Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
seed_organizations: list<item: string>
  child 0, item: string
news_sites: list<item: string>
  child 0, item: string
search_strings: list<item: string>
  child 0, item: string
search_locations: list<item: string>
  child 0, item: string
source_table: string
omitted_columns: struct<id: string, dedupe_key: string, relationship_timestamp: string, relationship_llm_model: strin (... 66 chars omitted)
  child 0, id: string
  child 1, dedupe_key: string
  child 2, relationship_timestamp: string
  child 3, relationship_llm_model: string
  child 4, actor_1_ai_stack_layer: string
  child 5, actor_2_ai_stack_layer: string
published_columns: struct<actor_1_type: struct<type: string, enum: list<item: string>>, actor_2_type: struct<type: stri (... 710 chars omitted)
  child 0, actor_1_type: struct<type: string, enum: list<item: string>>
      child 0, type: string
      child 1, enum: list<item: string>
          child 0, item: string
  child 1, actor_2_type: struct<type: string, enum: list<item: string>>
      child 0, type: string
      child 1, enum: list<item: string>
          child 0, item: string
  child 2, ai_stack_layers: struct<type: string, min_items: int64, max_items: int64, item_enum: list<item: string>, description: (... 8 chars omitted)
      child 0, type: string
      child 1, min_items: int64
      child 2, max_items: int64
      child 3, item_enum: list<item: string>
          child 0, item: string
      child 4, description: string
  child 3, relationship_type: struct<
...
       child 0, item: string
  child 5, directionality: struct<type: string, enum: list<item: string>, description: string>
      child 0, type: string
      child 1, enum: list<item: string>
          child 0, item: string
      child 2, description: string
  child 6, evidence_strength: struct<type: string, enum: list<item: string>>
      child 0, type: string
      child 1, enum: list<item: string>
          child 0, item: string
  child 7, ai_relevance: struct<type: string, enum: list<item: string>, description: string>
      child 0, type: string
      child 1, enum: list<item: string>
          child 0, item: string
      child 2, description: string
  child 8, ai_relationship_categories: struct<type: string, max_items: int64, item_enum: list<item: string>, description: string>
      child 0, type: string
      child 1, max_items: int64
      child 2, item_enum: list<item: string>
          child 0, item: string
      child 3, description: string
  child 9, relationship_pass: struct<type: string, description: string>
      child 0, type: string
      child 1, description: string
row_filter: struct<ai_relevance_in: list<item: string>, exclude_if_any_equals: struct<columns: list<item: string (... 18 chars omitted)
  child 0, ai_relevance_in: list<item: string>
      child 0, item: string
  child 1, exclude_if_any_equals: struct<columns: list<item: string>, value: string>
      child 0, columns: list<item: string>
          child 0, item: string
      child 1, value: string
to
{'published_columns': {'actor_1_type': {'type': Value('string'), 'enum': List(Value('string'))}, 'actor_2_type': {'type': Value('string'), 'enum': List(Value('string'))}, 'ai_stack_layers': {'type': Value('string'), 'min_items': Value('int64'), 'max_items': Value('int64'), 'item_enum': List(Value('string')), 'description': Value('string')}, 'relationship_type': {'type': Value('string'), 'enum': List(Value('string'))}, 'relationship_status': {'type': Value('string'), 'enum': List(Value('string'))}, 'directionality': {'type': Value('string'), 'enum': List(Value('string')), 'description': Value('string')}, 'evidence_strength': {'type': Value('string'), 'enum': List(Value('string'))}, 'ai_relevance': {'type': Value('string'), 'enum': List(Value('string')), 'description': Value('string')}, 'ai_relationship_categories': {'type': Value('string'), 'max_items': Value('int64'), 'item_enum': List(Value('string')), 'description': Value('string')}, 'relationship_pass': {'type': Value('string'), 'description': Value('string')}}, 'omitted_columns': {'id': Value('string'), 'dedupe_key': Value('string'), 'relationship_timestamp': Value('string'), 'relationship_llm_model': Value('string'), 'actor_1_ai_stack_layer': Value('string'), 'actor_2_ai_stack_layer': Value('string')}, 'source_table': Value('string'), 'row_filter': {'ai_relevance_in': List(Value('string')), 'exclude_if_any_equals': {'columns': List(Value('string')), 'value': Value('string')}}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 274, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              seed_organizations: list<item: string>
                child 0, item: string
              news_sites: list<item: string>
                child 0, item: string
              search_strings: list<item: string>
                child 0, item: string
              search_locations: list<item: string>
                child 0, item: string
              source_table: string
              omitted_columns: struct<id: string, dedupe_key: string, relationship_timestamp: string, relationship_llm_model: strin (... 66 chars omitted)
                child 0, id: string
                child 1, dedupe_key: string
                child 2, relationship_timestamp: string
                child 3, relationship_llm_model: string
                child 4, actor_1_ai_stack_layer: string
                child 5, actor_2_ai_stack_layer: string
              published_columns: struct<actor_1_type: struct<type: string, enum: list<item: string>>, actor_2_type: struct<type: stri (... 710 chars omitted)
                child 0, actor_1_type: struct<type: string, enum: list<item: string>>
                    child 0, type: string
                    child 1, enum: list<item: string>
                        child 0, item: string
                child 1, actor_2_type: struct<type: string, enum: list<item: string>>
                    child 0, type: string
                    child 1, enum: list<item: string>
                        child 0, item: string
                child 2, ai_stack_layers: struct<type: string, min_items: int64, max_items: int64, item_enum: list<item: string>, description: (... 8 chars omitted)
                    child 0, type: string
                    child 1, min_items: int64
                    child 2, max_items: int64
                    child 3, item_enum: list<item: string>
                        child 0, item: string
                    child 4, description: string
                child 3, relationship_type: struct<
              ...
                     child 0, item: string
                child 5, directionality: struct<type: string, enum: list<item: string>, description: string>
                    child 0, type: string
                    child 1, enum: list<item: string>
                        child 0, item: string
                    child 2, description: string
                child 6, evidence_strength: struct<type: string, enum: list<item: string>>
                    child 0, type: string
                    child 1, enum: list<item: string>
                        child 0, item: string
                child 7, ai_relevance: struct<type: string, enum: list<item: string>, description: string>
                    child 0, type: string
                    child 1, enum: list<item: string>
                        child 0, item: string
                    child 2, description: string
                child 8, ai_relationship_categories: struct<type: string, max_items: int64, item_enum: list<item: string>, description: string>
                    child 0, type: string
                    child 1, max_items: int64
                    child 2, item_enum: list<item: string>
                        child 0, item: string
                    child 3, description: string
                child 9, relationship_pass: struct<type: string, description: string>
                    child 0, type: string
                    child 1, description: string
              row_filter: struct<ai_relevance_in: list<item: string>, exclude_if_any_equals: struct<columns: list<item: string (... 18 chars omitted)
                child 0, ai_relevance_in: list<item: string>
                    child 0, item: string
                child 1, exclude_if_any_equals: struct<columns: list<item: string>, value: string>
                    child 0, columns: list<item: string>
                        child 0, item: string
                    child 1, value: string
              to
              {'published_columns': {'actor_1_type': {'type': Value('string'), 'enum': List(Value('string'))}, 'actor_2_type': {'type': Value('string'), 'enum': List(Value('string'))}, 'ai_stack_layers': {'type': Value('string'), 'min_items': Value('int64'), 'max_items': Value('int64'), 'item_enum': List(Value('string')), 'description': Value('string')}, 'relationship_type': {'type': Value('string'), 'enum': List(Value('string'))}, 'relationship_status': {'type': Value('string'), 'enum': List(Value('string'))}, 'directionality': {'type': Value('string'), 'enum': List(Value('string')), 'description': Value('string')}, 'evidence_strength': {'type': Value('string'), 'enum': List(Value('string'))}, 'ai_relevance': {'type': Value('string'), 'enum': List(Value('string')), 'description': Value('string')}, 'ai_relationship_categories': {'type': Value('string'), 'max_items': Value('int64'), 'item_enum': List(Value('string')), 'description': Value('string')}, 'relationship_pass': {'type': Value('string'), 'description': Value('string')}}, 'omitted_columns': {'id': Value('string'), 'dedupe_key': Value('string'), 'relationship_timestamp': Value('string'), 'relationship_llm_model': Value('string'), 'actor_1_ai_stack_layer': Value('string'), 'actor_2_ai_stack_layer': Value('string')}, 'source_table': Value('string'), 'row_filter': {'ai_relevance_in': List(Value('string')), 'exclude_if_any_equals': {'columns': List(Value('string')), 'value': Value('string')}}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Dataset Card for AI Supply Chain Relationships

A regularly updated dataset of relationships in the AI industry from late 2022, extracted from public data sources. This dataset maps the AI supply chain: the network of organizations that contribute to the development, deployment, and use of AI systems, including model developers, cloud providers, data vendors, investors, enterprise users, and more.

Dataset Details

Dataset Description

A regularly updated dataset of relationships in the AI industry, extracted from public data sources. This dataset maps the AI supply chain: the network of organizations that contribute to the development, deployment, and use of AI systems, including model developers, cloud providers, data vendors, investors, enterprise users, and more. These relationships are obtained by running targeted Google Search queries from late 2022 until the present, updated approximately weekly. Each row corresponds to a relationship, with a link to the original data source, the article_title, the document_date, and a relationship_description. The actors involved are named and annotated with types (actor_1_type and actor_2_type). The relationship is further annotated by relationship_type, relationship_status, directionality, ai_stack_layers, evidence, evidence_strength, ai_relevance, and ai_relationship_categories.

  • Curated by: Sarah H. Cen, Lindsey Gailmard, Joohan Kim, Rishi Bommasani, Daniel E. Ho, Percy Liang
  • Language(s): English
  • License: CC BY-NC-SA 4.0

Dataset Sources

Uses

Direct Use

Intended for non-commercial research, education, and analysis. This dataset is released under CC BY-NC-SA 4.0, which does not permit commercial use. Other licensing (including commercial) may be available by separate agreement with the authors.

Out-of-Scope Use

  • Commercial use, unless by a separate agreement with the authors
  • Treating extractions as ground-truth legal, financial, or regulatory facts (they are LLM extractions from web text)
  • Surveillance, doxxing, or other harmful targeting of individuals in the data
  • Claiming completeness of the AI industry (coverage is shaped by seed organizations, search queries, indexing, and what is publicly announced)

Dataset Structure

  • data/gsearch_relationship_results.parquet - the main dataset
  • metadata/search_config.json — seed organizations and Google Search collection parameters
  • metadata/label_schema.json — values for categorical fields
  • metadata/schema.md — field-by-field schema and label definitions

Each row is a relationship extracted from a document and indexed by the relationship pass over that document. Rows do not represent unique number of relationships because many documents may cover the same relationship. Plus, there may be more than one row per relationship per document when there are multiple relationship passes over the document.

Fields

Column Description
link Source document URL
article_title Title associated with the source page
document_date Parsed document / article date (when available)
actor_1_name Name of the first actor
actor_1_type Actor-type label for actor 1
actor_2_name Name of the second actor
actor_2_type Actor-type label for actor 2
relationship_type Relationship-type label
relationship_status Lifecycle status (e.g., proposed, established, ended)
directionality Directionality label (e.g., ACTOR_1_TO_ACTOR_2, ACTOR_2_TO_ACTOR_1, SYMMETRIC)
relationship_description Short natural-language description of the relationship
evidence Supporting evidence text from the source
evidence_strength Strength label for the evidence
ai_relevance AI-relevance label (AI_RELATIONSHIP or INVOLVES_AI_ACTOR)
ai_relationship_categories AI topical tags (≤3)
relationship_pass Extraction pass index (the pipeline runs multiple independent passes)
ai_stack_layers Relationship-level AI stack layers (1–2)

Example load

from datasets import load_dataset

ds = load_dataset("sarahcen/ai-supply-chain")
print(ds)
print(ds["train"][0])
import pandas as pd

df = pd.read_parquet(
    "hf://datasets/sarahcen/ai-supply-chain/data/gsearch_relationship_results.parquet"
)

Dataset Creation

Curation Rationale

This dataset helps to better understand relationships between actors in the AI industry, including but not limited to:

  1. Who provides models to whom
  2. Who provides data to whom
  3. Who proivdes compute to whom
  4. Who provides capital to whom
  5. Who competes with whom
  6. Litigation between organizations

This dataset provides continually updated mapping of those relationships. We intentionally use public web sources so that others can return to the source material.

Source Data

Data Collection and Processing

High-level pipeline (see paper for details):

  1. Google Search: run targeted queries for all seed organizations in search_config.json using relationship-oriented queries. For each organization, we run multiple broad searches plus site-restricted news domains such as TechCrunch, VentureBeat, Yahoo Finance.
  2. Document retrieval and classification: scrape pages (that are not blocked).
  3. Relationship extraction and categorization: run multiple passes on each document, extracting relationships, labeling AI relevance and evidence strength, then assigning actor types, relationship type/status/directionality, AI stack layers, and AI tags according to the taxonomy in label_schema.json and schema.md.

The released dataset only contains AI-related relationships.

Who are the source data producers?

The data sources are publicly available documents, such as news articles, press releases, blogs, and other search indexed pages.

Annotations

Annotation process

Relationships and actor categories are extracted and assigned automatically via LLM structured outputs from scraped document text. The public release keeps rows with ai_relevance in {AI_RELATIONSHIP, INVOLVES_AI_ACTOR} and excludes rows labeled CONTRADICTORY_INFORMATION on relationship type, status, or directionality. Multiple extraction passes may yield multiple rows for the same underlying document/relationship identity (relationship_pass).

Who are the annotators?

Annotations are either from Google Search metadata or post-processing done by LLMs (primarily, gpt-5.6-luna).

Personal and Sensitive Information

All information is extracted from public data sources and LLM post-processing.

Bias, Risks, and Limitations

The mapping by no means have complete coverage over the AI industry and supply chain. Source of coverage bias are the organizations we use to seed our searches; our use of Google Search and their ranking algorithm; that our searches are in English.

Our data sources are publicly available documents returned by Google Search. This introduces biases. For instance, articles behind paywalls are not included; high-quality but low-traffic text sources may be missed; the sources themselves may present information in a biased way (e.g., press releases can overstate, omit, or misrepresent events/relationships).

Document processing (such as relationship extraction and categorization) is performed by LLMs. There are limitations to this approach, including potential false negatives and positives, hallucinated results, and non-standard names for organizations.

This approach also results in actor "duplicates" that refer to the same organization under the different names. We chose to keep this behavior because the actor names generally mirror how they are referred to in the raw text. It is thus possible to post-process the names, using the desired "coarseness" (e.g., should Google DeepMind be collapsed to Google or Alphabet?).

The data also has duplicate relationships because we extract relationships per document, and multiple documents may cover the same relationship.

Some documents did not have a valid data in their metadata or a parseable date.

Recommendations

  • We recommend validating critical edges and findings based on them against the linked source.
  • We recommend deduplicating actor names as coarsely as desired before running analyses that draw findings from the numbers/trends in this dataset.
  • See the relationship_pass column to help merge duplicate relationships resulting from running multiple LLM passes over the same document.
  • To easily explore the data, please use: https://aisupplychains.org/

Citation

BibTeX:

@misc{aisc_mapping_2025,
    title={{AI} {S}upply {C}hain {M}apping: An {A}nalysis of the {C}omplex {R}elationships in the {AI} {E}cosystem},
    author={Sarah H. Cen and Lindsey Gailmard and Joohan Kim and Rishi Bommasani and Daniel E. Ho and Percy Liang},
    year={2026},
    journal={Forthcoming}
}

APA:

Cen, S. H., Gailmard, L., Kim, J., Bommasani, R., Ho, D. E., & Liang, P. (2026). Mapping the AI Supply Chain: An Analysis of the Complex Relationships in the AI Ecosystem. Forthcoming.

Dataset Card Authors

Sarah H. Cen

Dataset Card Contact

sarahcen@andrew.cmu.edu

Downloads last month
27