Methodology

Where the data comes from

USAspending.gov publishes reported prime awards, including procurement data from FPDS and assistance data from FABS. We query its award search API for new awards from 1 October 2023 with the keywords artificial intelligence, machine learning, large language model, generative AI, AI/ML, deep learning, computer vision, natural language processing, neural network and LLM. Then we pull each candidate’s full record from /api/v2/awards/<id>/: agency and sub-agency, recipient and parent company, obligations, base-and-all-options value, dates, PSC, NAICS, Assistance Listing, competition and pricing. USAspending’s spending guide explains the reporting sources and limitations; documentation accessed 11 November 2026.

What gets a row

Keyword search alone catches awards that merely mention AI somewhere in a recipient name or a long abstract. A row stays only when the award’s own description passes the AI term test: AI, machine learning, an LLM, generative AI, deep learning, computer vision, neural networks or natural language processing. Contracts and orders need at least $1M obligated, grants and cooperative agreements at least $5M obligated, and IDV vehicles a ceiling of $1M or more. These are the thresholds in our transformation script.

The collector searches award type codes A, B, C, D (BPA calls, purchase orders, delivery orders and definitive contracts); IDV_A, IDV_B, IDV_B_A, IDV_B_B, IDV_B_C, IDV_C, IDV_D, IDV_E (GWAC, indefinite-delivery contracts, schedules, BOAs and BPAs); and 02, 03, 04, 05 (block, formula and project grants and cooperative agreements). It does not query loans, direct payments, other financial assistance or the newer F001–F010 codes shown in the live type reference (accessed 11 November 2026).

Searches sort by award amount and stop after at most 10 pages of contracts or 6 pages each of IDVs and grants, at 100 results per page per keyword. Each search also stops at the collector’s amount floor. The transformer drops PSC construction and maintenance codes (Y and Z), three acronym false positives (SBA’s Loan and Lender Monitoring System, “LLMS”, and a Master of Laws course) and grants whose first 260 description characters fail the AI test. The result is a curated ledger of material federal AI awards, refreshed from the source on every pull.

Obligated, ceiling, never added

Obligated is the sum of every obligation on the award so far, de-obligations included: money the government has committed. Ceiling is the base-and-all-options value reported for a contract or vehicle. An IDIQ often shows a large ceiling and almost no obligation, because the money flows through orders placed against it. Orders appear as their own rows with the parent vehicle named, so adding a ceiling to obligations double counts. Grants carry no ceiling here. See USAspending’s award spending definitions (accessed 11 November 2026).

Dates and fiscal years

The action date is the date signed for contracts and IDVs and the base action date for assistance. The federal fiscal year starts on 1 October, so FY2025 covers October 2024 to September 2025. Performance start and end follow the award’s current period of performance; potential end shows the date with every option exercised.

SAM.gov and agency releases

Recipient names, parent companies and identifiers in USAspending.gov come from SAM.gov registrations. The SAM.gov opportunities and entity APIs need a registered key, so this ledger cites the USAspending.gov award page, which shows the same award record with its full transaction history.

Other transaction agreements (OTAs) and scope

USAspending’s award-type reference has no OTA code (checked 11 November 2026), so OTAs reach this ledger only when an agency reports one under a contract code and its description passes the AI test. Award HHSO100201700020C shows the pattern: its description reads “OTHER TRANSACTION AUTHORITY (OTA) AWARD” and the API files it as type D, definitive contract (signed 2017, outside our window). The award type on each row is USAspending’s classification. Rows are prime awards only: subawards are left out, and classified work never reaches public spending data. Headline OTA prototype deals live on the sister ledger, AI Contracts Tracker.

Corrections and refreshes

Agencies modify awards after signing, so obligations move. Each row carries the date we last pulled it. Send a correction or a missing award to hello@federalaicontracts.com with the award ID and we refresh it. Featured placement is paid and labelled; it never changes a row or its position in the sorted table.

Reuse

The dataset is free to reuse under CC BY 4.0 with a link to federalaicontracts.com. Download it as CSV or JSON.