← ALL DATASETS
DATASET / epoch_ai_company_compute_spend / V20261111

AI company compute spending (Epoch AI)

See how much leading AI companies are reported to spend on cloud compute, and how the spending divides between running models for users (inference) and research, development and training.

OPEN IN QUERY ↗AI / compute / spending / companies / Epoch AI
SOURCEEpoch AI ↗
GEOGRAPHYGlobal; selected frontier AI companies
TIME2022–2025
UPDATED2026-11-11

VISUAL EXPLORATION

Table preview is the curated default for this dataset.

TABLE PREVIEW

Loading preview…

METADATA

DATA DICTIONARY

source_rowSource row · integer · identifier

Position of this record among the data rows of the source CSV (1 = first data row). Unique within this snapshot version only; it is not a stable Epoch AI identifier.

report_labelReport label · string · identifier

Epoch AI's short label for the record. Labels are not guaranteed to be unique.

companyCompany · string · dimension

Company name as used by Epoch AI; matches company in epoch_ai_companies.

observation_dateObservation date · string · dimension

As-of date for the figure (YYYY-MM-DD); for annual spending, the last day of the year.

period_typePeriod type · string · dimension

Time frame or reporting period the spending amount corresponds to, as labelled by the source.

categoryCategory · string · dimension

Spending categories, comma-separated. Inference cloud compute: cloud spend running models for users. R&D cloud compute: cloud spend on research, development and training. The source also uses Data, which its field reference does not define.

amount_usdAmount usd · number · measure

Spending amount in US dollars for the record's categories and period. Epoch AI's field reference does not define this column. The US dollar unit is a human-confirmed decision (0014).

inference_compute_spend_usdInference compute spend usd · number · measure

Source column 'Inference compute spend' in US dollars, published as recorded. Epoch AI's field reference does not define this column.

rd_compute_spend_usdRd compute spend usd · number · measure

Source column 'R&D compute spend' in US dollars, published as recorded. Epoch AI's field reference does not define this column.

total_compute_spend_usdTotal compute spend usd · number · measure

Source column 'Total compute spend' in US dollars, published as recorded. Epoch AI's field reference does not define this column.

confidenceConfidence · string · dimension

Epoch AI's confidence rating. Confident: primary or official source with specific figures. Likely: credible secondary source such as news reports with insider sources. Uncertain: less credible source or a claim with significant uncertainty.

source_typeSource type · string · dimension

Type of the primary source. Company disclosure: official statement from the company or its executives. Media report: reported by an established outlet, usually from insider sources or documents. Other: other evidence, such as third-party estimates or reports from non-executive staff. A record can carry more than one label, comma-separated.

report_dateReport date · string · dimension

Publication date of the primary source (YYYY-MM-DD).

source_url_1Source url 1 · string · dimension

Primary source URL.

source_url_2Source url 2 · string · dimension

Additional source URL, if any.

source_url_3Source url 3 · string · dimension

Additional source URL, if any.

PROVENANCE & LIMITATIONS

Records ticked 'Exclude from graph view' by Epoch AI were removed. Columns renamed; amounts cast to DOUBLE and dates to ISO text. Notes dropped. Input: ai_companies_compute_spend.csv extracted unchanged from ai_companies.zip, which the human downloaded from https://epoch.ai/data/ai_companies.zip on 2026-11-11 (zip SHA-256 8fe95fdcb688988f3bd8cb7f3d2bc512f94107d9f278cea4deb36a1f948dce7b); the same table is offered at https://epoch.ai/data/ai_companies_compute_spend.csv. The original CSV is not stored in Git because it contains the Notes column; raw/ holds a pointer file with its URL, size and checksum. CSV SHA-256 afd48524a17d81c5bea17daf3e63acf97587c34171d59ab402bced2ffb462f82. 14 source rows, 10 published rows, 4 removed. Field meanings follow https://epoch.ai/data/ai-companies-documentation/records. License CC BY 4.0; cite: Epoch AI, 'Data on AI companies'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/data/ai-companies' [online resource].

Processing code: pipeline/sources/epoch_ai/process-ai-companies.py

  • Only OpenAI and Anthropic have records in this snapshot.
  • Epoch AI's field reference states no currency for the amount columns; they are published as US dollars by human decision, and no inflation or currency adjustment was applied here.
  • The source does not define the amount column, the three split columns, the period-type options or the 'Data' category; they are published as recorded.
  • Figures are ongoing compute expenses reported for different periods (year, quarter); check period_type before comparing or summing.
  • Categories overlap: a combined R&D-and-inference record should not be added to separate R&D and inference records for the same period.
  • Records that Epoch AI excludes from its graph as not comparable are omitted.
  • Many figures come from media reports rather than audited disclosures. Use the confidence field; Epoch AI states that individual data points should not be considered completely reliable.
  • The source's free-text Notes column (evidence and reasoning, often quoting third-party media) is not published here. Follow the source URLs or see https://epoch.ai/data/ai-companies for that context.
  • Epoch AI revises this database frequently; this is a dated snapshot, not a live feed.