AI Computing Keyword Candidate Study

Comparative structure study

The largest recorded reporting jurisdiction, Beijing, contains 105 records (25.4% of this 414-record cohort). The remaining cohort contains 309 records. The comparison concerns archive composition.

The recorded industry label Next-generation information technology accounts for 75.8% here (314/414), compared with 27.1% in the parent set (5,226/19,297), an absolute gap of 48.8 percentage points.

108/414 companies link to pages with existing source-supported excerpts (26.1% of the cohort). The other 306 records remain usable for archive composition, but do not support new operating case studies.

An ID-deduplicated archive snapshot, not the whole market, current recognition or operating results.

The parent set, All public archive records, contains 19,297 records. This cohort contributes 414 (2.1%); 18,883 records remain after excluding it.

Ratio of label participation, not market share; multi-label counts overlap.

Excluding this cohort, the recorded industry label Next-generation information technology covers 4,912/18,883 remaining records (26.0%), an absolute gap of 49.8 percentage points from this cohort.

The full reporting jurisdiction distribution has 30 labels; the top 3 contain 225/414 records (54.3%).Other known labels contain 189 records.

For 414 records with known labels, the effective category count is 7.9, versus 30 actual labels. Definitions and scope are in the methodology.

The table shows ten labels; 20 others are included in the calculations above.

Record counts do not rank supply capability or economic size.

The largest intersection, Beijing × Next-generation information technology, contains 85 records, or 81.0% of the 105 records carrying the Beijing label.

A cell contains company IDs carrying both labels. It does not establish procurement, supply or investment relationships between firms. Multi-label cells may overlap.

This page displays 8 of the full cross-tab’s 80 non-empty cells; the other 72 remain in the analysis.

Non-empty fields are not necessarily accurate or supported by originals.

Counts concern existing reviewed bilingual excerpts, not re-certification of originals or entire reports.

A rule hit prioritises review; it does not establish that the entire field is wrong. One record can trigger several rules.

An ID may match several fields, so counts may exceed the cohort total. Matches establish search relevance only.

Among the translated non-broad query terms, “artificial intelligence” has the most matches: 208/414. Name, product and sub-industry fields match 4, 181 and 35 records respectively; the largest field is product field. Fields can overlap, and a term remains a search link rather than verified business activity.

Selected broad terms and specific segment terms, not the full query. Terms and fields can overlap and must not be summed. Counts show lexical sensitivity; no records are removed without primary evidence.

71 of this cohort’s 414 records link through existing company-ID mappings to 2026 lists from all jurisdictions (17.1%). This intersection measures list linkage, not successful recognition or company growth.

List rows and company IDs are separate counts, not new firms or final recognition.

Model applications require task and data checks

The Artificial intelligence and data intelligence label covers 33 records (8.0% of this cohort). This archive clue raises a verification question about Task-bounded industrial models or application services with evaluation records; it does not establish that these companies already deliver it.

Shared records with Database Data Governance Keyword Candidate Study: 88; records in this cohort but outside that set: 326. The intersection/union ratio is 17.3%, a measure of label overlap rather than business collaboration. Read comparison

Shared records with Cloud Data Center Infrastructure Keyword Candidate Study: 79; records in this cohort but outside that set: 335. The intersection/union ratio is 17.0%, a measure of label overlap rather than business collaboration. Read comparison

Stable-ID samples, not rankings; metrics use the full cohort.