Report
The October 2026 Beige Book through the lens of AI
Economic growth remained broad but softened across sectors as AI-related Beige Book language concentrated on data-center construction, power, labor, and productivity.
Euphemios12 min read
Executive summary
The October Beige Book describes an economy that was still growing but with less breadth than in July. The mean district score for overall activity slipped to 0.75 from 0.83 in September and 0.92 in July. Consumer spending moved closer to flat, manufacturing remained positive but cooled, and real estate and construction stopped contracting on average without returning to July's strength. Prices were the clearest acceleration: their mean score rose to 1.25, from 1.08 in each prior edition. A positive prices score means prices increased; it is not a judgment that conditions improved. (Federal Reserve Board, Beige Book editions of July, September, and October 2026; Euphemios analysis.)
AI appeared less as a stand-alone technology story than as an economy-wide input. The October reports contained 53 manually reviewed occurrences of AI, automation, data centers, or related terms, nearly matching September's 54 and above July's 42. Data-center language accounted for 30 October occurrences, and construction or electricity demand accounted for 29 of all October contexts. Labor passages described both restrained hiring and demand for scarce skills; productivity passages described higher output or efficiency without equivalent head-count growth. These are textual occurrences, not projects, dollars, jobs, or causal estimates. (Federal Reserve Board, Beige Book, October 14, 2026; Euphemios analysis.)
The qualitative reports align directionally with published evidence showing rapid physical investment: five cloud and computing companies recorded $230.42 billion in quarterly cash purchases of property and productive assets, versus $105.77 billion a year earlier, while private data-center construction reached a $96.46 billion annual pace in September, up from $50.70 billion. The sources do not prove that AI caused the Beige Book sector patterns or that every dollar was AI-related. They show that the anecdotes and independently measured investment are pointing toward the same infrastructure buildout. (Euphemios analysis, 2026-11-20.)
Key findings
- Overall activity averaged 0.75 in October, versus 0.83 in September and 0.92 in July. All readings remained positive, but the direction across the three editions was softer. (Federal Reserve Board, three Beige Book editions; Euphemios analysis.)
- Consumer spending fell to a 0.09 mean score from 0.18 in September and 0.36 in July, while prices rose to 1.25 from 1.08. Household price sensitivity appeared repeatedly across districts. (Federal Reserve Board, three Beige Book editions; Euphemios analysis.)
- Manufacturing averaged 0.75, down from 0.92 in both earlier editions; real estate and construction averaged 0.00 after −0.08 in September and 0.58 in July. Data-center, defense, power, and advanced-manufacturing demand supported otherwise uneven conditions. (Federal Reserve Board, three Beige Book editions; Euphemios analysis.)
- October contained 53 reviewed AI-related occurrences: 30 data-center terms, 19 AI terms, and 4 automation terms. Construction or electricity demand supplied 29 contexts. (Federal Reserve Board, Beige Book, October 14, 2026; Euphemios analysis.)
- Data-center language appeared in 11 of 12 district reports. Dallas had 11 reviewed AI-related occurrences, Atlanta 9, and Cleveland 8, showing that the theme was geographically broad but concentrated. (Federal Reserve Board, Beige Book, October 14, 2026; Euphemios analysis.)
- Measured investment moved in the same direction as the anecdotes: cloud-company cash purchases were $230.42 billion versus $105.77 billion a year earlier, and private data-center construction rose 90.26% to a $96.46 billion annual pace. (Euphemios analysis, 2026-11-20.)
- The labor signal was mixed rather than uniformly negative: October included 13 hiring-or-labor contexts and 4 productivity contexts, while the district employment mean held at 0.58. (Federal Reserve Board, Beige Book, October 14, 2026; Euphemios analysis.)
Growth remained broad but narrowed
The three-edition comparison shows deceleration rather than contraction. Overall activity remained positive in most districts, but the mean fell in each edition. October's 0.75 reflected positive readings in Boston, New York, Philadelphia, Cleveland, Richmond, Atlanta, St. Louis, Minneapolis, and Dallas, with Chicago, Kansas City, and San Francisco characterized as flat or steady. The national summary similarly described moderate growth in two districts, modest growth in three, slight growth in four, and no significant change in three. October national summary
The cross-sector movement was less uniform. Consumer spending weakened to 0.09, as high-income and luxury spending often held up while value-oriented and discretionary demand softened. Manufacturing stayed positive at 0.75, helped in several districts by data centers, defense, energy, semiconductors, and advanced manufacturing. Real estate and construction improved from −0.08 to 0.00, but that average combined data-center and infrastructure strength with weak residential conditions. Banking and finance averaged 0.45 across the districts with directly comparable sections, down from 0.50 in September and 0.64 in July.
Employment held at 0.58, matching September and above July's 0.50. Wage scores were firmer at 0.92, although Boston described wage growth as muted. Price pressure moved the other way: the prices mean reached 1.25, with robust readings in Cleveland, St. Louis, and Dallas. Fuel, freight, materials, insurance, and specialized labor were recurring sources of pressure.
Average district sentiment by sector across three editions
- July 15
- September 2
- October 14
| Sector | July 15 | September 2 | October 14 |
|---|---|---|---|
| Overall activity | 0.92 | 0.83 | 0.75 |
| Consumer spending | 0.36 | 0.18 | 0.09 |
| Manufacturing | 0.92 | 0.92 | 0.75 |
| Real estate and construction | 0.58 | -0.08 | 0 |
| Banking and finance | 0.64 | 0.5 | 0.45 |
| Employment | 0.5 | 0.58 | 0.58 |
| Wages | 1 | 1 | 0.92 |
| Prices | 1.08 | 1.08 | 1.25 |
Means exclude missing district-sector observations rather than treating them as zero.
Source · Federal Reserve Board, Beige Book; Euphemios analysis · Official HTML for July 15, September 2, and October 14, 2026; fetched November 20, 2026 CST · As of
October's district-sector scorecard
The table reports the standardized sectors used for cross-district comparison. A dash means that the district did not publish a directly comparable named section; it is not a zero. The full score pack retains every additional named sector, including services, transportation, energy, agriculture, business spending, worker experience, and community conditions, with its supporting phrase and official URL.
| District | Overall | Consumer | Manufacturing | Real estate | Banking | Employment | Wages | Prices |
|---|---|---|---|---|---|---|---|---|
| Boston | 1 | 0 | 1 | 0 | 1 | 1 | 0 | 1 |
| New York | 1 | 1 | 1 | 0 | 1 | 0 | 1 | 1 |
| Philadelphia | 1 | — | 1 | −1 | 1 | 1 | 1 | 1 |
| Cleveland | 1 | 0 | 1 | 1 | 2 | 1 | 1 | 2 |
| Richmond | 1 | 1 | 0 | 1 | 0 | 1 | 1 | 1 |
| Atlanta | 1 | 1 | 0 | 1 | 1 | 0 | 1 | 1 |
| Chicago | 0 | 0 | 1 | −1 | −1 | 1 | 1 | 1 |
| St. Louis | 1 | 0 | 1 | −1 | 0 | 0 | 1 | 2 |
| Minneapolis | 1 | −1 | 1 | 1 | — | 1 | 1 | 1 |
| Kansas City | 0 | −1 | 0 | 1 | 0 | 0 | 1 | 1 |
| Dallas | 1 | 1 | 2 | −1 | 1 | 1 | 1 | 2 |
| San Francisco | 0 | −1 | 0 | −1 | −1 | 0 | 1 | 1 |
October district sentiment by sector
- Overall activity
- Manufacturing
- Real estate and construction
- Employment
- Wages
- Prices
| Federal Reserve district | Overall activity | Manufacturing | Real estate and construction | Employment | Wages | Prices |
|---|---|---|---|---|---|---|
| Boston | 1 | 1 | 0 | 1 | 0 | 1 |
| New York | 1 | 1 | 0 | 0 | 1 | 1 |
| Philadelphia | 1 | 1 | -1 | 1 | 1 | 1 |
| Cleveland | 1 | 1 | 1 | 1 | 1 | 2 |
| Richmond | 1 | 0 | 1 | 1 | 1 | 1 |
| Atlanta | 1 | 0 | 1 | 0 | 1 | 1 |
| Chicago | 0 | 1 | -1 | 1 | 1 | 1 |
| St. Louis | 1 | 1 | -1 | 0 | 1 | 2 |
| Minneapolis | 1 | 1 | 1 | 1 | 1 | 1 |
| Kansas City | 0 | 0 | 1 | 0 | 1 | 1 |
| Dallas | 1 | 2 | -1 | 1 | 1 | 2 |
| San Francisco | 0 | 0 | -1 | 0 | 1 | 1 |
Six sectors reported by all 12 districts. Consumer and banking coverage is retained in the score pack. A positive prices score means prices rose.
Source · Federal Reserve Board, Beige Book; Euphemios analysis · Official HTML for July 15, September 2, and October 14, 2026; fetched November 20, 2026 CST · As of
The scorecard makes the composition of growth visible. Dallas combined moderate overall growth with a sharp manufacturing pickup and strong price pressure, while its housing market remained weak. Cleveland had robust loan demand and cost pressure alongside moderate manufactured-goods demand. Chicago and San Francisco were flatter overall, with softer real estate and tighter or weaker financial conditions. Atlanta's overall activity, consumer spending, housing, lending, and energy were positive, while manufacturing and employment were flat.
AI language shifted toward the physical buildout
October's 53 occurrences were almost unchanged from September's 54, but the mix shifted. Data-center terms increased to 30 from 26 in September and 18 in July. AI terms declined to 19 from 25, and automation terms rose to 4 from 3. That mix matters: the Beige Book was describing transformers, power, water, permits, construction materials, skilled trades, lending, freight, and industrial demand, not only software adoption.
AI-related term occurrences by district and edition
- July 15
- September 2
- October 14
| Federal Reserve district or national summary | July 15 | September 2 | October 14 |
|---|---|---|---|
| National summary | 4 | 4 | 1 |
| Boston | 3 | 1 | 1 |
| New York | 2 | 5 | 3 |
| Philadelphia | 4 | 1 | 3 |
| Cleveland | 3 | 10 | 8 |
| Richmond | 1 | 9 | 4 |
| Atlanta | 8 | 8 | 9 |
| Chicago | 3 | 2 | 2 |
| St. Louis | 0 | 4 | 4 |
| Minneapolis | 3 | 3 | 1 |
| Kansas City | 1 | 3 | 2 |
| Dallas | 6 | 1 | 11 |
| San Francisco | 4 | 3 | 4 |
One record per boundary-safe term occurrence; multiple terms in one sentence count separately.
Source · Federal Reserve Board, Beige Book; Euphemios analysis · Official HTML for July 15, September 2, and October 14, 2026; fetched November 20, 2026 CST · As of
The October context classification reinforces that interpretation. Construction or electricity demand accounted for 29 occurrences, hiring or labor for 13, investment or capital spending for 6, productivity for 4, and other uses for 1. Atlanta referred to “AI implementation broadened across surveyed firms.” Cleveland contacts said automation and AI had “reduced staffing needs.” St. Louis described automation and data integration producing “meaningful productivity gains.” Dallas reported power, permit, water, labor, recruiting, and semiconductor effects around the same investment cycle.
The district distribution was broad. Every district except Minneapolis used a data-center term in October. Minneapolis still included an artificial-intelligence labor passage, reporting that the technology was not causing layoffs but was slowing job creation in some areas. The national summary separately mentioned firms increasing productivity by investing in AI. Because the national summary can restate district evidence, it is counted as its own scope rather than treated as an additional district.
The Beige Book and published AI evidence agree on direction, not cause
The strongest agreement is around physical investment. The Beige Book repeatedly described data-center building, electrical equipment constraints, power demand, specialized labor, freight, lending, and construction inputs. Separately, private data-center construction reached a $96.46 billion seasonally adjusted annual pace in September, compared with $50.70 billion a year earlier, a 90.26% increase. The construction measure covers building work and excludes servers and racks. The text count measures language, not spending. Their common direction supports the conclusion that the buildout was economically visible across regions, while neither source assigns the buildout wholly to AI. (Euphemios analysis, AI across sectors.)
Cloud-company spending supplies a second comparison. Microsoft, Alphabet, Amazon, Meta, and Oracle recorded $230.42 billion in combined quarterly cash purchases of property or productive assets, up from $105.77 billion in the comparable periods. Beige Book references to technology investment, AI tools, data preparation, automation equipment, and construction are consistent with that larger capital cycle. The company figures cover all operations, use issuer-specific cash-flow tags, and omit some noncash financing, so they cannot be converted into district AI spending.
Other published measures show diffusion beyond construction. Artificial-intelligence wording appeared in 3,810 of 6,258 original annual filings through 2026-11-20, or 60.88%, versus 3,067 of 6,237, or 49.17%, through the same date a year earlier. Among bank filings, 265 of 328, or 80.79%, matched the phrase. Federal contract transactions with explicit artificial-intelligence wording carried $498.0 million in fiscal-year net obligations, up from $412.8 million, or 20.64%. These measures agree that AI is present across corporate and public-sector language, but they do not establish deployment, budgets, savings, or economic effects inside any district. (Euphemios analysis, AI across sectors.)
The Colibrì event study adds a market comparison that does not line up cleanly with the operating anecdotes. Across 18 retained events, the AI-stock basket's five-session excess return averaged −0.19 percentage points; power stocks averaged −1.71 points, while large cloud operators averaged +0.68 points. Beige Book infrastructure demand therefore coexisted with mixed short-window relative returns. This is not a contradiction: the event study was small, overlapping news affected prices, and stock returns do not measure construction activity, facility utilization, or investment returns. (Euphemios analysis, Colibrì event study.)
Labor and productivity point to reallocation, not one outcome
October's labor passages do not support a single claim that AI was broadly eliminating work. New York contacts restrained hiring as they integrated AI. Cleveland banking contacts reported reduced staffing needs. Chicago described attrition alongside AI productivity gains. Dallas reported fewer entry-level hires in some cases, but also recruiting demand and scarce data-center labor. San Francisco described higher pay for electricians because of competition from data-center projects. Minneapolis said artificial intelligence was not causing layoffs but could slow job creation.
The aggregate scores remained positive: employment averaged 0.58 and wages 0.92. That combination is compatible with task substitution in some office functions, continuing demand for specialized technical and construction roles, and modest overall hiring. It does not measure net AI-attributable employment. The Beige Book contacts are not a random sample, and the relevant counterfactual—what staffing would have been without these technologies—is absent.
Methodology and definitions
The analysis uses three Federal Reserve Beige Book releases. Each edition is one release with a national summary and reports from the 12 Reserve Bank districts, the Federal Reserve’s regional reporting areas. The September 2 release is labeled “August 2026.” July 15, September 2, October 14, October national summary
A sector groups district headings on the same topic. The score uses a characterization, the wording that describes the direction and strength of conditions. The ordered rules check for strong declines and increases first, then modest or moderate wage inflation, increases, declines, flat or mixed language, and weak conditions. The first match determines the score on a −2 to +2 scale, with zero when no rule matches. Manual readings take precedence and use a Federal Reserve excerpt or a labeled summary. Positive prices scores indicate rising prices or cost pressure. A missing sector is omitted.
An occurrence is one reviewed match to an AI-related term. The scan counts each match in the main text of the national summaries and district reports, including repeated terms. It ignores case and checks for a boundary at each end. Each occurrence receives one primary context. Wording changes can affect scores and counts, and national summaries can repeat district material. See Limits.
To reproduce the scores, map each archived district heading to its sector and score the lead sentence of each named section. Use the Labor Markets lead for employment, and select the wage sentence from the district summary and Labor Markets using the wage-pattern order on the full methods page. Apply the manual readings and store the first 14 words of each automatic phrase. A district-sector record contains the edition, district, sector, original heading, score, supporting phrase, and source URL. Compute each unweighted mean by dividing the sum by the number of scores. To reproduce the occurrence counts, scan the archived report text with the published pattern, review the matches in context, and tally by term and edition.
For the scoring rules, manual readings, district coverage, and term counts, see Full methods and sources →.
External comparisons
All external figures are reproduced from the two published Euphemios reports dated 2026-11-20. The comparison is descriptive. Beige Book text, company cash-flow facts, construction spending, federal obligations, filing searches, and event-window returns have different units, populations, and observation periods.
Limits
Beige Book evidence is qualitative and based on contacts selected by each Reserve Bank rather than a random sample. Scores compress language and cannot recover the size of an economic change. A mean gives each available district equal weight, regardless of population or output. Composite sector sections can contain offsetting developments even when the lead sentence has one direction.
Occurrence counts are sensitive to wording. Repeated terms can refer to one project, and the national summary can restate district material. The counts do not measure adoption, investment, electricity use, productivity, employment, or value added. Context classification assigns one primary category even when a passage spans several.
The external comparisons are correlated signals. Cloud-company purchases are company-wide, construction spending excludes equipment, filing text can describe risks or plans, and federal obligations do not equal payments or AI-only spending. The event study is small and cannot isolate Colibrì from other news. None of these sources establishes that AI caused the sector scores.
FAQ
What does a sentiment score measure?
It translates the direction and intensity of a district's written characterization into a common scale. It does not measure output growth, dollars, jobs, or statistical significance.
Why is a positive prices score not necessarily favorable?
The score records direction. A positive value means reported prices or cost pressure rose; its economic effect depends on who bears those costs and how demand responds.
Why can an AI occurrence count exceed the number of passages?
The unit is a matched term. A sentence containing both AI and automation contributes two occurrences, each manually checked and classified.
Does the Beige Book show that AI is reducing employment?
It reports restrained hiring or lower staffing needs in some contacts and stronger demand for specialized labor elsewhere. It does not provide a net AI-attributable employment estimate.
How do the data-center anecdotes compare with construction data?
Both point to a large buildout. The anecdotes identify regional channels such as power, materials, labor, freight, and permits, while the construction series measures building work at a national annual pace.
Does the analysis show that the investment cycle is sustainable?
No. The evidence describes spending, construction, language, and short-window market movements. It does not measure future utilization, prices for computing services, or realized returns on new capacity.
Company and network names are trademarks of their owners. They are mentioned as facts about public filings and data, not as a relationship or an endorsement.
Sources
- Data as ofFederal Reserve Board, Beige Book (October 14, 2026)Run Official HTML national summary and 12 district reports, fetched November 20, 2026 CST
- Data as ofFederal Reserve Board, Beige Book (September 2, 2026)Run Official HTML national summary and 12 district reports, fetched November 20, 2026 CST
- Data as ofFederal Reserve Board, Beige Book (July 15, 2026)Run Official HTML national summary and 12 district reports, fetched November 20, 2026 CST
- Data as ofEuphemios analysis, AI across sectorsRun Published November 20, 2026; public-record evidence through November 20, 2026
- Data as ofEuphemios analysis, Colibrì event studyRun Published November 20, 2026; daily closes through November 19, 2026
- Data as ofFederal Reserve Board, Beige Book; Euphemios analysisRun Official HTML for July 15, September 2, and October 14, 2026; fetched November 20, 2026 CST