147 Generative AI Statistics October 2026

Generative AI statistics for 2026 show use running far ahead of payoff. At least 700 million people use leading AI systems every week (International AI Safety Report, Feb 2026), yet CEOs surveyed by IBM say only 25% of their AI initiatives delivered the expected return and only 16% have scaled across the enterprise (IBM Institute for Business Value, May 2025).

The page covers adoption, spending, funding, models and computing power, productivity and jobs, hallucination and misuse, and regulation. Each one links to the organisation that produced it: central banks, statistical offices, regulators, courts, company filings, academic papers and survey firms. Where a vendor or consultancy ran the survey, it is named in the citation so you can weigh it. Wider AI figures sit on our AI statistics page, chatbot traffic and AI answers on AI search statistics, and marketing use on AI marketing statistics.

How many companies and people have adopted generative AI?

Adoption is no longer the open question. Most employers have tried the tools, and the argument has moved to what they earn.

How many businesses use generative AI?

More than half of service firms and manufacturers in the New York region now use AI: 61% of service firms said so in August 2026, up from 40% a year earlier and 25% in 2024 (Federal Reserve Bank of New York, Sep 2026).

• 51% of regional manufacturers used AI, roughly double the 26% of 2025 and triple the 16% of 2024 (Federal Reserve Bank of New York, Sep 2026).
• Among adopters, the median share of workers using AI was just 17% at service firms and 7% at manufacturers (Federal Reserve Bank of New York, Sep 2026).
• Three-quarters of service firms and more than 90% of manufacturers described their AI investment as minimal to modest (Federal Reserve Bank of New York, Sep 2026).
• 46% of US small employer firms said their business or employees used AI tools in the 2025 Small Business Credit Survey, and another 15% planned to start (Federal Reserve Banks, Oct 2026).
• Nearly 60% of firms in an Atlanta Fed survey of more than 700 corporate executives invested in AI in 2025, and more than 80% expected to invest in 2026 (Federal Reserve Bank of Atlanta, Mar 2026).
• Deloitte found the share of workers with sanctioned AI tools grew from fewer than 40% to around 60% in one year (Deloitte AI Institute, Jan 2026).

Broad reach with shallow use is the pattern. Adoption rates count a firm once whether one person or the whole staff uses the tool.

Is generative AI paying off for businesses?

It depends on who is asked and how return is defined. 74% of enterprise respondents in Wharton's survey reported positive ROI from generative AI, and 72% formally measure it (Wharton Human-AI Research, Oct 2025).

• Nearly 6 in 10 leaders in KPMG's quarterly pulse report measurable business value, and productivity gains are the most common at 55% (KPMG, Sep 2026).
• 74% of organizations now include cost reviews in AI approval processes, up from 61% a quarter earlier, and 43% have usage or token budgets (KPMG, Sep 2026).
• Only 30% of organizations in Deloitte's survey are redesigning key processes around AI, and 37% use it only at a surface level (Deloitte AI Institute, Jan 2026).
• 77% of small firms reported no change in labor costs from using AI (Federal Reserve Banks, Oct 2026).

These surveys ask different people and define return differently, so the figures above should not be compared with each other. Read each as that survey's answer to its own question.

Are AI agents ready for business use?

Not yet at scale. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 because of cost, unclear business value or inadequate risk controls (Gartner, Jun 2025).

• Gartner estimates that only about 130 of the thousands of agentic AI vendors are real, and warns of "agent washing", the rebranding of existing products as agents (Gartner, Jun 2025).
• Only 16% of enterprise deployments and 27% of startup deployments qualify as true agents, meaning an LLM that plans, acts and adapts (Menlo Ventures, Dec 2025).
• Close to three-quarters of companies in Deloitte's survey plan to deploy agentic AI within two years, yet only 21% of them report a mature model for agent governance (Deloitte AI Institute, Jan 2026).
• Gartner expects 33% of enterprise software applications to include agentic AI by 2028, up from less than 1% in 2024 (Gartner, Jun 2025).

How many people use generative AI outside work?

The Stanford Digital Economy Lab study puts the US adult user base of generative AI chatbots at 115 million in March 2026, up from 98 million in July 2025, and finds the median user values access at $11.40 a month, up from $3.40 (Stanford Digital Economy Lab, Apr 2026).

• The same study put the mean amount US adults would need to be paid to give up chatbots for a month at $124.50, up 27% from $98 (Stanford Digital Economy Lab, Apr 2026).
• The International AI Safety Report says adoption likely remains below 10% across much of Africa, Asia and Latin America (International AI Safety Report, Feb 2026).

How much capital is flowing into generative AI?

The money has moved from software experiments to power, chips and cloud contracts. The largest checks now go to infrastructure.

How much do companies spend on generative AI?

Applications took more than half of it: $19 billion of enterprise spend went to AI apps in 2025, and the rest to infrastructure (Menlo Ventures, Dec 2025).

• 76% of AI use cases were bought rather than built in-house, against 53% bought in 2024 (Menlo Ventures, Dec 2025).
• 47% of AI deals went into production, against 25% for traditional SaaS (Menlo Ventures, Dec 2025).
• Coding reached $4.0 billion, 55% of departmental AI spend and the largest single category (Menlo Ventures, Dec 2025).
• Foundation model APIs took $12.5 billion of the $18 billion spent on AI infrastructure (Menlo Ventures, Dec 2025).
• The ambient scribe market for clinicians reached $600 million, 2.4 times the year before (Menlo Ventures, Dec 2025).

Menlo's market sizes are estimates, and the sample has limits worth knowing (Menlo Ventures, Dec 2025). Treat them as one view of the market.

How much venture money is going into AI companies?

Venture funding is at record levels and most of it follows AI. Global venture investment reached $205 billion in Q2 2026 across more than 5,000 startups, after $305 billion in Q1 (Crunchbase, Jul 2026).

• AI startups took more than 60% of Asian venture funding in Q2 2026, just over $26 billion and a record (Crunchbase, Jul 2026).
• Chinese startups raised just over $30 billion in the quarter, up 424% on a year earlier (Crunchbase, Jul 2026).
• DeepSeek raised $7.4 billion at a reported $50 billion valuation in June 2026 (Crunchbase, Jul 2026).
• European startups raised $24 billion, two-thirds more than in Q2 2025, with 65% going to 42 companies raising $100 million or more (Crunchbase, Jul 2026).
• SpaceX said it would acquire the AI coding tool maker Anysphere, which makes Cursor, for $60 billion (Crunchbase, Jul 2026).

What are the leading AI labs worth and earning?

Valuations now rest on revenue that is growing fast. Anthropic raised $30 billion in a Series G at a $380 billion post-money valuation in February 2026 (Anthropic, Feb 2026).

• The number of Anthropic customers spending over $1 million a year went from a dozen two years earlier to more than 500 (Anthropic, Feb 2026).
• Claude Code's run-rate revenue passed $2.5 billion, more than double its level at the start of 2026 (Anthropic, Feb 2026).
• Menlo counts at least 10 AI products with more than $1 billion in annual recurring revenue and 50 with more than $100 million (Menlo Ventures, Dec 2025).
• AI startups captured 63% of application-layer spending in 2025, up from 36% a year earlier (Menlo Ventures, Dec 2025).
• Mistral AI raised EUR 1.7 billion at a EUR 11.7 billion post-money valuation, led by chip equipment maker ASML (Mistral AI, Sep 2025).

Who earns the money from the AI buildout?

The chip and cloud suppliers, first. NVIDIA reported revenue of $96.2 billion for the quarter ended 26 July 2026, up 106% on a year earlier (NVIDIA, Aug 2026).

• NVIDIA guided to $108.0 billion for the next quarter without assuming any Data Center compute revenue from China (NVIDIA, Aug 2026).
• Oracle's remaining performance obligations rose $209 billion in a year to $664 billion (Oracle, Sep 2026).
• Oracle's cloud infrastructure revenue grew 121% to $7.4 billion in the quarter, while free cash flow was negative $5 billion (Oracle, Sep 2026).
• Oracle booked more than $30 billion of additional AI cloud contracts in the quarter and sold $20 billion of stock to help fund its buildout (Oracle, Sep 2026).
• CoreWeave's revenue was $2,575 million against $1,212 million a year earlier, with a net loss of $626 million (CoreWeave, Aug 2026).
• CoreWeave's revenue backlog was about $104 billion at 30 June 2026, and its active power reached 1.5 GW (CoreWeave, Aug 2026).
• NVIDIA and OpenAI signed a letter of intent for at least 10 gigawatts of NVIDIA systems, with NVIDIA intending to invest up to $100 billion in OpenAI (NVIDIA, Sep 2025).
• AMD agreed to supply OpenAI with 6 gigawatts of GPUs and issued OpenAI a warrant for up to 160 million AMD shares (AMD, Oct 2025).

Which models lead, and what does it cost to run them?

Progress still comes from scale, and the top of the leaderboard changes by the month. Cost per unit of capability keeps falling while the bill for the whole system keeps rising.

Which companies and models lead the market?

In enterprise LLM spending, Anthropic leads with an estimated 40% share, up from 12% in 2023, while OpenAI fell to 27% from 50% (Menlo Ventures, Dec 2025).

• Open-source models hold only 11% of enterprise LLM spending (Menlo Ventures, Dec 2025).
• Hugging Face uploads tripled since 2023, and GitHub hosts 5.6 million open-source AI projects (Stanford HAI, Apr 2026).
• Parameter counts have stayed near 1 trillion for three years, and several frontier labs stopped reporting them (Stanford HAI, Apr 2026).
• Epoch AI's database tracks over 3,600 machine learning models from 1950 to today (Epoch AI, Oct 2026).
• Frontier performance reaches consumer hardware in about 8 months (Epoch AI, Feb 2026).

How fast are generative AI models improving?

Faster since reasoning models arrived. Epoch's capability index for frontier models has advanced 14 points a year since September 2024, against 6 points a year for non-reasoning models (Epoch AI, Feb 2026).

• The task length that METR's agents complete half the time has doubled about every seven months since 2019 (METR, Mar 2025).
• METR warns that measurements above 16 hours are unreliable with its current task suite, so the benchmark is close to its ceiling (METR, Sep 2026).
• AI agents can now reliably complete some tasks that take a human programmer about half an hour, up from under 10 minutes a year earlier (International AI Safety Report, Feb 2026).
• On OSWorld, agent accuracy rose from roughly 12% to 66.3% in a year, within 6 points of human performance (Stanford HAI, Apr 2026).
• Gemini Deep Think scored 35 points, a gold medal, at the 2025 International Mathematical Olympiad, up from a 28-point silver in 2024 (Stanford HAI, Apr 2026).
• LLM context windows have grown 30 times a year since 2023 (Epoch AI, Feb 2026).
• A review of widely used benchmarks found invalid question rates from 2% on MMLU Math to 42% on GSM8K, so scores deserve caution (Stanford HAI, Apr 2026).

What does it cost to train and run models?

Each unit of capability gets cheaper while the total bill grows. The price of reaching GPT-4's performance on PhD-level science questions fell 40 times a year, and price drops ranged from 9 to 900 times a year depending on the milestone (Epoch AI, Mar 2025).

• Training compute among the top five models has grown by a factor of about 10,000 since 2020 (Epoch AI, Feb 2026).
• A typical 1 gigawatt AI data center costs roughly $38 billion up front and takes about two years to build (Epoch AI, Feb 2026).
• Five hyperscalers own over two-thirds of global AI compute (Epoch AI, Feb 2026).
• The largest known AI data center has a capacity equal to 1.1 million NVIDIA H100 chips, and Meta's Hyperion is expected to reach 3.7 million by January 2028 (Epoch AI, Feb 2026).
• The United States hosts 5,427 data centers, more than ten times any other country (Stanford HAI, Apr 2026).
• 54% of US adults say data centers are mostly bad for the environment, up from 39% in January 2026, and 50% say they are bad for home energy costs, up from 38% (Pew Research Center, Sep 2026).
• Just 4% of Americans think data centers have a mostly good impact on the environment, energy costs or quality of life (Pew Research Center, Sep 2026).

What does generative AI do to productivity and jobs?

Lab experiments show large gains on defined tasks. Field evidence is thinner, and the people who use the tools tend to overstate what they gain.

How much faster do people work with generative AI?

Consultants using GPT-4 on tasks inside the AI frontier completed 12.2% more tasks and finished 25.1% faster in a Harvard Business School experiment (Harvard Business School, Sep 2023).

• On a task outside the frontier, consultants using AI were 19% less likely to produce a correct solution (Harvard Business School, Sep 2023).
• In an experiment with 444 college-educated professionals, ChatGPT cut writing time by 0.8 standard deviations and raised quality by 0.4, helping lower-ability workers most (MIT, Mar 2023).
• Developers with GitHub Copilot finished an HTTP server task 55.8% faster than the control group (Peng et al., Feb 2023).
• In a follow-up with late-2025 tools, METR's raw estimate for returning developers was an 18% reduction in task time, with a confidence interval from 38% less to 9% more. METR called its own data an unreliable signal because developers refused to work without AI (METR, Feb 2026).
• In a survey of 349 technical workers, the median self-reported change in the value of work was 1.4 to 2 times, and the median self-reported speed change was 3 times (METR, May 2026).
• Respondents estimated a 1.3 times value gain in March 2025, 2 times in March 2026, and forecast 2.5 times for March 2027 (METR, May 2026).
• METR's earlier 2025 study found that people overestimated AI's effect on their time spent on tasks by 40 percentage points on average (METR, May 2026).

Which jobs are most exposed to generative AI?

Around 80% of the US workforce could have at least 10% of their tasks affected by large language models, and about 19% could see at least half of their tasks affected (Eloundou et al., Mar 2023).

• In high-income countries, the jobs at the highest risk of automation make up 9.6% of female employment and 3.5% of male employment (ILO, May 2025).
• Productivity growth is 40% higher at companies most exposed to AI than at the least exposed (PwC, Jun 2026).
• The top fifth of the most AI-exposed companies averaged 163% productivity growth since 2022 (PwC, Jun 2026).
• Skills needed for the most AI-exposed jobs are changing more than twice as fast as for the least exposed, a 75% widening of the gap from a year earlier (PwC, Jun 2026).
• The most AI-exposed junior roles are 7 times more likely to demand traditionally senior skills (PwC, Jun 2026).

Exposure measures tasks and says nothing about job losses. The ILO says the most likely outcome is transformation of jobs (ILO, May 2025).

Are companies cutting jobs because of generative AI?

Rarely so far. Only 4% of New York region service firms laid off workers because of AI in the six months to August 2026, while 15% hired fewer people and 13% hired more (Federal Reserve Bank of New York, Sep 2026).

• Federal Reserve Board staff found no evidence that industries or firms with higher AI adoption post fewer jobs (Federal Reserve Board, Mar 2026).
• Executives in the Atlanta Fed survey reported a 1.8% increase in output per worker from AI in 2025, though the gains implied by revenue were much smaller (Federal Reserve Bank of Atlanta, Mar 2026).
• Large companies in that survey expect AI to reduce their employment by 0.8% in 2026, while the average across all firms is close to zero (Federal Reserve Bank of Atlanta, Mar 2026).
• Executives expect the share of routine clerical workers to fall 0.76% in 2026 and 2.19% by 2028 (Federal Reserve Bank of Atlanta, Mar 2026).
• In 34 of 37 countries Pew surveyed in 2026, people tend to believe AI will lead to fewer jobs rather than more (Pew Research Center, Sep 2026).
• In Australia, South Korea and the United States, around seven in ten adults or more expect AI to cost jobs over the next 20 years, and the US share is up 7 points in two years (Pew Research Center, Sep 2026).
• For the first time, a majority of US adults under 30 (55%) say they are more concerned than excited about AI (Pew Research Center, Aug 2026).

How reliable and safe is generative AI?

Models are fluent and often wrong, and the failure is hard to see from the output alone. The same fluency is used for fraud and fills the web with machine-written text.

How often does generative AI hallucinate?

It depends on the task and the model. On Vectara's September 2026 summarization leaderboard, hallucination rates ran from 1.8% for the best model to 24.2% for the worst (Vectara, Sep 2026).

• AI legal research tools from LexisNexis and Thomson Reuters hallucinated between 17% and 33% of the time in a Stanford RegLab study (Stanford RegLab, May 2024).
• NewsGuard found the 10 leading chatbots repeated false claims on news topics 35% of the time in August 2025, up from 18% a year earlier (NewsGuard, Sep 2025).
• The chatbots' non-response rate fell from 31% to 0% as they adopted live web search, so they now answer every news prompt, right or wrong (NewsGuard, Sep 2025).
• When a false statement was framed as the user's belief, GPT-4o's accuracy fell from 98.2% to 64.4% and DeepSeek R1's from over 90% to 14.4% (Stanford HAI, Apr 2026).

If assistants get news and facts wrong, they get brands wrong too. What they say about a company depends on which sources they find and trust.

How is generative AI used for fraud and cyberattacks?

Investment scams carry most of the AI-linked losses in the FBI's 2025 data. Complaints with a reported AI link lost more than $632 million, out of more than $8 billion lost to investment scams overall (FBI IC3, Apr 2026).

• Businesses reported losses over $30 million to business email compromise scams involving AI in 2025 (FBI IC3, Apr 2026).
• In one study of audio deepfakes, listeners mistook AI-generated voices for real speakers 80% of the time (International AI Safety Report, Feb 2026).
• UK AI Security Institute testing found models complete apprentice-level cyber tasks 50% of the time on average, against just over 10% in early 2024 (UK AI Security Institute, Dec 2025).
• In 2025 the institute tested the first model able to complete expert-level cyber tasks that typically require over 10 years of experience (UK AI Security Institute, Dec 2025).
• The length of cyber tasks models can complete unassisted is doubling roughly every eight months (UK AI Security Institute, Dec 2025).
• The institute found a 40 times difference in expert effort needed to jailbreak two models released six months apart, and vulnerabilities in every system it tested (UK AI Security Institute, Dec 2025).
• In chemistry and biology, models exceed the institute's expert baseline for open-ended questions by up to 60%, and are up to 90% better than human experts at wet lab troubleshooting (UK AI Security Institute, Dec 2025).

Is AI-generated code secure?

Not by default. Veracode found 45% of code samples from over 100 LLMs failed security tests and introduced OWASP Top 10 vulnerabilities (Veracode, Jul 2025).

• Java had a 72% security failure rate, and models failed to defend against cross-site scripting in 86% of relevant samples (Veracode, Jul 2025).
• Security performance stayed flat regardless of model size or training sophistication, even as functional correctness improved (Veracode, Jul 2025).

How much of the web is AI-written, and who pays for training data?

Over one-third of webpages published after ChatGPT's release show signs of AI authorship in Pew's July 2026 snapshot (Pew Research Center, Aug 2026).

• In 2026 samples, about one in ten .com pages shows signs of AI authorship, against 4.6% of .org pages and about 1% of .edu and .gov pages (Pew Research Center, Aug 2026).
• The analysis covered 490,000 English-language webpage texts from Common Crawl, scored with the Open Pangram detector, which Pew notes is not perfect on individual pages (Pew Research Center, Aug 2026).
• The court granted final approval of the Anthropic authors' settlement on 20 July 2026, with first payments expected on or before 15 November 2026 (Bartz v. Anthropic settlement administrator, Jul 2026).

Readers and AI systems now meet machine-written text often.

How is generative AI regulated?

Law is arriving in three layers: binding rules in the EU, a fight over who sets the rules in the United States, and a patchwork of national laws and voluntary company commitments elsewhere.

What does the EU AI Act require, and what are the penalties?

The Act presumes a general-purpose model has high-impact capabilities, the trigger for systemic-risk duties, when its training computation exceeds 10^25 floating point operations (European Union, Jul 2024).

• Supplying incorrect, incomplete or misleading information to authorities carries fines up to EUR 7.5 million or 1% of turnover (European Union, Jul 2024).
• The Commission can fine providers of general-purpose AI models up to 3% of turnover or EUR 15 million (European Union, Jul 2024).
• The European Parliament approved the Act on 13 March 2024 with 523 votes in favour, 46 against and 49 abstentions (European Parliament, Mar 2024).
• The Commission lists more than 20 providers, including Amazon, Anthropic, Google, IBM, Microsoft, Mistral AI and OpenAI, as signatories of the GPAI Code of Practice, and says xAI signed only the Safety and Security chapter (European Commission, Oct 2026).

How are US federal and state governments responding?

With more rules and more conflict. US federal agencies took 58 AI-related regulatory actions in 2025, up from one in 2016 (Stanford HAI, Apr 2026).

• Executive Order 14365 of 11 December 2025 told the Attorney General to set up an AI Litigation Task Force within 30 days to challenge state AI laws (The White House, Dec 2025).
• The order directs Commerce to identify onerous state AI laws within 90 days and makes states with such laws ineligible for remaining BEAD non-deployment funds to the extent federal law allows (The White House, Dec 2025).
• California's SB 53, signed on 29 September 2025, is described by the governor as the first-in-the-nation frontier AI safety law, and it sets up a consortium called CalCompute to design a public computing cluster (Governor of California, Sep 2025).
• SB 53 also creates a mechanism to report critical safety incidents to California's Office of Emergency Services and a civil penalty enforceable by the Attorney General (Governor of California, Sep 2025).
• Texas's Responsible Artificial Intelligence Governance Act was signed on 22 June 2025 and took effect on 1 January 2026 (Texas Legislature, Jun 2025).
• AI-related witnesses in US congressional hearings rose from 5 in 2017 to 102 in 2025 (Stanford HAI, Apr 2026).
• The US invested about $20.4 billion in AI-related public contracts and grants between 2013 and 2024 (Stanford HAI, Apr 2026).

How are regulators and companies elsewhere responding?

Enforcement is uneven and sometimes reversed. Italy's data protection authority fined OpenAI EUR 15 million in December 2024 over ChatGPT, and a Rome court upheld an appeal against the decision in a judgment published on 18 March 2026 (Garante per la protezione dei dati personali, Mar 2026).

• European public commitments to AI contracts reached about $3.7 billion between 2013 and 2024, of which the UK accounted for $1.6 billion (Stanford HAI, Apr 2026).
• State-backed AI supercomputing clusters in Europe and Central Asia grew from 3 to 44 between 2018 and 2025 (Stanford HAI, Apr 2026).
• In 2025, 12 companies published or updated Frontier AI Safety Frameworks, documents describing how they plan to manage risks from more capable models (International AI Safety Report, Feb 2026).
• The share of businesses with no responsible AI policies fell from 24% to 11%, while the top obstacles remain knowledge gaps (59%), budget (48%) and regulatory uncertainty (41%) (Stanford HAI, Apr 2026).

Methodology and Sources

We included a number only after opening the page that publishes it and confirming the figure and its date. Figures quoted in roundups were followed back to the original report, filing, paper or dataset, or dropped. Most numbers come from central banks, statistical and regulatory bodies, courts, company filings and academic or non-profit research; survey firms, consultancies and vendors (Menlo Ventures, PwC, KPMG, Deloitte, Gartner, Crunchbase, Veracode) are named in every citation. Where two sources measure different things, such as firms that use AI and workers who use it, we say so instead of blending them. Several figures come from live dashboards (Vectara, Epoch AI, the court-case database), so their dates are the month we read them. The EU, US and national rules change often, so legal dates are taken from the official text or press release. This page is checked and refreshed every month.

Sources

• AMD, AMD and OpenAI Announce Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs
• Anthropic, Series G announcement
• Bartz v. Anthropic Settlement Administrator
• Bartz v. Anthropic Settlement Administrator, Frequently Asked Questions
• Brynjolfsson, Collis, Eggers, Kazinnik and Nguyen, What is Generative AI Worth? (Stanford Digital Economy Lab)
• CoreWeave, Second Quarter 2026 Results
• Crunchbase News, China And AI Lead Asia's Startup Funding To Multiyear Peak In Q2
• Crunchbase News, Europe Posted Its Strongest Venture Funding Quarter In 4 Years
• Crunchbase News, Global Startup Investment Hit Record $510B In H1 2026
• Damien Charlotin, AI Hallucination Cases Database
• Deloitte AI Institute, State of AI in the Enterprise 2026
• Eloundou, Manning, Mishkin and Rock, GPTs are GPTs (arXiv)
• Epoch AI, AI Models database
• Epoch AI, LLM inference prices have fallen rapidly but unequally across tasks
• Epoch AI, Trends in AI
• European Commission, Signatories of the GPAI Code of Practice
• European Parliament, Artificial Intelligence Act: MEPs adopt landmark law
• European Union, Regulation (EU) 2024/1689 (AI Act), Article 101
• FBI Internet Crime Complaint Center, 2025 Internet Crime Report
• Federal Reserve Bank of Atlanta, How Might AI Change the Workplace? Evidence from Corporate Executives
• Federal Reserve Bank of New York, Regional Business Surveys
• Federal Reserve Banks, Small Business Credit Survey (via New York Fed)
• Federal Reserve Board, AI Adoption and Firms' Job-Posting Behavior (FEDS Notes)
• Garante per la protezione dei dati personali, ChatGPT: the Authority closes the preliminary investigation
• Gartner, Over 40% of Agentic AI Projects Will Be Canceled by End of 2027
• Harvard Business School Working Paper, Navigating the Jagged Technological Frontier
• IBM Institute for Business Value, CEO Study
• ILO and NASK, Generative AI and Jobs: A Refined Global Index of Occupational Exposure
• International AI Safety Report 2026 (chaired by Yoshua Bengio)
• KPMG, Quarterly AI Pulse Survey Q3 2026
• Menlo Ventures, 2025: The State of Generative AI in the Enterprise
• METR, Measuring AI Ability to Complete Long Tasks (arXiv)
• METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity
• METR, Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker Productivity
• METR, Task-Completion Time Horizons of Frontier AI Models
• METR, We are Changing our Developer Productivity Experiment Design
• Mistral AI, Series C announcement
• NewsGuard, AI False Information Rate Nearly Doubles in One Year
• Noy and Zhang, Experimental Evidence on the Productivity Effects of Generative AI (MIT)
• NVIDIA, OpenAI and NVIDIA Announce Strategic Partnership to Deploy 10 Gigawatts of NVIDIA Systems
• NVIDIA, Second Quarter Fiscal 2027 Results
• Office of Governor Gavin Newsom, Governor Newsom signs SB 53
• Oracle, Q1 FY27 Results
• Peng et al., The Impact of AI on Developer Productivity: Evidence from GitHub Copilot (arXiv)
• Pew Research Center, Americans' views of data centers have turned more negative
• Pew Research Center, Globally, More People Expect AI to Cause Job Loss Than Growth
• Pew Research Center, How Much of the Internet Is Written With AI?
• Pew Research Center, Young adults in the U.S. are increasingly wary of AI
• PwC, 2026 AI Jobs Barometer
• PwC, 29th Global CEO Survey
• Stanford HAI, AI Index Report 2026
• Stanford RegLab and HAI, Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (arXiv)
• Texas Legislature Online, HB 149 bill history
• The White House, Executive Order 14365 on a national policy framework for AI
• UK AI Security Institute, Frontier AI Trends Report
• Vectara, Hallucination Leaderboard (HHEM-2.3)
• Veracode, 2025 GenAI Code Security Report
• Wharton Human-AI Research and GBK Collective, Accountable Acceleration

Glossary

• Agentic AI: AI systems that plan and carry out multi-step tasks with limited human input.
• ARR: annual recurring revenue.
• BEAD: the US Broadband Equity, Access, and Deployment program.
• Confabulation / hallucination: a model stating false content with confidence.
• ECI: Epoch Capabilities Index, a combined score of model performance across benchmarks.
• FLOP: a floating point operation, the unit used to measure training compute.
• GPAI: general-purpose AI model, the EU term for broad foundation models.
• H100-equivalent: computing capacity expressed in NVIDIA H100 chips.
• LLM: large language model.
• RPO: remaining performance obligations, contracted revenue not yet recognised.
• Time horizon: the length of task, measured by human expert time, that an AI agent completes at a stated success rate.