Industry produced most notable AI models Stanford counted in 2023

The signal

Stanford HAI's 2024 AI Index Report, published 15 April 2024 as part of the index covering Signal Atlas's AI sector, counted 51 notable machine-learning models produced by industry in 2023, against 15 from academia and 21 from industry-academia collaborations. On that count, industry alone produced more notable models than academia and joint efforts combined. The report does not set out every criterion for 'notable' in the passage cited here, so the figure should be read as this index's count under its own method, not a universal measure of research output.

The evidence

A companion HAI article, published the same day, adds a related but distinct figure: industry accounted for 72% of new foundation models in 2023, a narrower model category than the 51/15/21 count of all notable machine-learning models. The two figures should not be treated as interchangeable measures of the same thing. A separate tracker, Epoch AI's notable-models dataset, monitors more than 3,600 models against its own criteria -- state-of-the-art benchmark performance, high citation counts, historical significance, or more than a million monthly active users -- and does not publish an industry-versus-academia split on the page examined here. That difference matters: two reputable trackers can each report a defensible 'notable models' count without agreeing on what qualifies, which is itself evidence of how contested the underlying category is.

Timeframe and confidence

This is a single year's snapshot rather than a time series; the passages consulted here give no equivalent industry-versus-academia breakdown for 2021 or 2022 against which to compare 2023. Read as an editorial framework, a one-year gap of this size is suggestive of a resourcing divide -- large models increasingly require compute budgets that few universities carry alone -- but a single year cannot establish whether the gap is widening, stable, or already narrowing.

What would change the reading

A published multi-year series, using one consistent definition of 'notable,' showing academia's share falling further would support a compute-divide reading. A rebound in academic-led notable models, or growth in industry-academia collaboration's share specifically, would weaken it.

• Does the counting method change between report editions in ways that affect year-over-year comparison?
• How much of the industry total comes from a small number of well-funded labs rather than the sector broadly?
• What happens to research verification when the costliest models are produced outside academic peer-review norms?

The report's count is specific and dated. Whether it marks the start, middle, or plateau of a shift depends on data this edition alone cannot supply.

Source trail

1. The 2024 AI Index Reporthai.stanford.edu · Source publication: 2024-04-15 · Retrieved 2026-09-16

   States the 2023 counts of 51 industry, 15 academic, and 21 industry-academia collaboration notable machine-learning models.

2. AI Index: State of AI in 13 Chartshai.stanford.edu · Source publication: 2024-04-15 · Retrieved 2026-09-16

   Adds the distinct figure that industry produced 72% of new foundation models in 2023, and gives per-company model counts since 2019.

3. Data on AI Modelsepoch.ai · Source publication: not established · Retrieved 2026-09-16

   Documents a differently defined notable-models dataset (over 3,600 models) as a methodological contrast, without an industry/academia split.

Read across the evidence

• AlexNet's 2012 entry cut the ImageNet error rate by ten points
• The AI Index put a boundary around cheaper model use
• A publisher's lawsuit made AI training data a licensing question

Adoption & operating reality · All 100 historical records