State of AI Industries 2026
Adoption, spending and measured outcomes in healthcare, finance, software, retail, manufacturing, education, legal, marketing and government, 2024 to 2026
Thank you for reading this post, don't forget to subscribe!
- Ramp’s AI Index showed 55.9% of US businesses on its platform paying for AI tools in September 2026; in August, technology firms were at 80.9% and accommodation and food services at 33.1%, a gap of about 2.4 to 1.
- The US Census Bureau’s Business Trends and Outlook Survey found 19.8% of US employer businesses using AI as of 3 May 2026, with the information sector at 39.7% and finance and insurance at 33.9%.
- The American Medical Association found that 81% of US physicians used AI in their practice in a survey of 1,692 physicians fielded from 15 January to 2 February 2026, up from 66% in November 2024.
- The FDA’s AI-Enabled Medical Device List reached 1,614 entries in its 4 September 2026 update, after a record 335 AI device authorizations with 2025 decision dates (other counts of 2025 range from 295 to 333).
- HEPI found that 95% of UK full-time undergraduates used AI in December 2025, and Clio’s 2026 Legal Trends Report found that 85% of legal professionals reported some level of AI adoption at their firms.
- IDC estimated that banks spent about $31.3 billion on AI in 2024, more than any other industry, while Menlo Ventures found healthcare the largest buyer of vertical generative-AI software at $1.5 billion in 2025.
- Controlled studies show smaller gains than adoption rates suggest: a JAMA multisite study published in April 2026 found ambient AI scribes cut documentation time by about 16 minutes a day, and METR’s July 2025 trial found experienced developers 19% slower with early-2025 AI tools.
Download the data (CSV) · 28 metrics with source, date and confidence for every value. Free to quote with attribution and a link.
- 1. Adoption across sectors — Every comparable survey ranks information, finance and professional services first and hospitality, construction and retail last, with a gap of more than two to one
- 2. Healthcare — Healthcare moved from pilots to practice: 81% of US physicians use AI and the FDA authorized a record number of AI devices in 2025, but measured gains remain narrow
- 3. Financial services and software — Finance and software lead on adoption and spending, but the best controlled evidence on productivity is mixed
- 4. Education and legal — In education and law, individual use is near-universal while institutions, assessment rules and verification lag behind
- 5. Retail, marketing, manufacturing and government — Outside the knowledge industries adoption is lower and harder to measure, and the best data track customer behaviour rather than firm deployment
- 6. Spending and outcomes — Spending is concentrating in banking, software and healthcare, but sector-level returns remain mostly unmeasured
- Comparison
- The numbers, 2024–2026
- Outlook: what to watch in 2027
- FAQ · About the research
Every comparable survey ranks information, finance and professional services first and hospitality, construction and retail last, with a gap of more than two to one
Three survey families measure AI adoption by industry, and each answers a different question. Ramp’s AI Index records whether a business on its card and bill-pay platform pays for an AI subscription or contract. The US Census Bureau’s Business Trends and Outlook Survey (BTOS) asks a representative sample of employer businesses whether they used AI in any of 15 business functions. Eurostat asks EU enterprises with ten or more employees whether they use specific AI technologies. Their levels differ by a factor of two to three: in mid-2026 Ramp put paid adoption at 55.9%, the Census Bureau put use at 19.8% and Eurostat’s latest reading, for 2025, was 19.95%. Because of that spread, this report never places two of these surveys on one chart. Within each survey the sector ranking is consistent, and that ranking is the most reliable finding available.
Ramp’s August 2026 breakdown is the most detailed single-survey view. Technology firms led at 80.9%, followed by finance at 73.7% and manufacturing at 61.1%. Retail stood at 48.8%, health care at 43.4%, construction at 42.2% and accommodation and food services at 33.1%, against 55.2% for all businesses. The top sector is about 2.4 times the bottom one. Manufacturing’s position above retail and health care is notable: Ramp counts any paid AI tool, and manufacturers buy software for design, maintenance and back-office work even where the factory floor is unchanged. Health care’s low firm-level figure sits beside very high individual use by physicians, a contrast discussed in chapter 2. Ramp’s customers lean towards technology-forward and venture-backed firms, so the absolute levels overstate adoption in the wider economy.
The Census Bureau’s representative survey gives the same order at lower levels. As of 3 May 2026, 39.7% of information-sector businesses and 33.9% of finance and insurance businesses used AI, against a national rate of 19.8%; retail trade was at about 14%. Firm size matters as much as industry: 37% of businesses with 250 or more employees used AI. Eurostat’s 2025 data show the same shape in Europe. The information and communication sector reached 62.52% and professional, scientific and technical services 40.43%, while every other activity was below 25%, with construction at 10.79%. Only the top two EU sectors could be verified at sector level; other values circulating online could not be matched to Eurostat’s table and are not used here.
View the data
| Paying for AI tools, Aug 2026 | |
|---|---|
| Technology | 80.9 |
| Finance & insurance | 73.7 |
| Manufacturing | 61.1 |
| All businesses | 55.2 |
| Retail | 48.8 |
| Health care | 43.4 |
| Construction | 42.2 |
| Accommodation & food | 33.1 |
Notes: Single survey only. Ramp assigns firms to sectors with a model based on NAICS; its own table labels the top sector ‘information’ at 80.0% for August 2026 (79.9% in September). Ramp’s customer base leans towards technology-forward and venture-backed firms, so levels are higher than in representative surveys such as Census BTOS (19.8% nationally, May 2026). Ramp expanded its sample in June 2026 and revises history. Ratio 80.9 / 33.1 = 2.4 (DERIVED).REPORTAI report-ai.org
The trend lines show steady rather than explosive growth. Ramp’s overall measure rose from 36.4% in November 2024 to 46.6% in December 2025 and 55.9% in September 2026. Finance and insurance climbed faster, from 51.2% to 58.1% to 73.7% in August 2026, overtaking manufacturing and closing on technology. The Census national rate stayed between 17% and 20% from December 2025 to May 2026 on its new question. Two cautions apply. Ramp expanded its sample to more than 70,000 businesses in June 2026 and revises past months, and the Census Bureau changed its question in November 2025, which roughly doubled the measured rate overnight. Neither break invalidates the ranking, but both make year-on-year growth rates less certain than the headline figures imply.
The pattern is consistent with what AI tools currently do. Text drafting, coding, search and summarization are useful wherever work is digital and done at a desk, which describes most of the information, finance and professional-services workforce. In hospitality, construction and retail, much of the work is physical or customer-facing, and AI arrives first in back-office functions such as marketing, finance and customer service. McKinsey’s 2026 survey found 89% of organizations using AI in at least one function, but that survey covers mostly large firms, and only 37% of respondents reported any EBIT impact. Industry adoption is no longer a question of whether firms use AI at all; it is a question of how many functions, how intensively and with what measurable effect.
View the data
| All businesses | Finance & insurance | |
|---|---|---|
| Nov 2024 | 36.4 | 51.2 |
| Dec 2025 | 46.6 | 58.1 |
| Aug 2026 | 55.2 | 73.7 |
Notes: Only dated readings are plotted; the path between them was not linear (for example, a February 2026 reading shows 49.1% overall and 66.6% for finance). September 2026 overall: 55.9%. Information-sector firms were at 65.9% in November 2024 and 80.0% in August 2026; no December 2025 value was found, so that series is not plotted. Ramp expanded its sample in June 2026. 55.2 – 36.4 = 18.8 points (DERIVED).REPORTAI report-ai.org
Ramp measures payment: a firm counts if it buys an AI subscription through Ramp, and Ramp’s customers are disproportionately technology-forward, so its levels run high (55.9% in September 2026). The Census Bureau’s BTOS measures use: a representative sample of US employer businesses is asked whether AI was used in any of 15 functions, and many small firms say no (19.8% in May 2026). Eurostat measures specific technologies among EU firms with ten or more employees (19.95% in 2025). Each method has a different denominator and a different definition of ‘use’, so their levels cannot be compared. All three, however, rank information and professional or financial services first and construction, hospitality and retail last.
Healthcare moved from pilots to practice: 81% of US physicians use AI and the FDA authorized a record number of AI devices in 2025, but measured gains remain narrow
Physician adoption has risen faster than in almost any other profession. The American Medical Association found 38% of US physicians using AI in 2023, 66% in November 2024 and 81% in a survey of 1,692 physicians fielded from 15 January to 2 February 2026. The average physician now reports 2.3 use cases, up from 1.1 in 2023, and the share using AI to summarize medical research rose from 13% in 2024 to 39% in 2026. These are self-reported uses and mostly administrative or informational: drafting documentation, summarizing literature and answering questions. They do not mean that 81% of diagnoses involve AI. No AMA wave was fielded in calendar 2025, so the series has a gap between the November 2024 and early-2026 readings.
Institutions are moving more slowly than their clinicians. Menlo Ventures estimated in October 2025 that 22% of US healthcare organizations had implemented domain-specific AI tools, which it described as seven times the 2024 level, implying roughly 3% a year earlier. Health systems led at 27%, outpatient providers were at 18% and payers at 14%. Menlo put healthcare AI spending at $1.4 billion in 2025, nearly triple 2024, and its enterprise survey ranked healthcare first among vertical generative-AI buyers at $1.5 billion. At firm level the sector still trails: Ramp’s August 2026 data show 43.4% of health-care businesses paying for AI tools, below the 55.2% average, because many small practices do not buy AI software even when their physicians use general-purpose chatbots.
The regulatory pipeline has grown every year since 2019. The September 2026 snapshot of the FDA’s AI-Enabled Medical Device List attributes 80 authorizations to 2019, 114 to 2020, 130 to 2021, 163 to 2022, 226 to 2023, 235 to 2024 and 335 to 2025. The cumulative list stood at 1,016 entries in December 2024, about 1,430 by the end of 2025 and 1,614 in its 4 September 2026 update, which covers decisions through 29 June 2026. Third-party quarterly counts put the first half of 2026 at roughly 167 to 178 authorizations, broadly in line with the 2025 pace. Counts of 2025 itself differ by method, from 295 for 510(k) clearances alone to 335 on the full list, so the year is rated CONTESTED. About three-quarters of all listed devices are in radiology.
View the data
| FDA AI device authorizations | |
|---|---|
| 2019 | 80 |
| 2020 | 114 |
| 2021 | 130 |
| 2022 | 163 |
| 2023 | 226 |
| 2024 | 235 |
| 2025 | 335 |
Notes: Counts by decision date from the list as it stood in September 2026; the FDA back-fills the list, so earlier snapshots show lower counts (2022 was first reported as 91). 2025 is CONTESTED: other counts are 331, 333 and 295 (510(k) only). Partial 2026 (January-June: about 167-178) is excluded because it is not a full year. 335 / 80 = 4.2 (DERIVED).REPORTAI report-ai.org
Outcome evidence is real but narrower than the adoption figures. The strongest result remains the MASAI randomized trial of more than 100,000 Swedish women, in which AI-supported mammography reading detected 29% more cancers without more false positives and cut radiologists’ reading workload by 44% (site page 863). For ambient AI scribes, the most widely deployed generative tool, a JAMA multisite study of more than 1,800 users and 6,770 controls published in April 2026 found documentation time fell by about 16 minutes a day, roughly 10%, and total electronic-record time by about 13 minutes, or 3%. Only 32% of users were heavy users, and they saw larger gains. Earlier pre/post studies reported burnout falling from 51.9% to 38.8% after 30 days, but they had no control group and low response rates.
The data cannot yet show whether AI improves patient outcomes at scale. The FDA list counts authorizations, not deployments or clinical effects, and the agency itself says the list is not comprehensive. Most authorized devices reach market through the 510(k) pathway, which requires equivalence to an existing device rather than new outcome trials. Physician surveys measure use, not accuracy. Patient attitudes also lag: a Pew Research Center survey found 60% of US adults would be uncomfortable if their provider relied on AI for diagnosis and treatment (2023; site page 863). The next stage of evidence will need trials that link AI deployment to diagnoses, readmissions or mortality, not only to time saved. Nor does any public source track how often clinicians override or ignore AI output.
View the data
| Physicians using AI | |
|---|---|
| 2023 | 38 |
| Nov 2024 | 66 |
| Jan-Feb 2026 | 81 |
Notes: Self-reported use of any AI in practice, including administrative and research tasks. No wave was fielded in calendar 2025, so the gap between bars is uneven.REPORTAI report-ai.org
The FDA’s AI-Enabled Medical Device List records devices authorized for marketing in the US that the agency identifies as using AI, dated by decision. It is updated periodically and back-filled, so the count for any past year tends to rise in later snapshots; 2022, for example, was first reported as 91 and now shows about 163. Third parties count the same list differently: some include only 510(k) clearances, others all pathways. The FDA states that the list is not a comprehensive resource. An authorization means a device may be sold, not that it is in use or that it improves outcomes. We therefore plot one snapshot consistently and label 2025 CONTESTED.
Finance and software lead on adoption and spending, but the best controlled evidence on productivity is mixed
Financial services is the most AI-intensive industry outside technology itself. Ramp’s data show finance and insurance firms paying for AI rising from 51.2% in November 2024 to 58.1% in December 2025 and 73.7% in August 2026. The Census Bureau found 33.9% of finance and insurance businesses using AI as of May 2026, second only to the information sector. Gallup’s May 2026 workplace data show the same order for daily use by employees: 42% in technology, 27% in finance and 22% in professional services, with the other seven industries between 9% and 15%. IDC estimated that banks spent about $31.3 billion on AI in 2024, more than 20% of all AI spending tracked by its industry guide, making banking the largest-spending vertical.
The largest banks report deployment at scale, mostly through their own disclosures. JPMorgan Chase expects about $19.8 billion of technology spending in 2026, up roughly 10%; reports of the AI-specific share differ, at $1.2 billion or about $2 billion, so the allocation is rated CONTESTED. Its president put the annual business value of AI tools at $1 billion to $1.5 billion, and the bank said its internal LLM Suite was used by about 150,000 employees weekly in February 2026. Bank of America’s Erica assistant handled 676 million client interactions in 2024 and nearly 700 million in 2025, passing 3.2 billion cumulative interactions by the end of 2025. Annual Erica use grew only about 3% in 2025, a sign that mature consumer-facing assistants grow with the customer base rather than with AI capability.
Fraud detection is the oldest and best-documented banking use, but most outcome figures come from vendors. Mastercard says its Decision Intelligence Pro model improves fraud detection by about 20% on average and cuts false positives by up to 85%, scoring transactions in under 50 milliseconds; Danske Bank reported roughly 60% fewer false positives after moving to machine-learning models (site page 468). These are company analyses on their own data. The scale of the problem is independently measured: the Nilson Report put global card fraud losses at about $33.41 billion in 2024, slightly lower than in 2023. A falling loss total is consistent with better detection but cannot be attributed to AI alone. Losses are concentrated in the United States, which accounts for about a quarter of card volume but close to 42% of losses.
| Indicator | 2024 | 2025 | 2026 to date | Source |
|---|---|---|---|---|
| Finance & insurance firms paying for AI | 51.2% (Nov) | 58.1% (Dec) | 73.7% (Aug) | Ramp AI Index |
| Finance & insurance firms using AI | n/a | n/a | 33.9% (May) | Census BTOS |
| Employees using AI daily: tech / finance | n/a | n/a | 42% / 27% (May) | Gallup |
| Erica cumulative interactions | 2.5B+ | 3.2B+ | 3.4B+ | Bank of America |
| Developers using AI tools | 62% | 78% | ~83% (DERIVED) | Stack Overflow |
| Developers trusting AI accuracy | 43% | 33% | n/a | Stack Overflow |
| GitHub Copilot paid subscribers | ~2.7M (DERIVED) | 4.7M | not disclosed | Microsoft |
| Time on tasks with AI, METR trial | n/a | +19% (slower) | n/a | METR |
Notes: Gallup daily-use figures: technology 42%, finance 27%, professional services 22%, other industries 9-15% (Gallup, May 2026). Census uses a new question from November 2025. Stack Overflow 2026 = 100% – 17.2% reporting no AI tools (DERIVED; wording changed). Copilot 2024 = 4.7M / 1.75 (DERIVED from Microsoft’s +75% year-on-year). Some outlets report 2025 trust as 29%. Erica 2026 from CMSWire; a 3.6B figure for August 2026 is unverified.REPORTAI report-ai.org
Software development has the highest practitioner adoption of any occupation surveyed. Stack Overflow’s developer survey found 62% of developers using AI tools in 2024 and 78% in 2025; in 2026 only 17.2% said they used no AI tools, implying about 83% use on a reworded question. Daily use rose from 47.1% in 2025 to 58% in Stack Overflow’s March 2026 analysis. GitHub Copilot had 4.7 million paid subscribers at the end of 2025, about 75% more than a year earlier, and Microsoft reported 50 million Copilot users in mid-2026. Menlo Ventures found coding accounted for about 55% of departmental generative-AI spending in 2025 (site page 376). Coding agents are spreading too: an arXiv study found traces of coding-agent use in 22% to 29% of GitHub projects as of February 2026 (site page 862).
Productivity evidence in software is more cautious than adoption suggests. Trust in the accuracy of AI tools fell from 43% of developers in 2024 to 33% in 2025, and in 2026 only 6.6% said they would trust AI output for important decisions. The only randomized trial of experienced developers in their own codebases, published by METR in July 2025, found that 16 open-source developers took 19% longer on 246 tasks when allowed early-2025 AI tools, even though they expected to be faster. The study is small and the tools have since improved, but it is a reminder that self-reported time savings and measured output can differ. No large controlled study of AI’s effect on software output across firms has yet been published.
Many sector outcome figures come from the company selling or running the system: Mastercard’s 20% detection gain, Stripe’s 32% fraud reduction, JPMorgan’s $1-1.5 billion of annual AI value. These figures are not necessarily wrong, but they are measured on the vendor’s own data, against a baseline the vendor chooses, and they are rarely replicated. We label them as vendor-reported and do not add them up across firms. Independent measures, such as total card fraud losses or randomized trials like METR’s, are rarer but carry more weight, even when they are small. Where only vendor figures exist, the honest conclusion is that a gain is plausible, not that it has been measured.
In education and law, individual use is near-universal while institutions, assessment rules and verification lag behind
University students adopted AI faster than any other group tracked in this report. HEPI’s annual survey of UK full-time undergraduates found 66% using AI in the 2023-24 academic year, 92% in 2024-25 and 95% in 2025-26, the last from fieldwork by Savanta in December 2025 with 1,054 respondents. In the latest wave 94% used generative AI for assessed work, and 12% said they included AI-generated text directly in assessed work, up from 3% in 2024 and 8% in 2025. Institutional provision has grown but remains far behind: 38% of students said their university now supplies generative-AI tools, up from 23% and 9% in the two previous years. Students are, in effect, bringing their own tools to institutions that are still writing the rules.
In US schools the picture is similar but measured less consistently. Pew Research Center found that 13% of US teens had used ChatGPT for schoolwork in fall 2023 and 26% in fall 2024; in fall 2025, on a wider question covering any chatbot, 54% had used one for schoolwork, and 10% used chatbots for all or most of it. The jump therefore mixes real growth with a change of definition. Among teachers, Gallup and the Walton Family Foundation found 60% had used AI in the 2024-25 school year and 32% weekly, with weekly users saving about 5.9 hours a week. Their 2025-26 survey found 59% of K-12 educators using AI, but it covered a wider population including paraprofessionals, so the flat reading is not evidence of decline.
View the data
| Use AI in any way | University provides tools | |
|---|---|---|
| 2023-24 | 66 | 9 |
| 2024-25 | 92 | 23 |
| 2025-26 | 95 | 38 |
Notes: Single survey series. Academic years are labelled by the year they describe; each survey was fielded early in that academic year (for example, December 2025 for 2025-26). In this report’s data file, 2025-26 is filed under 2026.REPORTAI report-ai.org
Legal adoption has reached a similar plateau. Clio’s Legal Trends Reports found 19% of legal professionals using AI in 2023, 79% in 2024 and 79% again in 2025; the 2026 report found 85% reporting some level of AI adoption at their firms. Large firms led at 87% in 2025, against 71% of solo practitioners. Spending has followed: Menlo Ventures estimated legal generative-AI spending at $350 million in 2024 and $650 million in 2025, the second-largest vertical after healthcare. Thomson Reuters’ CoCounsel reported more than 1 million users in early 2026 (site page 472). Other survey series report much lower legal adoption, around 30% in 2024, because they measure firm-level tool deployment rather than individual use; the two should not be mixed.
The constraint in both sectors is verification rather than access. A 2024 Stanford RegLab and HAI study found leading legal research tools hallucinated on 17% to 33% of queries, despite vendor claims of hallucination-free citations, and the 2023 Mata v. Avianca sanctions showed the professional cost of filing unverified AI output (site page 472). In education, Turnitin reported in 2024 that about 11% of more than 200 million papers it reviewed contained at least 20% AI-written text, while detection accuracy remains disputed. The outcome evidence for learning is thin but positive in controlled settings: a World Bank trial in Nigeria estimated gains equivalent to 1.5 to 2 years of schooling from a six-week AI tutoring programme (site page 865). One trial does not establish a system-wide effect.
Education surveys are fielded in one calendar year and describe an academic year that spans two. HEPI’s December 2025 fieldwork describes 2025-26; Pew’s September-October 2025 survey describes the 2025-26 school year; Gallup-Walton’s July-August 2026 survey looks back on 2025-26. To keep like with like, we label each by the academic year it describes and file it under the calendar year in which that academic year ends. This matches the labels on Report AI’s education page but differs from the publication year some outlets use. Where a survey changed its question, as Pew did in 2025 by widening ‘ChatGPT’ to ‘chatbots’, we mark a break in the series rather than compute growth.
Outside the knowledge industries adoption is lower and harder to measure, and the best data track customer behaviour rather than firm deployment
Retail is a large AI spender but a modest adopter by firm count. IDC estimated retail AI spending at about $25 billion in 2024, third among industries. Yet Ramp found 48.8% of retail businesses paying for AI in August 2026, below the 55.2% average, and the Census Bureau found about 14% of retail businesses using AI in May 2026, among the lowest rates of any sector. The difference reflects concentration: a handful of large chains account for most retail AI spending, while hundreds of thousands of small stores do not buy AI tools. Retail’s AI story is therefore told mainly through large-retailer disclosures and through measures of shopper behaviour, not through representative adoption surveys. IDC does not publish retail AI spending after 2024.
Shopper behaviour is changing faster than retailer adoption. Adobe Analytics found that traffic from generative-AI sources to US retail websites rose 1,300% in the 2024 holiday season and 693.4% in the 2025 season, with AI-driven revenue per visit more than doubling; by September 2026 the monthly year-on-year increase had slowed to 71% from a much larger base. Salesforce estimated that AI and agents influenced $229 billion of global online holiday sales in 2024, about 19% of orders, and $262 billion in 2025, about 20%. Retailers with their own agents grew sales 6.2% against 3.9% for peers. ‘Influenced’ is Salesforce’s attribution model, not a causal estimate, and the growth rates come from a small base, so these figures show direction rather than size of effect.
| Indicator | 2024 | 2025 | 2026 to date | Source |
|---|---|---|---|---|
| Gen-AI traffic to US retail sites, YoY | +1,300% (holiday) | +693.4% (holiday) | +71% (Sep, not comparable) | Adobe Analytics |
| AI-influenced global holiday sales | $229B (19%) | $262B (~20%) | season in progress | Salesforce |
| Retail firms paying for AI | n/a | n/a | 48.8% (Aug) | Ramp AI Index |
| Retail firms using AI | n/a | n/a | ~14% (May) | Census BTOS |
| Marketing activities using gen AI | 7.0% | 15.1% | 22.4% | The CMO Survey |
| Manufacturing firms paying for AI | n/a | n/a | 61.1% (Aug) | Ramp AI Index |
| Manufacturing operations AI-augmented | n/a | n/a | 34% | Rockwell Automation |
| US federal AI use cases | 2,133 | 3,611 | not yet published | OMB inventory |
Notes: Adobe’s September 2026 figure is a single month and not comparable with holiday-season windows. Salesforce’s ‘influenced’ is an attribution model. The 2025 federal inventory added R&D use cases and nine new reporting agencies, so the 69% rise is partly definitional. Rockwell’s survey (May 2026, about 1,500 respondents in 17 countries) is a self-assessment.REPORTAI report-ai.org
Marketing is where AI use inside firms is best measured over time. The CMO Survey, run by Duke’s Fuqua School of Business with Deloitte and the American Marketing Association, found generative AI used in 7.0% of marketing activities in spring 2024, 15.1% in 2025 and 22.4% in 2026; marketers expect more than half within three years, a forecast rather than an outcome. Menlo Ventures put creator-tool spending at $360 million in 2025. McKinsey’s surveys name technology, media and telecom as the sectors furthest along with AI agents. Manufacturing has no comparable three-year series. Ramp found 61.1% of manufacturers paying for AI in August 2026, and Rockwell Automation’s 2026 survey of about 1,500 manufacturers estimated that 34% of operations are AI-augmented, but neither measures what happens on the factory floor.
Government adoption is measured by inventories, not surveys. US federal agencies reported 2,133 AI use cases across 41 agencies in 2024 and 3,611 across 56 agencies in the 2025 inventory released in April 2026, an increase of about 69%. Much of that rise is a reporting change: research and development uses were added, NASA’s count went from 18 to 425, and nine agencies reported for the first time. High-impact use cases rose from 351 to 445 but fell as a share, from 16.5% to 12.3%, under a new definition. Federal AI contract obligations totalled $5.6 billion across fiscal years 2022 to 2024, 72% of it at the Department of Defense (GAO; site page 374). No comparable series exists for state, local or non-US governments.
Government AI inventories and corporate ‘use case’ tallies count projects an organization chooses to report, not how much work AI does. A single chatbot pilot and a fraud-scoring system that processes millions of transactions each count as one. Counts also move with reporting rules: the 2025 US federal inventory rose 69% partly because research uses were added and nine agencies reported for the first time. Use-case counts are useful for tracking breadth and for spotting high-impact systems that need oversight, but they are not a measure of adoption depth, spending or outcome, and they should not be compared across organizations that define a use case differently.
Spending is concentrating in banking, software and healthcare, but sector-level returns remain mostly unmeasured
Total AI spending is measured well; spending by industry is not. Gartner put worldwide AI spending at $987.9 billion in 2024 in its September 2025 edition and now puts 2025 at about $1.79 trillion, restated from the $1.48 trillion it published a year earlier. Its September 2026 forecast for 2026 is about $2.67 trillion, up 49.5%, with about 56% going to infrastructure. These totals include data-centre hardware, AI-enabled devices and services, and they are not split by buying industry. They are also forecasts for a year in progress, so they should not be compared with the measured adoption figures elsewhere in this report. Gartner’s January and May 2026 editions put 2026 at $2.52 trillion and $2.59 trillion; each revision has been upwards, driven mostly by infrastructure.
IDC’s industry guide is the only public source that ranks industries by total AI spending, and its public figures stop at 2024. Banking led at about $31.3 billion, retail followed at about $25 billion, and software and information services completed the top three, which together accounted for about $89.6 billion, roughly 38% of global AI spending tracked by IDC. An earlier IDC edition put 2023 banking spending at $20.6 billion, but editions differ in scope, so the two years are not compared here. Manufacturing and healthcare are not separately disclosed in public IDC materials. The industry ranking by total spend is therefore a 2024 snapshot, not a trend. Until IDC or another tracker publishes 2025 and 2026 industry totals, any claim that one sector’s AI spending is growing faster than another’s rests on survey data or company disclosures.
View the data
| 2024 | 2025 | |
|---|---|---|
| Healthcare | 500 | 1,500 |
| Legal | 350 | 650 |
Notes: Survey-based estimates (2025: n=495 US enterprise decision-makers, 7-25 November 2025). Healthcare 1,500 / 500 = 3.0x; legal 650 / 350 = 1.9x (DERIVED). Menlo’s separate healthcare report gives $1.4B for 2025. 2024 figures come from secondary coverage of the 2024 edition. Not comparable with IDC’s total AI spending, which covers all AI, not generative-AI software only.REPORTAI report-ai.org
For generative-AI software bought for a specific industry, Menlo Ventures’ annual enterprise survey gives a two-year view. Healthcare spending rose from about $500 million in 2024 to $1.5 billion in 2025, 43% of a $3.5 billion vertical-AI total, and legal from $350 million to $650 million. Menlo’s separate healthcare report put 2025 healthcare AI spending at $1.4 billion. Across departments rather than industries, coding took about 55% of departmental generative-AI spending in 2025. These are survey-based estimates from about 500 US enterprise buyers, and Menlo restated its earlier years to exclude inference, so growth multiples should be read as approximate. The vertical total itself, $3.5 billion, is small next to the roughly $37 billion Menlo estimates enterprises spent on generative AI overall in 2025, most of it on horizontal tools and model access.
Measured returns lag both spending and adoption. McKinsey’s 2026 survey found 89% of organizations using AI in at least one function but only 37% reporting any EBIT impact, and about 6% qualifying as high performers. At sector level, the controlled evidence comes from a short list of studies: the MASAI mammography trial (29% more cancers detected), the JAMA scribe study (about 16 minutes a day of documentation saved), the METR developer trial (19% slower) and the Stanford legal-tool test (17-33% hallucination rates). Each covers one task in one setting. None measures industry-wide productivity, and the sectors spending most, banking and software, publish the fewest independent outcome measures. Until more sectors publish controlled results, adoption and spending figures should be read as measures of activity, not of value.
Gartner, IDC and Menlo Ventures measure different things. Gartner’s worldwide total ($2.67 trillion forecast for 2026) includes AI-optimized servers, data centres, AI-enabled phones and PCs, and services, and is not split by buying industry. IDC tracks vendor revenue for AI software, hardware and services by the industry that buys them, which is why banking appears as a $31.3 billion line. Menlo surveys enterprise buyers about generative-AI software only, so its healthcare figure is $1.5 billion. Each is internally consistent; none can be added to or ranked against another. Gartner also revises its past years, as with 2025, restated from $1.48 trillion to about $1.79 trillion.
Between 2024 and 2026 the sector ranking barely moved; what changed was the level of practitioner use and the volume of regulated AI products
Comparing 2024, 2025 and 2026 shows three patterns. First, the order of sectors is stable. Information and technology firms led on Ramp in November 2024 (65.9%) and still led in 2026 (about 80%); finance was second throughout; Eurostat’s top two sectors were the same in 2024 and 2025. Second, practitioner use in regulated professions converged at high levels: physicians went from 66% to 81%, UK undergraduates from 66% to 95% over three academic years, and legal professionals from 79% to 85% on Clio’s measure. Third, the regulated product pipeline grew, with FDA AI device authorizations rising from 235 in 2024 to 335 in 2025 and continuing at a similar pace in the first half of 2026.
Across sectors in 2026 the clearest contrast is between individual and organizational adoption. In healthcare, 81% of physicians use AI but only 43.4% of health-care businesses on Ramp pay for it and 22% of healthcare organizations had implemented domain-specific tools by 2025. In education, 95% of UK students use AI but 38% get tools from their university. In finance and software the gap is smaller, because firms buy tools centrally. Retail, construction and hospitality show low figures on both measures. Outcome evidence follows a different order: healthcare has the most controlled studies, while banking and software, the largest spenders, rely mainly on vendor or company figures.
Several apparent year-on-year jumps are partly definitional and should not be quoted as growth. The Census Bureau’s measured rate roughly doubled when it changed its question in November 2025; Pew’s teen figure doubled when the question widened from ChatGPT to any chatbot; the federal use-case count rose 69% when research uses were added; and Gartner restated 2025 spending upwards by about $300 billion. Where a series broke, this report shows both readings and does not compute a growth rate.
| Sector | Firms paying for AI (Ramp, Aug 2026) | Firms using AI (Census, May 2026) | Practitioner use (latest) | Spend signal | Measured outcome |
|---|---|---|---|---|---|
| Software & IT | 80.9% | 39.7% | ~83% of developers (2026, DERIVED) | Coding ~55% of departmental gen-AI spend (2025) | Experienced devs 19% slower (METR RCT, 2025) |
| Finance & insurance | 73.7% | 33.9% | 27% of employees use AI daily (May 2026) | Banking ~$31.3B AI spend (IDC, 2024) | Vendor-reported only (e.g. +20% fraud detection) |
| Manufacturing | 61.1% | n/a | 34% of operations AI-augmented (2026) | Not disclosed | No controlled evidence found |
| Retail | 48.8% | ~14% | n/a | ~$25B AI spend (IDC, 2024) | AI-influenced holiday sales $262B (attribution, 2025) |
| Healthcare | 43.4% | n/a | 81% of physicians (Jan-Feb 2026) | $1.5B vertical gen AI (Menlo, 2025) | Scribes save ~16 min/day (JAMA, 2026); +29% cancers detected (MASAI) |
| Education | n/a | n/a | 95% of UK undergraduates (2025-26) | n/a | Weekly teacher users save 5.9 h/week (self-reported, 2025) |
| Legal | n/a | n/a | 85% at firms with AI (Clio, 2026) | $650M vertical gen AI (Menlo, 2025) | 17-33% hallucination in research tools (Stanford, 2024) |
| Marketing & media | n/a | n/a | 22.4% of marketing activities (2026) | $360M creator tools (Menlo, 2025) | n/a |
| US federal government | n/a | n/a | 3,611 use cases (2025 inventory) | $5.6B AI contracts FY2022-24 (GAO) | n/a |
Notes: Read down the columns, not across: each column uses one source and its own definition. Ramp’s ‘technology’ sector is compared with Census’s ‘information’ sector. ‘Practitioner use’ rows come from different surveys with different populations and are not ranked. Outcome entries are single studies or company figures, not sector-wide estimates. Sources for figures not in the data file: Gallup daily use (May 2026); Rockwell (May 2026); Menlo coding share and creator tools (site page 376); MASAI (site page 863); Stanford (site page 472); GAO (site page 374).REPORTAI report-ai.org
View the data
| Event | Group | Start | End |
|---|---|---|---|
| Pew teens fieldwork | Education | 2025-09 | 2025-10 |
| Holiday season 2025 | Retail | 2025-11 | 2025-12 |
| BTOS question change | Cross-industry | 2025-11 | |
| HEPI fieldwork | Education | 2025-12 | |
| AMA fieldwork | Healthcare | 2026-01 | 2026-02 |
| FDA decisions counted | Healthcare | 2026-01 | 2026-06 |
| Federal inventory out | Government | 2026-04 | |
| Census BTOS reading | Cross-industry | 2026-05 | |
| Gallup-Walton survey | Education | 2026-07 | 2026-08 |
| Ramp latest reading | Cross-industry | 2026-09 | |
| FDA list update | Healthcare | 2026-09 | |
| Stack Overflow results | Software | 2026-10 | |
| Eurostat 2026 due | Cross-industry | 2026-12 |
Notes: Pew fielded 25 Sep-9 Oct 2025; holiday windows are 1 Nov-31 Dec 2025; BTOS new question from the 17 Nov 2025 collection; AMA fielded 15 Jan-2 Feb 2026; the 4 Sep 2026 FDA list update covers decisions through 29 Jun 2026; Census reading as of 3 May 2026; Gallup-Walton fielded 28 Jul-14 Aug 2026; Stack Overflow results published 6 Oct 2026. Eurostat’s December 2026 release is a scheduled date, not a reading.REPORTAI report-ai.org
Year over year, every comparable measure
| Measure | 2024 | 2025 | 2026 (to date) | 2025→26 |
|---|---|---|---|---|
| US businesses paying for AI tools (Ramp AI Index) | 36.4% | 46.6% | 55.9% | +20% |
| Finance and insurance businesses paying for AI tools (Ramp) | 51.2% | 58.1% | 73.7% | +27% |
| Organizations using AI in at least one business function (McKinsey) | 78% | 88% | 89% | +1% |
| FDA AI-enabled medical device authorizations per year | 235 | 335 | ~167-178 (Jan-Jun) | |
| Cumulative FDA-authorized AI-enabled medical devices | 1,016 | 1,430 | 1,614 | +13% |
| UK full-time undergraduates using AI (HEPI) | 66% | 92% | 95% | +3% |
| UK undergraduates whose university provides generative-AI tools (HEPI) | 9% | 23% | 38% | +65% |
| Legal professionals using AI (Clio Legal Trends) | 79% | 79% | 85% | +8% |
| Developers currently using AI tools in their development process (Stack Overflow) | 62% | 78% | ~83% | +6% |
| Bank of America Erica cumulative client interactions | 2.5B+ | 3.2B+ | 3.4B+ | +6% |
| Share of marketing activities using generative AI (The CMO Survey) | 7.0% | 15.1% | 22.4% | +48% |
- Paid business adoption on Ramp rose from 36.4% (Nov 2024) to 46.6% (Dec 2025) to 55.9% (Sep 2026); finance and insurance moved from 51.2% to 58.1% to 73.7% (Aug 2026), the steepest climb among large sectors.
- FDA AI device authorizations went from 235 (2024) to a record 335 (2025, contested range 295-335) to roughly 167-178 in January-June 2026; the cumulative list grew from 1,016 to 1,614 entries between December 2024 and September 2026.
- Practitioner adoption converged at a high level: physicians from 66% (Nov 2024) to 81% (early 2026), UK undergraduates from 66% to 92% to 95% across three academic years, legal professionals from 79% to 79% to 85% (Clio).
- The measures themselves changed: Census BTOS switched questions in November 2025, Pew widened its question from ChatGPT to all chatbots, Ramp expanded its sample in June 2026, and Gartner restated 2025 spending from $1.48T to about $1.79T, so several apparent jumps are partly definitional.
All eleven themes side by side: State of AI, year over year.
The numbers, 2024 to 2026
Every measure in this report with its value for each year, as-of date, source and confidence. Change compares 2025 with 2026 to date where both are numeric.
Cross-industry adoption
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| US businesses paying for AI tools (Ramp AI Index) % of US businesses on Ramp | 36.4% Nov 2024, Ramp AI Index (FRED series RAMPAIALL) MEDIUM | 46.6% Dec 2025, Ramp AI Index, January 2026 update MEDIUM | 55.9% Sep 2026 (to date), Ramp AI Index MEDIUM | +20% |
| US businesses using AI (Census Business Trends and Outlook Survey) % of US employer businesses | ~5% mid-2024, Federal Reserve Bank of Minneapolis analysis of Census BTOS DERIVED | 17.3% Nov 2025, US Census Bureau BTOS (new AI question) MEDIUM | 19.8% 3 May 2026 (to date), US Census Bureau, ‘Large Firms With at Least 20 Employees Biggest AI Users’, 26 May 2026 MEDIUM | +14% |
| EU enterprises (10+ employees) using AI (Eurostat) % of enterprises | 13.5% 2024, Eurostat (reported by ANSA, 23 Jan 2025) MEDIUM | 19.95% 2025 (published 11 Dec 2025), Eurostat, Use of artificial intelligence in enterprises (dataset isoc_eb_ain2) MEDIUM | — | |
| Organizations using AI in at least one business function (McKinsey) % of survey respondents’ organizations | 78% 2024 survey, McKinsey, The State of AI MEDIUM | 88% 2025 survey (published Nov 2025), McKinsey, The State of AI in 2025 MEDIUM | 89% 2026 survey (1,719 respondents), McKinsey, The State of AI in 2026: On the road to ROI (secondary coverage) MEDIUM | +1% |
Financial services
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| Finance and insurance businesses paying for AI tools (Ramp) % of finance and insurance businesses on Ramp | 51.2% Nov 2024, Ramp AI Index (FRED series RAMPAIFINANCE) MEDIUM | 58.1% Dec 2025, Ramp AI Index (FRED series RAMPAIFINANCE) MEDIUM | 73.7% Aug 2026 (to date), Ramp AI Index (MacroMicro chart; FRED shows about 73.7% for Sep 2026) MEDIUM | +27% |
| Bank of America Erica cumulative client interactions interactions (cumulative since 2018) | 2.5B+ end-2024 (released 24 Feb 2025), Bank of America newsroom MEDIUM | 3.2B+ end-2025 (released 10 Mar 2026), Bank of America newsroom MEDIUM | 3.4B+ 2026 (to date), CMSWire, quoting Bank of America MEDIUM | +6% |
| JPMorgan Chase: technology budget and AI value US$ | — | ~$2B a year on AI 2025 shareholder letter, JPMorgan Chase (via site page 376) MEDIUM | $19.8B tech budget; $1-1.5B annual AI value 2026 guidance; Apr 2026, JPMorgan Chase management (Fortune and others) CONTESTED | +890% |
Healthcare
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| FDA AI-enabled medical device authorizations per year devices (by decision year) | 235 decision year 2024 (FDA list, Sep 2026 snapshot), FDA AI-Enabled Medical Device List, as tabulated by The Medical Futurist and aiadoption.org MEDIUM | 335 decision year 2025 (FDA list, Sep 2026 snapshot), FDA AI-Enabled Medical Device List, as tabulated by The Medical Futurist and aiadoption.org CONTESTED | ~167-178 (Jan-Jun) Jan-Jun 2026 (to date), Innolitics quarterly counts; other third-party analyses CONTESTED | |
| Cumulative FDA-authorized AI-enabled medical devices devices (cumulative, since 1995) | 1,016 20 Dec 2024 list update, US FDA, FDA Roundup, 20 Dec 2024 MEDIUM | 1,430 through Dec 2025, JMIR 2026 cross-sectional analysis of the FDA list MEDIUM | 1,614 4 Sep 2026 list update (decisions through 29 Jun 2026), FDA AI-Enabled Medical Device List, as reported by Complizen and OneDayAdvisor MEDIUM | +13% |
| US physicians using AI in their practice (AMA) % of physicians surveyed | 66% Nov 2024 fieldwork (published Feb 2025), American Medical Association, Augmented Intelligence Research (n=1,183) MEDIUM | — | 81% fielded 15 Jan-2 Feb 2026 (n=1,692), American Medical Association, 2026 Physician Survey on Augmented Intelligence MEDIUM | |
| US healthcare organizations that have implemented domain-specific AI tools (Menlo) % of healthcare organizations | ~3% 2024, Menlo Ventures, 2025: The State of AI in Healthcare DERIVED | 22% Oct 2025, Menlo Ventures, 2025: The State of AI in Healthcare MEDIUM | — |
Spending
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| Healthcare vertical generative-AI spend (Menlo enterprise survey) US$ | $500M 2024 (published 20 Nov 2024), Menlo Ventures, 2024: The State of Generative AI in the Enterprise MEDIUM | $1.5B 2025 (survey 7-25 Nov 2025, n=495), Menlo Ventures, 2025: The State of Generative AI in the Enterprise MEDIUM | — | |
| Banking AI spending, worldwide (IDC) US$ | ~$31.3B 2024, IDC Worldwide AI and Generative AI Spending Guide MEDIUM | — | — | |
| Worldwide AI spending, all categories (Gartner) US$ | $987.9B Gartner, 17 Sep 2025 edition, Gartner MEDIUM | ~$1.79T base year in Gartner’s 16 Sep 2026 edition, Gartner MEDIUM | ~$2.67T (forecast) Gartner, 16 Sep 2026 edition, Gartner MEDIUM | +49% |
Legal
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| Legal vertical generative-AI spend (Menlo enterprise survey) US$ | $350M 2024, Menlo Ventures, 2024: The State of Generative AI in the Enterprise MEDIUM | $650M 2025, Menlo Ventures, 2025: The State of Generative AI in the Enterprise MEDIUM | — | |
| Legal professionals using AI (Clio Legal Trends) % of legal professionals surveyed | 79% Clio Legal Trends Report 2024, Clio MEDIUM | 79% Clio Legal Trends Report 2025, Clio (via 2Civility summary) MEDIUM | 85% Clio Legal Trends Report 2026 (publication date not confirmed), Clio press release MEDIUM | +8% |
Education
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| UK full-time undergraduates using AI (HEPI) % of students surveyed | 66% HEPI 2024 survey (academic year 2023-24), HEPI / Kortext, Student Generative AI Survey 2024 MEDIUM | 92% fieldwork Dec 2024, published Feb 2025 (n=1,041; academic year 2024-25), HEPI / Kortext, Student Generative AI Survey 2025 MEDIUM | 95% fieldwork Dec 2025 by Savanta, published 12 Mar 2026 (n=1,054; academic year 2025-26), HEPI / Kortext, Student Generative AI Survey 2026 (Report 199) MEDIUM | +3% |
| UK undergraduates whose university provides generative-AI tools (HEPI) % of students surveyed | 9% HEPI 2024 survey (academic year 2023-24), HEPI / Kortext, Student Generative AI Survey 2026 (trend data) MEDIUM | 23% HEPI 2025 survey (academic year 2024-25), HEPI / Kortext, Student Generative AI Survey 2026 (trend data) MEDIUM | 38% fieldwork Dec 2025, published 12 Mar 2026 (academic year 2025-26), HEPI / Kortext, Student Generative AI Survey 2026 MEDIUM | +65% |
| US teens (13-17) who have used AI chatbots for schoolwork (Pew) % of teens | 13% fall 2023 survey (school year 2023-24), Pew Research Center MEDIUM | 26% fielded 18 Sep-10 Oct 2024 (n=1,391; school year 2024-25), Pew Research Center, 15 Jan 2025 MEDIUM | 54% fielded 25 Sep-9 Oct 2025 (n=1,458; school year 2025-26), Pew Research Center, How Teens Use and View AI, 24 Feb 2026 MEDIUM | +108% |
| US K-12 teachers/educators who used AI during the school year (Gallup-Walton) % of respondents | — | 60% school year 2024-25 (published Jun 2025; n=2,232 public-school teachers), Gallup / Walton Family Foundation, Teaching for Tomorrow MEDIUM | 59% school year 2025-26 (surveyed 28 Jul-14 Aug 2026; n=1,952), Gallup / Walton Family Foundation (released 14 Oct 2026) MEDIUM | -2% |
Software & IT
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| Developers currently using AI tools in their development process (Stack Overflow) % of survey respondents | 62% Stack Overflow Developer Survey 2024, Stack Overflow MEDIUM | 78% Stack Overflow Developer Survey 2025, Stack Overflow MEDIUM | ~83% Stack Overflow Developer Survey 2026 (results 6 Oct 2026), Stack Overflow blog DERIVED | +6% |
| GitHub Copilot paid subscribers subscribers | ~2.7M Dec 2024, Microsoft FY26 Q2 earnings (Jan 2026) DERIVED | 4.7M quarter to 31 Dec 2025 (reported Jan 2026), Microsoft FY26 Q2 earnings MEDIUM | — |
Outcomes
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| Change in task completion time for experienced open-source developers using AI (METR randomized trial) % change in time taken | — | +19% (slower) early-2025 tools; published 10 Jul 2025, METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity MEDIUM | — | |
| Documentation time saved by ambient AI scribes (controlled multisite study) minutes per clinician per day | — | — | -16 min/day (-10%) published Apr 2026, JAMA multisite study of 1,800+ scribe users vs 6,770 controls (via News-Medical) MEDIUM |
Retail
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| Global online holiday sales influenced by AI and agents (Salesforce) US$ (1 Nov-31 Dec) | $229B (19%) holiday season 2024, Salesforce MEDIUM | $262B (~20%) holiday season 2025 (released 6 Jan 2026), Salesforce MEDIUM | — | |
| Growth in traffic from generative-AI sources to US retail websites (Adobe) % change year on year | +1,300% 1 Nov-31 Dec 2024 vs 2023, Adobe Analytics MEDIUM | +693.4% holiday season 2025 (released Jan 2026), Adobe Analytics MEDIUM | +71% Sep 2026 (single month, YoY), Adobe Analytics MEDIUM | -90% |
Media & marketing
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| Share of marketing activities using generative AI (The CMO Survey) % of marketing activities | 7.0% Spring 2024, The CMO Survey (Duke Fuqua, Deloitte, AMA) MEDIUM | 15.1% Spring 2025, The CMO Survey MEDIUM | 22.4% Spring 2026 (35th edition), The CMO Survey Highlights and Insights Report 2026 MEDIUM | +48% |
Government
| Measure | 2024 | 2025 | 2026 (to date) | Change |
|---|---|---|---|---|
| US federal agency AI use cases (OMB inventory) use cases reported | 2,133 2024 inventory (41 agencies), OMB Federal Agency AI Use Case Inventory (via CDT, ExecutiveGov) MEDIUM | 3,611 2025 inventory (56 agencies; released Apr 2026), OMB Federal Agency AI Use Case Inventory (via Nextgov, Brookings) MEDIUM | — |
Numbers to stop quoting
| Widely repeated | What the evidence supports |
|---|---|
| “The FDA has authorized about 1,250 AI medical devices (1,247).” | 1,247 was the July 2025 list update. The list reached 1,614 entries in its 4 September 2026 update (decisions through 29 June 2026). · detail |
| “Gartner forecasts $2.59 trillion of AI spending in 2026, up 47% from about $1.5 trillion in 2025.” | Gartner’s 16 September 2026 forecast is about $2.67 trillion for 2026 (+49.5%), and it restated 2025 to about $1.79 trillion from the $1.48 trillion it published in September 2025. The figure is a forecast, not an outcome. · detail |
| “Generative-AI traffic to retail sites is up 1,300%.” | +1,300% was the 2024 holiday season (Adobe). The 2025 holiday season was +693.4%, and the latest monthly reading, September 2026, was +71% year on year from a much larger base. |
| “Teen use of ChatGPT for schoolwork doubled from 26% to 54%.” | Pew’s 26% (fall 2024) asked about ChatGPT only; the 54% (fall 2025) asks about any chatbot. The question changed, so the rise is not a like-for-like doubling. · detail |
| “Only about 5-10% of US businesses use AI, according to the Census Bureau.” | That was the old BTOS question (AI used to produce goods or services). From November 2025 the Census asks about any business function, and the rate was 19.8% as of 3 May 2026; the two series are not comparable. |
| “Only about 30% of lawyers use AI.” | That figure (about 11% in 2023 to 30% in 2024) comes from a different survey series than Clio’s, which found 79% of legal professionals using AI in 2024 and 85% reporting firm-level adoption in 2026. Cite the survey and its definition with the number. · detail |
In 2027 the test for industry AI shifts from adoption to measured outcomes, and the earliest signals will come from healthcare, finance and representative business surveys
Adoption by practitioners is close to its ceiling in medicine, higher education, law and software, so further growth in those headline figures will tell us little. The more informative indicators for 2027 are representative business adoption, which at about one firm in five in the Census survey still has room to rise, the gap between the most and least digital sectors, and the volume and type of regulated AI products. A narrowing of the sector gap would mean AI tools have found uses in physical and customer-facing work; a stable gap would mean adoption is saturating within knowledge work.
The bigger open question is outcomes. By the end of 2027 there should be more controlled studies of ambient scribes, coding agents and AI tutoring, and the second year of McKinsey-style EBIT questions under higher adoption. If controlled studies keep finding single-digit or low double-digit time savings, and the share of organizations reporting EBIT impact stays below 40%, the case that AI is transforming sector productivity will remain unproven, whatever the adoption rates. If replicated trials show larger effects on output or patient outcomes, that view would change.
What to watch in 2027
| Indicator | Where it stands | What would change the picture |
|---|---|---|
| Census BTOS national AI use rate (new question) | 19.8% as of 3 May 2026 | A sustained reading above 25% would mean adoption is spreading beyond knowledge sectors; a reading still below 22% by mid-2027 would mean a plateau. |
| Ramp gap between technology and accommodation and food services | 80.9% vs 33.1% (about 2.4 to 1), August 2026 | A ratio below 2 to 1 would mean AI tools are finding uses in physical and customer-facing work. |
| FDA AI device authorizations per year | 335 in 2025 (CONTESTED, 295-335); about 167-178 in January-June 2026 | More than 400 in 2026 would mean acceleration; fewer than 300 would mean the pipeline has levelled off. |
| Controlled outcome studies of ambient AI scribes | About 16 minutes a day (about 10%) of documentation time saved (JAMA, April 2026) | Replicated trials showing 20% or more time saved, or effects on patient outcomes, would change the assessment of clinical AI. |
| Organizations reporting any EBIT impact from AI (McKinsey) | 37% in the 2026 survey | Above 50% would mean adoption is turning into measurable financial returns; flat at about 37-39% would mean it is not. |
| US federal AI use cases, high-impact share | 3,611 use cases, 12.3% high-impact (2025 inventory) | A 2026 inventory with a rising high-impact share would mean government use is moving from pilots into decisions that affect people. |
- Ramp AI Index
- The monthly share of US businesses on Ramp’s corporate card and bill-pay platform that pay for at least one AI subscription or contract, reported overall and by industry. It measures paid adoption among Ramp’s customers, not use across the whole economy.
- Business Trends and Outlook Survey (BTOS)
- A US Census Bureau survey of employer businesses fielded every two weeks. Since the 17 November 2025 collection it asks whether a business used AI in any of 15 business functions; before that it asked about AI used to produce goods or services.
- FDA AI-Enabled Medical Device List
- The FDA’s public list of medical devices authorized for US marketing that it identifies as using AI, dated by decision. It is updated periodically, back-filled and, by the FDA’s own description, not comprehensive.
- 510(k) clearance
- The FDA pathway for devices shown to be substantially equivalent to a device already on the market. Most AI devices are cleared this way, which does not require new clinical outcome trials.
- Domain-specific AI tool
- In Menlo Ventures’ healthcare survey, AI software built for a healthcare task (such as ambient documentation or coding) rather than a general-purpose chatbot.
- Vertical AI
- AI software sold for one industry, such as healthcare or legal. Menlo Ventures estimates vertical generative-AI spending separately from horizontal tools and coding.
- Ambient AI scribe
- Software that listens to a clinical consultation and drafts the visit note for the clinician to review and sign.
- AI use case inventory
- The annual list of AI applications that US federal agencies report to the Office of Management and Budget. It counts reported projects, not usage volume or spending.
- AI-influenced sales
- Salesforce’s estimate of online orders in which shoppers interacted with AI or agents during the purchase journey. It is an attribution model, not a measure of sales caused by AI.
Frequently asked
Which industry uses AI the most in 2026?
Information and technology businesses lead on every comparable survey. Ramp’s AI Index showed 80.9% of technology firms on its platform paying for AI in August 2026, against 55.2% for all businesses, and the US Census Bureau found 39.7% of information-sector businesses using AI as of 3 May 2026, against a national rate of 19.8%. Finance and insurance is second on both (73.7% on Ramp; 33.9% on Census).
How many AI medical devices has the FDA authorized?
The FDA’s AI-Enabled Medical Device List reached 1,614 entries in its 4 September 2026 update, covering decisions through 29 June 2026, up from 1,016 in December 2024. The September 2026 snapshot attributes a record 335 authorizations to 2025 and 235 to 2024; other counts of 2025 range from 295 to 333. Roughly three-quarters of all listed devices are in radiology.
What share of doctors use AI?
The American Medical Association found that 81% of US physicians used AI in their practice in a survey of 1,692 physicians fielded from 15 January to 2 February 2026, up from 66% in November 2024 and 38% in 2023. The average physician reported 2.3 use cases, most commonly summarizing research and drafting documentation.
Which industry spends the most on AI?
IDC ranked banking first, at about $31.3 billion of AI spending in 2024, with retail at about $25 billion and software and information services in the top three; together the three reached about $89.6 billion. For vertical generative-AI software, Menlo Ventures found healthcare largest at $1.5 billion in 2025, 43% of the $3.5 billion vertical total, ahead of legal at $650 million. No public source updates IDC’s industry split for 2025 or 2026.
Is AI improving results in these industries?
Controlled evidence is limited and mixed. A JAMA multisite study published in April 2026 found ambient AI scribes cut clinicians’ documentation time by about 16 minutes a day (about 10%); METR’s July 2025 randomized trial found experienced open-source developers took 19% longer with early-2025 AI tools; and McKinsey’s 2026 survey found only 37% of organizations reporting any EBIT impact from AI. Most sector-level gains published so far are vendor-reported.
Figures are compiled from company disclosures, regulators, research firms and reputable reporting, and cross-checked against Report AI’s own index pages. Each value carries its as-of date, source and a confidence rating: HIGH where the primary source was read, MEDIUM where it rests on consistent reputable secondary reporting (most figures, because many primary pages cannot be opened from our research environment), DERIVED where it is our own arithmetic, and CONTESTED where credible sources disagree. A year means the calendar year the figure describes; 2026 values are the latest available and marked “to date”. Weekly and monthly user counts, and forecasts and outcomes, are never compared as if they were the same measure. Exhibits plot only dated, sourced readings; lines are not interpolated between them, and values with different definitions are never placed on one axis. Known gaps in this edition: Manufacturing has no three-year series: only Ramp’s 61.1% of manufacturing firms paying for AI (Aug 2026) and Rockwell’s estimate that 34% of operations are AI-augmented (May 2026). Eurostat’s 2025 manufacturing share could not be verified (17.3% circulates but conflicts with Eurostat’s statement that all sectors outside the top two were below 16% in 2024).; IDC’s industry split is paywalled: no public 2025 or 2026 banking, retail or healthcare spending figure; Gartner does not split by industry.; Government: the 2026 federal AI use-case inventory is not yet published; no comparable state/local or non-US government adoption series was found. The 2025 federal count (3,611) is inflated by a reporting change.; Census BTOS sector rates were found only for information (39.7%), finance and insurance (33.9%) and retail (about 14%) as of May 2026; the 24.6% national reading for October 2026 is unverified.; Ramp’s information-sector value for December 2025 was not found; Ramp expanded its sample in June 2026 and revises history, so earlier readings may move.; No AMA physician survey was fielded in 2025; no 2026 Menlo healthcare or enterprise vertical figures yet (expected December 2026); Eurostat 2026 data due December 2026.; FDA counts by year differ by snapshot and method (2025: 295-335; Q1 2026: 81 vs 92); FDA’s own page could not be read, so all FDA figures are from third-party tabulations of the list.; Media and marketing has only the CMO Survey series; no newsroom or publisher AI adoption series was found.; Primary sites for Eurostat, the Census Bureau and FRED were unreachable from this environment; their figures are rated MEDIUM from consistent secondary reporting.; Clio’s 2026 Legal Trends Report publication date is unconfirmed, and its 2026 wording (‘some level of AI adoption at their firms’) may differ from earlier years’ individual-use question.; Site consistency: pages 863 and 938 still show 1,247 FDA devices; page 376 shows Gartner’s superseded $2.59T and ‘2025 ~$1.5T’; page 865 labels Pew’s 13% as 2023 (survey year) while this report files school surveys by academic year.. These reports compile and verify secondary data with original analysis; they are not original surveys. See our editorial standards, corrections log and fact-check register.
State of AI Usage · State of AI Spending · State of AI Finance · State of AI Work · State of AI Models & Prices · State of AI Tokens · State of AI Infrastructure & Energy · State of AI Geography · State of AI Industries · State of AI Risk · State of AI Regulation · Year over year · AI Market Pulse · AI statistics that are wrong