The most reliable record of what changed in AI during 2022 is the 2023 AI Index from the Stanford Institute for Human-Centered AI, which counted rather than guessed. It found that industry produced 32 significant machine learning models in 2022 against three from academia, that global private investment in AI fell 26.7 percent to USD 91.9 billion, and that passed bills mentioning artificial intelligence rose from one in 2016 to 37 in 2022 across 127 countries.
Key takeaways
- Companies, not universities, built most of the notable models in 2022: 32 industry models against three from academia.
- Global private AI investment dropped to USD 91.9 billion, down 26.7 percent from 2021 but 18 times the 2013 level.
- The share of companies using AI has more than doubled since 2017 but has stayed between 50 and 60 percent in recent years.
- Recorded AI incidents and controversies grew 26 times between 2012 and 2022, and public opinion differed sharply by country.
- The 2026 AI Index shows what moved on since: industry builds over 90 percent of notable frontier models, organizational adoption reached 88 percent, and documented incidents hit 362 in 2025.
What the AI Index is and why it suits a trends question
Articles titled "the biggest AI trends of a year" usually rest on the author's opinion. The AI Index counts instead. Each annual report collects data on research output, model development, investment, jobs, policy, and public opinion, and states where each number comes from. The 2023 edition covers 2022, so it can be checked against the year it describes. Its summary lists ten takeaways, and seven of them describe movements that can be written as a number with a direction. Those seven are used below. They are a reading of one report, not a ranking of everything that happened in AI that year.
Who built the models: industry, by a wide margin
The report counts "significant" machine learning models, a label that the index assigns using its own criteria. For 2022 it found 32 produced by industry and three produced by academia. The practical meaning is about resources. Training large models takes data and computing capacity that most university labs do not have, so the notable systems of that year came mostly from companies. The same report notes that year-over-year improvement on many benchmarks was marginal, and that benchmarks were saturating faster. A benchmark is saturated when top systems score so near the ceiling that the test no longer separates them, which is a limit of the test as much as of the models.
Money and adoption: a dip in funding, a plateau in use
Global private investment in AI was USD 91.9 billion in 2022, down 26.7 percent from 2021, though still 18 times the 2013 figure. A fall after a record year says little about whether the technology works. Funding follows markets as well as results. On use, the index reports that the share of companies adopting AI has more than doubled since 2017 but has held between 50 and 60 percent in recent years. In the United States, the average share of job postings that were AI-related rose from 1.7 percent in 2021 to 1.9 percent in 2022. The rise is 0.2 percentage points, small but measured across many sectors.
Policy and incidents: two counts that moved fast
Lawmaking attention grew quickly. Passed bills containing the words "artificial intelligence" went from one in 2016 to 37 in 2022 across 127 countries, and legislative mentions of AI in 81 countries grew nearly 6.5 times since 2016. A passed bill that mentions AI can be narrow, so the count shows attention, not the strength of any rule. The index also reports that the AIAAIC database of AI incidents and controversies grew 26 times from 2012. The caveat is that such a database counts what was reported, and more reporting can come from more use and more public awareness as well as from more harm.
Opinion: the same technology, different verdicts
Surveys cited in the report found that 78 percent of respondents in China said AI products and services had more benefits than drawbacks, against 35 percent of respondents in the United States. The gap shows how much attitudes depend on the country, the question asked, and what people have personally seen of the technology. The index also reports that training the BLOOM model emitted 25 times more carbon than one air traveler on a one-way flight from New York to San Francisco, a comparison that shows the scale of one training run without settling the wider question of AI energy use.
What changed since: the 2026 AI Index
Four years on, the same report series gives a few points of comparison. Its 2026 edition covers 2025 and states that industry produced over 90 percent of notable frontier models that year, which extends the 2022 pattern of company-built models. Organizational adoption now stands at 88 percent, well above the 50 to 60 percent plateau described for 2022, although the editions may define adoption differently. The edition counts 362 documented AI incidents in 2025, up from 233 in 2024. In U.S. private AI investment alone, it reports USD 285.9 billion for 2025, a figure that is not comparable with the 2022 global total of USD 91.9 billion because the geographic scope differs.
The 2026 report also records a gap in opinion between experts and the public: 73 percent of experts expect AI to affect how people do their jobs positively, against 23 percent of the public. The 2022 figures above are therefore a starting point, not the current state, and the 2026 numbers should be read from the report itself.
The readings in one table
| Reading | 2022 figure | Limit of the figure |
|---|---|---|
| Industry leads model building | 32 industry models, three academic | "Significant" is the index's own label |
| Benchmark gains are slowing | Marginal year-over-year gains on many tests | Saturated tests stop separating systems |
| Investment fell | USD 91.9 billion, down 26.7 percent | Reflects markets as well as results |
| Adoption plateaued | 50 to 60 percent of companies | Depends on how "adopting AI" is defined |
| Lawmaking attention rose | 37 passed bills, from one in 2016 | Counts mentions, not rule strength |
| Recorded incidents rose | 26 times the 2012 level | Counts reports, not total harm |
| Opinion split by country | 78 percent in China, 35 percent in the US | Survey wording and sampling differ |
How to read any list of AI trends
- Is each trend attached to a number, a year, and a named source?
- Does the number come from a stated method, or from the author's impression?
- Does the list describe the year it claims to cover, or later events written backward?
- Are limits stated, such as how a term like "significant" or "adoption" is defined?
FAQ
Did investment in AI collapse in 2022?
No. The AI Index reports USD 91.9 billion in global private investment, down 26.7 percent from 2021. That is a decline from a high year, and the total was still 18 times the 2013 level.
Does 50 to 60 percent adoption mean half of all workers use AI?
No. The figure describes the share of companies reporting that they adopt AI, which says nothing about how many employees use it or how often.
Why did companies build most of the notable models?
The report records the counts, not the cause. Training large models requires large datasets and computing capacity, which companies were better placed to pay for than most academic labs.
Are the 2022 figures still current?
No. They describe one year. The 2026 edition of the AI Index covers 2025, and later editions revise methods and update the counts, so a 2022 number should be cited with its year.
Related reading
- Review of AI trends in education
- A trend report and its predictions
- Trustworthy AI and the challenges around it
- Major companies in the AI market
More in AI Basics.
Last verified: October 2026
Sources
- Artificial Intelligence Index Report 2023, top takeaways, Stanford Institute for Human-Centered Artificial Intelligence, 2023
- The 2026 AI Index Report, top takeaways, Stanford Institute for Human-Centered Artificial Intelligence, 2026