Stanford's AI Index: in 2022, industry produced 32 notable ML models compared to academia's three, global private investment fell 26.7% YoY to $91.9B, and more
An annual report on AI progress has highlighted the increasing dominance of industry players over academia and government in deploying and safeguarding AI applications.
Context & Ripple Effects
The 2019 edition of Stanford's AI Index measured a field still defined by research output — a surge in papers and self-driving cars topping private investment. Four years later, the 2023 report marks the inversion: industry produced 32 notable ML models in 2022 against academia's three, while global private AI investment fell 26.7% year over year to $91.9B.
That combination — fewer dollars chasing models only a few players can afford to build — is the thread the later editions pick up: the 2024 report documents sharply rising training costs, and the 2025 edition finds US companies producing 40 frontier models versus 15 from China and three from Europe.
First-order effects
- Academic and government labs are effectively priced out of notable model production — with 32 of 35 notable ML models coming from industry in 2022, the frontier is now an industrial product, not a research artifact.
- Investors face a repriced market: $91.9B in global private AI investment, down 26.7% YoY, shifts funding discipline onto startups that can no longer assume capital follows any credible model roadmap.
Second-order effects
- As training costs climb and investment tightens simultaneously, model-building consolidates into the handful of firms that can fund it themselves, forcing everyone else to license, fine-tune, or build on top of industry models rather than compete at the base layer.
- Universities lose their claim as the source of state-of-the-art systems, pushing academic AI value toward evaluation, safety, and policy work — precisely the 'deploying and safeguarding' dimension the Index flags as increasingly industry-dominated too.
Third-order effects
- If the pattern holds across editions, AI becomes structurally an oligopoly market measured annually by Stanford HAI: the Index's own evolution from counting papers in 2019 to counting frontier models by country in 2025-2026 turns it into a standing scoreboard for US-China industrial AI competition rather than a research survey.
The trend: Annual AI benchmarking has shifted from measuring research activity to tracking industrial concentration, with each Index edition documenting model production consolidating further into a small set of well-funded companies.