link to github.io
• MLPerf AlgoPerf (algorithms_results) not machine-readable: anonymised team numbers, not organisations
• MLCommons members (mlcommons.org/our-members) not machine-readable: logo grid of images; no names or links in the HTML
• Hugging Face verified organisations not machine-readable: no public list; /api/organizations HTTP 401
• Ray ecosystem / integrations (docs.ray.io) not machine-readable: docs prose page, a mention list not a register

Own = declares at least one integration itself; covered = visible only because an aggregator, standards body, regulator or partner contract names it; distribution-only = listed only on cloud marketplaces it sells through, a channel reported beside, never counted as its own declared integrations; undeclared = nothing we can read, so no number; declares_compatibility = providers with a `compatible` shape vote, report only (kin, 2026-10-13): never counted as own, never in depth or position.

Taxonomy how this market maps onto apis.io

Derived 2026-10-15 from the 67 providers in this market we could resolve in the catalog. The percentage is the share of those providers carrying each one; it describes the market, not the tag or industry. Tag groups have no page of their own yet, so they are listed unlinked.

share of resolved cohort carrying it; tags >=0.20 & >=3 members (max 12, junk dropped), tag_groups >=0.20 (cooccurrence, non-cross-cutting; max 8), industries >=0.20 (max 6), areas >=0.10 (max 6; areas are keyword-matched and sparse), capabilities L1 >=0.20 (max 8); derive-market-taxonomy.py

Leaders by position (provisional)

Provisional: grouped by position, depth compared within a position only. Distribution (cloud marketplaces) and iPaaS bridges are counted beside position, never inside it. Unknown-builder listings are left out of position shares and counted here.

builds-on 5 · publishes apps on other platforms' stores

covered 6 · reached mainly through aggregators

regulated 1 · named in a regulator's register or partners' contracts

standard 24 · listed by a voluntary standard's own directory

long-tail 1 · reached only through iPaaS bridges

distribution 4 · only sells through cloud marketplaces

undeclared 26 · declares nothing we can read: no number