AI Share of SBIR/STTR Awards · National Science Foundation

## What this measures

AI-related awards as a percent of the National Science Foundation's SBIR and STTR awards in the fiscal year (at least 50 awards to print). An award counts as AI-related when its title or abstract contains any of these phrases (case-insensitive): artificial intelligence, machine learning, machine-learning, deep learning, deep-learning, reinforcement learning, supervised learning, unsupervised learning, self-supervised learning, federated learning, transfer learning, artificial neural network, deep neural network, convolutional neural network, recurrent neural network, graph neural network, natural language processing, large language model, language model, generative ai, foundation model, computer vision, expert system (with plural forms), or the capitalised word "AI", "A.I." or "LLM". Plain "neural network" does not count, because in NIH abstracts it usually means circuits in the brain. In a hand check of 40 random flagged awards from FY2015 on, 38 clearly used AI and 2 only mentioned it; of 40 random unflagged awards, 38 clearly did not involve AI and 2 might have (an autonomous vehicle and an emotion-detection system).

## Why it matters

It is a direct read of how far AI has spread through federally funded startup research: what share of the seed bets agencies place now involve it.

## How to read it

The series starts in FY2002 at 2.5%. The latest reading, for FY2025, is 28.7%. The high is 32.2% (FY2024) and the low 1.3% (FY2003).

## What it does not say

Agencies upload awards to SBIR.gov with a lag, so a fiscal year or quarter appears six months after it ends and can still rise a little on later monthly refreshes. Program authority lapsed from 1 October 2025 until the reauthorization signed on 13 April 2026 (P.L. 119-83), and few awards were made in between, so the quarters of the lapse read near zero and FY2026 will be thin for that reason as well. A keyword rule finds awards that describe AI in their title or abstract; it misses projects that use AI without naming it and counts some that only mention it. The series start in FY2002, because most awards before FY2001 have no abstract in the file.

## Source

SBIR.gov award data file (with abstracts), published by the U.S. Small Business Administration and refreshed monthly: https://data.www.sbir.gov/mod_awarddatapublic/award_data.csv. SBA: "Government information on SBA.gov is in the public domain." Source: U.S. Small Business Administration, SBIR.gov. The AI classification is Palanor's: keyword rules on each award's title and abstract (the abstracts are read, never stored).