AI Adoption by City: Do Businesses Outside San Francisco Use AI?

Quick answer

Yes. In the Census Bureau's survey of AI use in any business function (general AI, not agents), the 24 other large metros average 22.5% of businesses using AI, against 26.3% in the San Francisco-Oakland-Berkeley metro, rank 4 of 25. Silicon Valley itself is not measured.

Key takeaways

• The 24 other large metros average 22.5% of businesses using AI (95% CI 21.0 to 24.0); the San Francisco-Oakland-Berkeley metro averages 26.3%.
• That metro is not first: it is rank 4 of 25.
• The three other California metros in the file are far apart: San Diego is rank 2 of 25, Los Angeles rank 19 and Riverside rank 25.
• Across states, the Census measure and one AI vendor's usage per working-age person move together (r = 0.70), an association across states: the two sources point the same way.
• The figures are general AI use, not agents, and Silicon Valley proper is not in the file.

Do businesses outside San Francisco use AI?

Yes. In the 24 other large metros, 22.5% of businesses report using AI, against 26.3% in the San Francisco-Oakland-Berkeley metro, which is rank 4 of 25. That is the Census Bureau's survey question about AI in any business function over the last two weeks. It measures general AI use. It is general AI use, not agents.

Which large metros report the most AI use?

Denver-Aurora-Lakewood is first at 27.4% of businesses and Riverside-San Bernardino-Ontario is last at 15.6%; the San Francisco row is fourth.

Shares are the share of businesses using AI in any function (not agents), averaged over 22 two-week periods. They are unweighted means of group shares, not adjusted for industry mix or firm size, and multi-metro firms are excluded. San Antonio has fewer published periods (10). Weighting each period by its precision moves the San Francisco metro to rank 6 of 25, because neighboring metros are close.

What did we measure, and what did we start from?

We measured the share of businesses that said "Yes" to using AI in any of their business functions in the last two weeks, in the Census Business Trends and Outlook Survey. It is one yes-or-no question about AI in general. We used the 25 largest metros and all states and DC, over 22 biweekly periods from the current question wording. The sample after dropping unpublished cells was n = 538 metro-period cells and n = 1070 state-period cells. A "95% CI" is a confidence interval: a range the true figure very likely sits in, since the Census surveyed a sample.

We started from the assumption that, outside the San Francisco area, business AI is hardly used. Here is what the data showed: it is used, at levels close to the San Francisco area. The direction was visible to us before we froze the analysis plan. The cut-offs we set (a 10% floor and a ratio of 0.50 for the Census data, and 0.50 for the Anthropic index below) came after that, so they are a modest bar, not a hard test.

Technical detail: panels overlap, so the 538 metro cells carry design effect 10.7 (effective n = 50) and the 1070 state cells design effect 8.5 (effective n = 126). In plain words, the 538 readings behave like about 50 independent ones, and the 1070 state readings like about 126.

How does the San Francisco area compare with other metros?

The San Francisco-Oakland-Berkeley metro averages 26.3% using AI, a few points above the 22.5% of the other 24 metros, but it is not first. San Jose-Sunnyvale-Santa Clara is not among the 25 metros in the Census file, so Silicon Valley itself is not measured; this metro is the nearest unit.

The 3.52 percentage points and the 22.5% against 26.3% are two different comparisons: the first is within the same period against the 21 metros outside California, the second compares unweighted averages with all 24 other metros. Do not subtract one from the other.

The ratio of 0.82 has a bootstrap interval of 95% CI 0.76 to 0.91, but it is too narrow: it varies only the 24 comparison metros and ignores the sampling error of the single San Francisco estimate.

Treat every San Francisco and California difference here as a description, not a verdict. Only one metro, or one state, is being singled out, so we make no claim that a gap is statistically meaningful. These differences are not adjusted for industry mix or firm size, so they may reflect what kinds of businesses are there. Multi-state and multi-metro firms, which include many large tech firms, are left out of the metro and state tables.

Are the other California metros alike?

No. The three other California metros in the file are very different from one another. San Diego is rank 2 of 25, Los Angeles rank 19 and Riverside rank 25. As a group they come out 2.12 percentage points below the 21 metros outside California, with 95% CI -8.10 to 3.87. That interval includes zero and is itself approximate, since only three metros are compared, so the honest reading is no clear difference.

Is California ahead of other states?

California is a couple of points above other states, as a description. The difference is 2.04 percentage points in the share using AI in any function, within the same period, against the other 49 states and DC, across 1070 state-period cells (about 126 independent ones). It is one state, so it carries no interval, and it is not adjusted for industry mix or firm size; multi-state firms are excluded.

One more check uses the Anthropic Economic Index, which reports Claude.ai chat and Cowork usage (Free, Pro and Max plans; not the API) per working-age person, for one vendor. It is not a measure of businesses. California's index is 1.59, where 1.0 means usage in proportion to its working-age population, and it ranks 2 of 51. The median across the 49 other states and DC is 0.765, and 38 of 50 of them have an index of at least 0.50.

Do the Census and Anthropic measures move together across states?

They rise together: across states, the Census share and the Anthropic usage index rise together, with r = 0.70 (95% CI 0.50 to 0.82), n = 50. In words, states where more businesses report using AI are also, on the whole, states where people use this one AI product more. That is an association across states, not a test of the assumption we started from. The 50 are 49 states and DC; North Dakota had too few published periods. The Census side uses periods 202608 to 202611 (April and May 2026), the same months as the Anthropic side.

Technical detail: the Spearman rank correlation is rho = 0.72 (95% CI 0.52 to 0.85), the distance correlation is distance correlation 0.70, and with 50 states the smallest correlation we could detect is minimum detectable r = 0.39.

A correlation says the two measures move together; both may track industry mix, how urban a state is or internet access. It does not say that either explains the other.

What can this data not show?

It cannot show agents, value, or Silicon Valley itself. The limits, in full:

• San Jose-Sunnyvale-Santa Clara is not in the Census metro file, so Silicon Valley is not measured; the nearest unit is the San Francisco-Oakland-Berkeley metro.
• Only four California metros are published; San Francisco and California are each a single unit.
• The survey asks about AI in any business function, not agents, and nothing about value or results.
• Multi-state and multi-metro firms, which include many large tech firms, are excluded from the metro and state tables, as are sole proprietors without employees. Small towns, rural areas and metros outside the 25 largest are not in the sample.
• The cut-offs were fixed after the direction was visible (see the method). Biweekly panels overlap, so cells are not independent.
• Metros and states are not adjusted for industry mix or firm size.
• State comparison shows association only; the Anthropic index is one vendor's chat plans, not businesses.
• Window: Census periods 202524 to 202619 (collection December 2025 to September 2026), files retrieved 6 October 2026; Census refreshes every two weeks, so a download today adds periods.

For employee count and industry, see our earlier look at AI use by business size and industry.

How can you check these numbers?

Open the public files and filter to the AI question. The files are the Top 25 metro workbook and the State workbook, from the Business Trends and Outlook Survey (U.S. Census Bureau, retrieved 6 October 2026). The sheets with estimates and standard errors hold the figures; the methodology document explains which firms are excluded.

The Anthropic data are the sixth release of the Economic Index dataset, licensed CC-BY. Cite it as Massenkoff et al., "Anthropic Economic Index report: Cadences", 26 June 2026, published here.

What can you do with this?

Find your metro in the table. If you are in one of the 25 metros, find your row: shares run from 15.6% to 27.4%, not adjusted for industry mix or firm size. Outside those metros, the state file is the closest published number. That says nothing about whether a tool pays off for you; for that, see our guide to tools sold as hands-off AI, and start with one task you already do every week.

Analysis by Deivy Hernandez, from public data files.

FAQ

Does this tell me what kind of AI businesses outside the Bay Area use?

No. The Census question is about AI in any business function, not agents or any one tool. The data show some AI is in use, not what kind or how deeply.

Is my own city one of the 25 metros?

Only if it is among the 25 largest. The table above lists them, and the state file covers all states and DC. If your metro is not listed, the state figure is the closest published number.

Why do the Census and the Anthropic numbers both appear?

They measure different things: the Census asks employer businesses whether they used AI; the Anthropic index counts one vendor's chat usage per working-age person. We compared them across states only to see whether they point the same way.

Does a higher share mean AI pays off there?

No. The survey only asks whether a business used AI in any function in the last two weeks. It says nothing about value, results or which tools.