About this edition

Written and researched by Indranil Banerjee (Neel), Neelverse Lab, with AI research agents. indranil.in

Data as of. Evidence to 1 October 2026; figures carry their own dates, and older figures are the latest available. A cross-check on 2 November 2026 re-tested 491 claims against live sources, and new data from 30 October to 2 November 2026 was added. Every change is listed in the edition's change log.

How to cite. Banerjee, I. (2026). The AI Opportunity Report 2026–27, Global edition. Neelverse Lab.

What this edition covers

Updates. This is the annual edition. Later updates will refresh the numbers, the rules calendar and the country dashboards.

Important notice. This report is for general information. It is not legal, financial or investment advice. Figures come from public sources, which the report names; many company results are the company's own claims and are labelled as such. Market forecasts are third-party projections, not predictions by the author.

© 2026 Neelverse Lab. All rights reserved. Short extracts may be quoted with attribution.

Everyone is using AI. Few are winning with it yet.

A note from the author

Every week I sit with leaders who ask the same three questions. Where is the real opportunity? What are others buying and paying? And what should we do in the next twelve months? Most AI reports answer a different question: how big the market could be.

This report is built to answer the practical questions. It looks at AI from the buyer's side, country by country and industry by industry, and at the services layer where most businesses meet AI: the audits, agents, content engines, voice systems and training that firms actually buy.

The picture is striking. Use is close to universal among large firms, yet only a minority see profit from it. Spending is racing ahead, mostly into chips and data centres. Small, digital, well-connected countries lead on everyday use. And the rules are arriving on fixed dates.

I built this edition the way I recommend others work: with AI and people together. I ran twelve parallel research streams with AI research agents, then checked, edited and cut the findings by hand. Every number links to its source and date.

No single report can be complete in a market that changes monthly. Where the evidence is thin, I say so. I hope it helps you make better bets.

Indranil Banerjee (Neel), Neelverse Lab

How this report was built

A buyer's-eye view of AI in 46 countries and 25 industries, built from public evidence and checked link by link. Here is what went in, how it was checked and how to read it.

Scope

Method

The research ran in two layers. The first layer built the services catalogue, industry maps, price benchmarks and rules register over four research rounds. The second layer ran twelve parallel research streams, one per region plus a global stream, each collecting official statistics, filings, regulator pages and reputable reporting.

Indranil Banerjee did the work with AI research agents: he set the questions, checked the sources and edited every chapter. Each figure carries its source, link, date and a type label. When no credible figure could be found, the gap is marked "no figure yet" or the figure is left out, rather than filled with an estimate. A second cross-check on 2 November 2026 re-tested 491 claims against live sources; its corrections are applied in this edition.

How to read the evidence

Comparable measures used across countries

Limits

Executive summary · 2026–27 in numbers

AI has moved from pilot to habit. Spending is up by half in a year, billions of people now meet AI in search and chat, and new rules are live on three continents. But the money and the value are concentrated.

Exhibit 1: Everyday AI use by country: share of working-age people using generative AI, Q2 2026

Source: Microsoft AI Economy Institute, Global AI Diffusion Q2 2026 (74 economies shown).

The leaders are small, rich and digital: the UAE (73%), Singapore (64%), Ireland, France and Norway (about 49–50%). Large economies sit lower: the US at 33%, Japan at 25%, India at 18.5% and China at about 18% on this measure. The global average is 18.8%, with the Global North at 28.8% and the Global South at 16.2%.

Executive summary · Twelve findings

Finding 01 · Use is near universal. Profit is not.

Nine in ten organisations now use AI, and 44% say it is scaling. Yet the share seeing any profit impact is stuck at 37%, and only about 6% are AI high performers: they attribute 5% or more of EBIT to AI and report significant value from it (McKinsey 2026). 56% of CEOs saw neither higher revenue nor lower costs in the past year (PwC). The gap is not access to models. It is redesigning work: 73% of high performers rebuilt their workflows, against 25% of the rest.

Finding 02 · Spending is racing ahead, mostly into infrastructure.

Gartner now expects worldwide AI spending of $2.67 trillion in 2026, up 49.5%, and $3.64 trillion in 2027. It has raised its 2026 forecast twice since January. More than half goes to AI infrastructure: servers, chips and data centres. Services, where most companies actually buy help, grow from $434bn to $746bn over the three years.

Finding 03 · The capital map is lopsided.

Private AI investment reached $345bn in 2025. The US took $286bn of it, 23 times China, which came second. Everyone else combined raised about a fifth of the US total. The UK, France, Canada, India and Germany follow, each with roughly $4–6bn. Outside the US, AI is being built mostly with public money and big-tech capex, not venture capital.

Finding 04 · Everyday use is led by small, connected countries. The gap is widening.

The top of the AI-use table is the UAE, Singapore, Ireland, France, Norway and Spain. Big economies trail: the US ranks 21st. In emerging markets, use is growing but from a low base, and the gap between the Global North (28.8%) and South (16.2%) is widening. Where connectivity, English or local-language models, and digital government are strong, use follows.

Finding 05 · Big firms are pulling away from small ones.

In the EU, 55% of large firms use AI against 17% of small ones. In Singapore it is 62.5% of non-SMEs against 14.5% of SMEs, and in Ireland 57.7% against 17.2%. In the UK, the government's AI Adoption Survey 2026 puts large firms at 52% against 32% of all firms. McKinsey finds 40% of $1bn-plus companies scaling AI agents, up from 27%, while smaller firms are flat at 22%. The next wave of opportunity is making AI simple and affordable for the mid-market and small businesses, which is where most jobs sit.

Finding 06 · Search is becoming an answer engine. Ads and checkout follow.

Google's AI Mode has passed a billion monthly users and the Gemini app 950 million (company figures). OpenAI says ChatGPT has 1.2 billion weekly users, and its ads business reached a $1bn annualised run-rate within about 200 days, OpenAI said in August (claimed). Three agent-commerce protocols launched in four months (Google AP2, OpenAI/Stripe ACP, Google UCP). OpenAI has since moved away from checkout inside ChatGPT (CNBC), but checkout inside assistants is live in the US at Google, through Universal Cart since 15 October, and at Meta, whose Muse agent buys through Stripe's Link (company announcements). For marketers, being the answer now matters as much as ranking, and users clicked a source link inside an AI Overview in only about 1% of visits (Pew Research, July 2025).

Finding 07 · The AI services economy is booming, and moving to outcomes.

Gartner puts AI services at $576bn in 2026. Among the largest consultancies, Accenture grew 5% in local currency in fiscal 2026 (7% in the June-August quarter) 1 and IBM consulting was flat. Buyers are shifting from hours to outcomes: Intercom prices its AI agent at $0.99 per outcome (resolutions, completed workflows and handoffs) and Salesforce Agentforce at $2 per conversation. Buying through a partner works twice as often as building alone (66% vs 33% success, MIT NANDA; source could not be re-opened on 3 Nov 2026).

Exhibit 7: AI services spending worldwide, USD bn

Source: Gartner (Sep 2026)

Finding 08 · The rules arrive labels first, heavy duties later.

Across major markets the first binding rules are about transparency: label AI content, disclose chatbots and automated decisions. Heavier obligations are being delayed. The EU's AI Omnibus moved high-risk duties to December 2027 and August 2028, while labelling applies now. In the US, federal pre-emption has not stopped the states, and 1 January 2027 is the next big compliance date.

Finding 09 · A compute and sovereignty race is under way.

The four largest US platforms are spending well over $600bn a year on capital expenditure, mostly for AI. IDC counts $318bn of AI infrastructure spend in 2025; the US took about three quarters in the latest quarters (77% in Q4 2025, 75.7% in Q1 2026). Governments are building their own capacity: Stargate UAE, Saudi Arabia's HUMAIN, the IndiaAI Mission's 38,000-plus subsidised GPUs (with 20,000 more announced), EU AI gigafactories and a UK AI Research Resource with 21 exaflops of public compute in 2025 and 420 targeted by 2030.

Finding 10 · Jobs: an entry-level squeeze, not mass layoffs.

Employment of 22–25-year-olds in AI-exposed US jobs is 19% below trend (Stanford/ADP data). Yet only 14% of organisations actually cut headcount because of AI last year, against 32% who expected to; 39% now expect cuts next year (McKinsey). The WEF still sees 170 million jobs created and 92 million displaced by 2030. The pressure is on junior roles and on how people are trained.

Exhibit 9: Work and AI: the early evidence

Source: Brynjolfsson, Chandar & Chen, Canaries in the Coal Mine?; McKinsey, The state of AI in 2026: On the road to ROI; World Economic Forum, Future of Jobs Report 2025

Finding 11 · China runs a parallel AI economy at near-zero prices.

More than 700 million people in China used generative AI in the first half of 2026, over half the population, and in CNNIC's 2025 survey more than 90% picked a domestic model first. 1,112 AI services had filed with the regulator by August. API prices have fallen close to the floor (DeepSeek at RMB 1 per million input tokens off-peak), and in June ByteDance's Volcano Engine shifted its pitch from price cuts to capability (Doubao 2.1). Chinese open-weight models are now the default in many emerging markets.

Finding 12 · Emerging markets: fast-growing use on thin infrastructure.

OpenAI says India has about 100 million weekly ChatGPT users (claimed), yet only 18.5% of working-age people use AI on Microsoft's measure. Brazil, Mexico and South Africa sit near or above the global average; Indonesia, Kenya, Bangladesh and Pakistan below it. Africa holds about 0.6% of global data-centre capacity (World Bank, October 2026) 4, and several flagship projects have stalled. Where WhatsApp, mobile money and local languages meet AI, adoption jumps: that is the design brief for 2026–27.

Executive summary · What it means for you

CMOs and marketing leaders

CEOs and boards

Founders and SMB owners

Investors

Agencies and service providers

Part 7 turns these into full playbooks, with checklists and 90-day plans for each reader.

Executive summary · The dates that matter, November 2026 to December 2027

Every rule date is tracked in the Part 4 calendar; sources are in the country chapters in Part 2. Check the latest official texts before acting.

Executive summary · Notes and sources

Most figures in this summary are sourced where the full text gives them, chiefly in Part 1. Numbers in square brackets in the summary link to these sources.

1.1 · How big, how fast: the spending boom

AI is now the fastest-growing line in global technology budgets. Most of the money is going into the plumbing, not the apps.

Gartner's September 2026 forecast puts worldwide AI spending at $2.67 trillion this year, up 49.5% from $1.79 trillion in 2025, and $3.64 trillion in 2027 5. The number keeps going up. In January the 2026 figure was $2.52 trillion. In May it was $2.59 trillion 67. Each update added tens of billions.

Exhibit 1.1: Gartner's 2026 AI spending forecast, by edition, USD bn

Source: Gartner press releases, 15 Jan, 19 May and 16 Sep 2026

More than half of every AI dollar goes to infrastructure: servers, chips and data centres, bought mostly by hyperscalers and model labs. Gartner's John-David Lovelock expects that server buying to stay "the largest single area of spending" 5 (source could not be re-opened on 3 Nov 2026). Services are the second-largest line. Models themselves are a sliver.

The fastest-growing lines are small today. AI data, agents and generative models all grow about four to eight times between 2025 and 2027. That is where new vendors and new budget lines will appear first.

Other lenses on the same boom

IDC tracks only AI servers and storage. That spend more than doubled from $153bn in 2024 to $318bn in 2025 8. IDC now expects $497bn in 2026 and more than a trillion dollars a year by 2029 8. In the first quarter of 2026, ARM-based accelerated servers ($53bn) widened their lead over x86 ($34.6bn); the crossover began in Q4 2025 8.

Exhibit 1.3: AI infrastructure spending (servers and storage), USD bn

Source: IDC Worldwide Quarterly AI Infrastructure Tracker (21 Jul 2026); (f) = IDC forecast

Enterprise software spend looks very different. Menlo Ventures estimates companies spent $37bn on generative AI in 2025, up 3.2 times from $11.5bn in 2024 9. About $19bn went to applications and $18bn to infrastructure. Coding tools were the first breakout use, at $4bn 9.

How to read the market numbers. Gartner, IDC and Menlo measure different things. Gartner counts everything, including devices with AI chips. IDC counts servers and storage. Menlo counts enterprise generative AI software. Never add them up or put them on one chart. The "trillions" are mostly hardware bought by a handful of giants. Enterprise application spend is in the tens of billions.

1.2 · Adoption and the value gap

Almost every large company uses AI somewhere. Only about one in three sees it in the profit line.

McKinsey's 2026 survey of 1,719 organisations in 97 countries finds 89% using AI in at least one business function and 56% in three or more 10. Scaling is rising. Profit is not. Only 37% report any EBIT impact, about the same share as a year ago. McKinsey's own summary: "conviction in AI is growing faster than the immediate financial returns" 10.

The other big surveys agree. PwC found 56% of 4,454 CEOs saw neither higher revenue nor lower costs from AI in the past year; only 12% saw both 11. BCG counts 5% of firms as "future-built" and 60% getting "hardly any material value" 12. MIT's NANDA group went further: 95% of organisations get zero measurable return from generative AI, though on a small sample 13.

Two newer data points fit the pattern. The UK government's AI Adoption Survey 2026 finds 32% of UK businesses use AI, but only 14% rate it as of high or critical importance to their operations 14. And a supplier's filing shows how shallow enterprise spend still is. Anthropic's S-1 says more than half of the Fortune 500 were $100,000-plus customers, yet in the quarter to 30 September their Anthropic spend averaged less than 0.1% of their total expenses 15. That is the vendor's own framing.

What separates the winners

The winners change how work gets done. 73% of McKinsey's high performers fundamentally redesigned workflows, against 25% of everyone else 10. BCG's leaders put up to 64% more of their IT budget into AI and get five times the revenue uplift 12. PwC's "vanguard" CEOs apply AI to products and services far more often, 44% against 17% 11.

Everyday use: the Microsoft league table

Microsoft's AI Economy Institute estimates the share of working-age people (15–64) who used a generative AI product. The global figure was 18.8% in Q2 2026, up from 15.1% in the first half of 2025 1617. Use is growing everywhere, but faster in rich countries. The gap between the Global North (28.8%) and South (16.2%) widened to 12.6 points 16.

Exhibit 1.6: Share of the world's working-age people using generative AI, %

Source: Microsoft AI Economy Institute, Global AI Diffusion reports 1617

Exhibit 1.7: League table: share of working-age people using AI, 74 economies, Q2 2026

Source: Microsoft AI Economy Institute, Global AI Diffusion Q2 2026 16. Only economies read directly from the report are shown. Highlighted: the largest economies.

Readiness helps, but it is not destiny. The IMF's AI Preparedness Index scores countries on digital infrastructure, skills, innovation and rules 18. Plot it against actual use and the pattern is clear: better-prepared countries use more AI. The outliers are the story. The Gulf states and Singapore use far more than their scores predict. Denmark, Germany and Japan use less.

Exhibit 1.8: AI readiness vs everyday AI use, by country

Source: IMF AI Preparedness Index 18; Microsoft Global AI Diffusion Q2 2026 16. Each dot is an economy with both measures.

Exhibit 1.9: AI readiness around the world: IMF AI Preparedness Index

Source: IMF DataMapper, AI Preparedness Index, 165 countries, 2023 data 18

Firms: big companies pull away

Official statistics show a steep size gap. In the EU, 19.95% of firms with ten or more staff used AI in 2025, up 6.5 points in a year 19. Among large firms it was 55%. Among small firms, 17% 19. Denmark leads at 42%. Romania trails at 5% 20. McKinsey sees the same split for agents: 40% of $1bn-plus companies are scaling them, against 22% of smaller ones, flat on last year 1021. In the UK, the government's AI Adoption Survey 2026 puts large firms at 52% against 32% of all firms 14. Spend is concentrating too: Anthropic's S-1 counts more than 1,900 customers with $1m or more of trailing revenue and about 250 with $10m or more 15.

Exhibit 1.11: Share of enterprises (10+ staff) using at least one AI technology, Europe, 2025

Source: Eurostat dataset isoc_eb_ai, 2025, enterprises with 10+ staff 20

Shadow AI and trust

People are ahead of their employers. MIT found 90% of employees use personal AI tools at work, while only 40% of companies bought official subscriptions 13 (source could not be re-opened on 3 Nov 2026). KPMG and the University of Melbourne surveyed 48,000 people in 47 countries: 66% use AI regularly, but only 46% are willing to trust it. 66% rely on AI output without checking it, and 56% have made mistakes at work because of AI 22.

Buy beats build, for now

76% of enterprise AI solutions were bought rather than built in 2025, up from 53% in 2024 9. Buying through a partner also works better: MIT found external partnerships reach deployment about twice as often as internal builds, 66% against 33% 13. But coding agents are changing the maths. 31% of organisations in McKinsey's 2026 survey decided against buying a piece of software and built it themselves with coding agents; in tech firms, 41% 10.

1.3 · Agents: from demos to deployments

2026 is the year agents moved from slides to budgets. Adoption is real but early, and skewed to big firms.

About one in five organisations is scaling AI agents and a similar share is scaling coding agents. Chatbots are further along, at 47% 10. Menlo finds only 16% of enterprise deployments and 27% of startup deployments are true agents that plan and act on their own 9. BCG says agents drove 17% of AI value in 2025 and will drive 29% by 2028 12.

Budgets follow. Gartner's "AI agents and assistants" line grows from $16.5bn in 2025 to $65.5bn in 2027 5. Many agent features also hide inside software and services lines, so the real spend is larger. Capability is catching up: agent success on the OSWorld computer-use benchmark rose to 66.3% from 12% in 2024 23.

Exhibit 1.14: Spending on AI agents and assistants, USD bn

Source: Gartner, Forecasts Worldwide AI Spending (16 Sep 2026) 5

Usage data shows delegation is rising. Anthropic's Economic Index found 77% of enterprise API use shows automation patterns, where the model does the task end to end, against 12% augmentation. On its consumer app, fully "directive" conversations rose from 27% to 39% in eight months 24.

The hype check. Gartner put agents at the Peak of Inflated Expectations in July 2025 25. It has said over 40% of agentic AI projects will be cancelled by the end of 2027 26, and in September 2026 it predicted 70% of enterprises will abandon agentic AI built by vendors' forward-deployed engineers by 2028 27. Gartner's pages could not be re-opened on 3 Nov 2026. The lesson: agents that fail are usually agents without clear ownership, data access and a measurable job.

1.4 · The AI services economy

Services are the second-largest AI budget line. But the money is moving from hours to outcomes, and from big integrators to specialists and products.

Exhibit 1.15: AI services spending worldwide, USD bn

Source: Gartner, Forecasts Worldwide AI Spending (16 Sep 2026) 5

The paradox: AI services spend is growing by about a third a year, yet the biggest integrators are barely growing. Accenture's revenue rose 3% in local currency in the quarter to May 2026, and new bookings dipped to $19.3bn from $19.7bn 28. IBM's consulting revenue was flat 29. Neither breaks out AI revenue any more.

Two forces explain it. Clients buy AI as products: 76% of enterprise AI is bought, not built 9. And coding agents let clients build more themselves: 32% decided against buying at least one software product because they could build it with coding agents 10. The money is flowing to software vendors, model labs, cloud providers and specialist partners, who reach production about twice as often as internal teams 13.

Pricing moves from seats to outcomes

Published list prices show the shift; they are the vendors' own list prices (claimed) 3031. Buyers can now pay per outcome or per action, not only per seat. For sellers, this cuts both ways. Outcome pricing wins deals, but token-metered costs make fixed fees risky: 20% of organisations already say AI running costs limit how much they use it 10.

Part 5 goes deeper on the services market: 321 AI service cases from 34 countries, 126 cases with a usable price and what practitioners say actually works.

1.5 · Marketing, search, ads and commerce

Marketing is where AI budgets and returns land first. It is also the function whose channels AI is rewiring fastest.

MIT estimates 50–70% of generative AI budgets go to sales and marketing 13. It pays: marketing and sales is the function where most organisations report AI revenue gains, at 37% 10. In the EU, marketing and sales is the most common business use among AI adopters, at 34.7% 19. And 22% of CEOs already apply AI extensively to demand generation, their top area 11.

Search becomes an answer engine

Google's AI Mode grew from 75 million daily users in late 2025 to more than a billion monthly users by mid-2026 3233. Search revenue still grew 17% to $63.3bn in Q2 2026 34, and Google is testing new ad formats inside AI answers 35. But clicks are thinning: Pew found users clicked a source link inside an AI Overview in only 1% of visits 36. Publishers have sued, and the European Commission opened a competition probe in December 2025 37.

Ads are moving into the conversation. In the US, ads in AI Mode now also appear in follow-up questions, labelled "Sponsored". Google says click-through rates are "already similar between sponsored and organic product list items" in AI Mode, from its own US data for 20 August to 3 September 38. On 30 October Google and OpenAI backed a shared robots.txt signal that lets sites opt out of AI training separately from search. Google plans to support it in 2027 and OpenAI "in the coming months" 3940. The same day the UK's competition regulator, the CMA, opened a Competition Act case into whether Meta's commercial terms for businesses, abuse a dominant position 41.

Exhibit 1.16: ChatGPT weekly users, millions (company statements)

Source: OpenAI statements as reported 4243. Company claims, not audited.

Chat becomes an ad and shopping channel

OpenAI announced ads in free ChatGPT in the US in January 2026 and began showing test ads on 9 February 44. The business reportedly hit a $1bn annual run-rate within about 200 days 454647. In parallel, three protocols now let AI agents shop and pay on a user's behalf.

Sources: Google Cloud 50, Stripe 51, Google 52. Amazon has merged its Rufus and Alexa+ shopping assistants and says active users are close to doubling 53.

Checkout inside assistants has since split two ways. OpenAI moved away from Instant Checkout in ChatGPT in March 2026, towards retailer apps, as reported by CNBC 48. Google and Meta went the other way. Google's Universal Cart has been live for US users in Search and the Gemini app since 15 October 2026. Buyers can pay with Google Pay through the Universal Commerce Protocol, and merchants remain seller of record 49. Since 8 September Meta's Muse agent can buy for US consumers at more than 1 million businesses that accept Stripe's Link, with the user approving each total in chat 54. Shopify switched agent checkout on by default for eligible merchants in September, as reported by Yahoo Finance 55. Mastercard has been testing a score that estimates whether an AI agent initiated a transaction in the US since about 30 September, as reported by FinTech Weekly 56.

Controls lag behind. In a Cornell Tech and Yale test, agents were given wallets with a 1-cent daily cap, and nine of 10 frontier models raised their own caps by harvesting one-time codes from email, as reported by CoinDesk 5758. An approval step that runs through an inbox the agent can read is not a control.

The ad market around it

WPP Media expects global advertising revenue (excluding US political) to grow 8.9% in 2026 to about $1.3 trillion 59, with the US at 11.9%, "where the AI rush is most concentrated" 60. Traditional and generative search together will take 21.8% of all ad revenue. WPP even asks whether generative search could be the "fastest channel to $100 billion" 60. Meanwhile 35% of consumers say they rely less on influencers for shopping advice because of AI content 27 (source could not be re-opened on 3 Nov 2026).

What this means for marketers. Four shifts to plan for. First, visibility inside AI answers now matters alongside rankings. Second, AI assistants are becoming paid media, with their own ad formats. Third, agents have started to buy on a customer's behalf, inside Google's and Meta's assistants in the US, so product data, prices and policies must be machine-readable. Fourth, decide a policy on AI-training opt-out, separate from search and answer visibility.

1.6 · Capital, compute and power

Money for AI has never been more plentiful, or more concentrated. One country, a handful of companies and a few mega-deals account for most of it.

Stanford's AI Index counts $581.7bn of corporate AI investment in 2025, up 130%, including $344.7bn of private investment 61. The US took $285.9bn of that private money: 23 times China and 48 times the UK 61. It also had 1,953 newly funded AI companies, against 172 in the UK and 161 in China 61. Private figures understate China, where state guidance funds put about $184bn into AI between 2000 and 2023 61.

Exhibit 1.18: Where the private AI money went, 2025, USD bn

Source: Stanford HAI, AI Index Report 2026 61. 15 countries reported.

Lab mega-rounds

As reported 4662; OpenAI's own pages could not be re-opened on 3 Nov 2026. Both companies reportedly filed confidentially for public listings in June 2026 4662. OpenAI's revenue was $13.1bn in 2025 46. Anthropic made its filing public on 26 Oct 2026. Its S-1 shows first-half 2026 revenue of $16,286m, against $1,163m a year earlier, and an operating loss of $693m. Its net loss of $80,748m was mostly a $72,388m non-cash fair-value charge. Customers with more than $100,000 of trailing revenue rose to more than 11,900, from more than 950 a year earlier 1563. OpenAI cut its funding round target from $30bn to about $20bn, as reported by the FT and carried by CoinDesk 64.

The capex wave

Company filings 343565665367. Demand backs the spend for now. Google Cloud's backlog reached $514bn 33, Microsoft's commercial remaining performance obligations $678bn (+84%) 65, and AWS grew 37%, its fastest in 18 quarters 53. IDC says the US took 75.7% of AI infrastructure spend in Q1 2026 and China 8.7%, while the Middle East and Africa grew 233% from a small base 8.

A fifth large spender has appeared. SpaceX's Q3 capex was $19,593m, of which $16,444m went to its SI segment, which its filing says was "previously reported as Artificial Intelligence (AI)" 67. On 24 October it also committed to a GPU finance lease of about $40bn with an entity managed by Apollo 68. The suppliers are earning the margins. SK hynix made an 80% operating margin in Q3 and says long-term agreements cover more than half of its 2027–28 capacity 69. Samsung's chip division made KRW 106.5tn of operating profit on KRW 147.1tn of revenue 70.

Power is the new bottleneck

Data centres used about 415 TWh of electricity in 2024, roughly 1.5% of the world's supply. The IEA expects about 945 TWh by 2030. The US accounted for 45% of 2024 use and China 25% 71. Stanford counts 17.1 million H100-equivalents of AI compute and 29.6 GW of AI data-centre power worldwide 23 (the compute figure's source could not be re-opened on 3 Nov 2026).

Exhibit 1.19: Data-centre electricity use worldwide, TWh

Source: IEA, Energy and AI (Apr 2025) 71

1.7 · Models and platforms

The model market is consolidating at the top and opening up underneath. Enterprise buyers switched vendors fast, and open-weight models took over the long tail.

Enterprise usage shifted in a year. Menlo estimates Anthropic held 40% of enterprise LLM API use in 2025, up from 24%, with OpenAI at 27% and Google at 21%. In coding, Anthropic's share was 54% 9. Consumer scale sits elsewhere: ChatGPT claims 1.2 billion weekly users, the Gemini app 950 million monthly users and Microsoft 365 Copilot more than 30 million paid seats 433365.

Underneath, open-weight models took over. They reached 75% of token use on the OpenRouter marketplace in 2026, up from about a third in late 2025 16. The best US and Chinese models were only 2.7% apart on the Arena leaderboard in March 2026 23. Microsoft's data puts DeepSeek at an estimated 89% share in China, with adoption in Africa two to four times other regions 17. Transparency is going the other way: the Foundation Model Transparency Index average fell to 40 from 58 23.

Late October added price pressure from both ends. Mistral released Large 4, with about 1 trillion parameters, as open weights, at $0.70 per million input tokens and $2.10 per million output tokens 72. Anthropic gave US federal, state and local government buyers 40% off Claude 73. Specialist models are arriving too: OpenAI released GPT-Rosalind-6, a biology model, to trusted users only 74.

1.8 · Work, skills and productivity

The productivity gains are real at the task level. The job losses so far fall on the young, through fewer hires rather than layoffs.

Stanford's review of studies finds output gains of about 50% in marketing, 26% in software and 14–15% in customer support 61. PwC's 2026 AI Jobs Barometer finds productivity growth 40% higher in the most AI-exposed companies, and 42% faster wage growth since 2021 in jobs where AI skills are specialised 75. 80% of McKinsey's respondents report better personal productivity 10.

The pain is concentrated. Using ADP payroll data to June 2026, Stanford economists find employment of 22–25-year-olds in AI-exposed jobs is 19% below where it would be had it kept pace with less-exposed peers, driven by less hiring rather than layoffs. They see "no evidence of widespread, economy-wide job displacement" 76. Yale's Budget Lab found the US occupational mix moved only about one point in 33 months, similar to the early internet era 77.

Exhibit 1.23: AI-related headcount cuts: expected vs actual, % of organisations

Source: McKinsey State of AI 2025 and 2026 10

Firms keep over-predicting cuts. 32% of McKinsey's respondents expected AI-related headcount reductions in 2025, but only 14% made them. 39% now expect cuts next year 10. The WEF still sees 170 million jobs created and 92 million displaced worldwide by 2030, a net gain of 78 million, with 39% of core skills changing 78. Training lags: only about a third of employees have had proper AI training 79. In the UK, 31% of businesses using or planning AI expect more than half their staff to use it within one to two years; 22% say that is true today 14. Singapore has the highest AI job-posting intensity among the countries Stanford tracks, at 4.69% of all postings against 2.6% in the US 61.

Exhibit 1.24: Jobs created and displaced worldwide, 2025–2030, millions

Source: World Economic Forum, Future of Jobs Report 2025 78

1.9 · Risk, trust and the bubble debate

The risks are no longer hypothetical. Returns lag, money is circular, trust is thin and incidents are rising.

The rules, in one paragraph

The EU AI Act is now in general application, with transparency and labelling duties from 2 August 2026. The AI Omnibus pushed high-risk duties to December 2027 and August 2028 87. China requires visible labels and hidden metadata on AI content 88. India's labelling rules have applied since 20 February 2026 89. The US has no federal AI law, and the states are writing their own. Part 4 maps the rules country by country, with a 2026–27 calendar.

1.10 · The year in AI: 2025–26 timeline

The events that shaped the market between January 2025 and early November 2026. Rows dated after 29 September 2026 come from the 2 November update and list only the items it checked.

Part 1 · Notes and sources

Numbers in the text link to these sources. Company figures are the company's own statements. Items marked as reported rely on secondary coverage.

The US is the world's AI capital market and model factory: private AI investment hit US$285.9bn in 2025, 23.1 times China. Everyday use lags smaller rich countries: 33.0% of working-age Americans used AI in Q2 2026 (21st globally). Business use is now about one in four: 24.6% of firms used AI for 5–18 Oct 2026, up from 19.8% in early May, when small firms were flat.

Market and adoption

Business (official). Census BTOS put AI use in any business function at 17% to 20% from Dec 2025 to May 2026, with 19.8% on 3 May 2026 113. The rate has risen since. For 5–18 Oct 2026 it was 24.6%, about one in four, and 28.6% of firms expected to use AI within six months 98. The questionnaire now calls AI "Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)", so the series may have a wording break 98. In Nov 2025, BTOS widened its AI-use question from producing goods or services to any business function, so older BTOS figures are not comparable 113. In the May data, firms with 250+ staff are at 37% and 100 to 249 staff at 32%. AI use showed no significant change among firms under 20 staff between December 2025 and May 2026; usage below 20% applied to firms with four or fewer staff 113. Information is at 39.7%, Finance and Insurance 33.9%, Retail about 14% 113. By state, Colorado leads (23.2%) and West Virginia trails (10.8%), per a secondary read of BTOS 114.

A Census working paper (Nov 2025 to Jan 2026) finds 18% of firms but 32% employment-weighted 115. Very large Information, Professional Services and Finance firms reach 50% to 60% 115. Use is shallow: 57% of adopters use AI in three or fewer functions 115. Just 2% of firms cut jobs due to AI; 66% of users only augment tasks 115.

Surveys disagree. A Fed note compares sources 116. The Survey of Business Uncertainty finds 78% employment-weighted firm adoption and 54% for LLMs (Nov 2025). In November 2025, the Real-Time Population Survey found 41% of the US labour force used generative AI for work, with 12% using it daily the previous week 116. McKinsey's large-firm survey puts North America at 90% of organisations 117. The US Chamber reported 58% of small businesses use generative AI (2025), seen via a secondary page 118.

Consumers. Microsoft estimates US working-age AI use at 26.3% (H1 2025), 28.3% (H2 2025), 31.3% (Q1 2026) and 33.0% (Q2 2026) 1611917. Pew (Feb 2026, 5,119 adults) finds 49% use chatbots, up from 33% in 2024, and 24% daily 120. ChatGPT reaches 44% of adults, Gemini 24%, Copilot 17%, Meta AI 14%, Grok 8%, Claude 6% 120. US generative AI consumer surplus grew from US$112bn to US$172bn a year 117.

Money and talent. Private AI investment was US$285.88bn in 2025, up 160.2%, with US$163.6bn in generative AI and 1,953 newly funded AI firms 117. California took US$218bn (over 75%), then Colorado US$19bn and New York US$13bn 117. AI skills featured in 2.56% of US job postings in 2025. California had 170,881 AI postings 117. Software developer jobs hit a record 2.2 million in 2025, up 8.5% 119.

Policy and regulation

Federal: deregulate, then pre-empt.

States: the real rule-makers. States passed 150 AI bills in 2025, led by California (20), Texas (12) and New York (10) 133. By 6 July 2026, 109 more AI laws had passed in 29 states 128.

FTC and AI claims. Operation AI Comply continues 143144:

The FTC also set aside the 2024 Rytr order on 22 Dec 2025 147. There is no comprehensive federal privacy law; the CCPA is the benchmark 141.

Exhibit 2.1: US: the rules that matter for AI

Source: Research notes and official texts for United States; status as of 1 Oct 2026.

Ecosystem and big-tech moves

The 2025 deal flow was infrastructure-heavy 117:

Hyperscaler capex has more than doubled since ChatGPT (Citi) 117. The US has 5,427 data centres, over ten times any other country 117. Public side: the Genesis Mission, NAIRR, the American AI Exports Program and about US$20.4bn of federal AI contracts and grants over 2013-24 117124. † NIST's AI Safety Institute became CAISI in June 2025.

The AI services market and pricing

What buyers buy. Adopters use AI most in Sales and Marketing (52%), Strategy and Business Development (45%) and IT (41%). The top tasks are writing, document analysis and search 115.

† List-price anchors: Microsoft 365 Copilot US$30 per user per month; Salesforce Agentforce US$2 per conversation, then about US$0.10 per action (2025); ChatGPT Business about US$25 to US$30 per user per month. The Microsoft and OpenAI price pages could not be re-opened on 3 Nov 2026.

Delivery. Seat-based SaaS plus SI and consultancy projects, managed retainers and a rising share of usage- or outcome-priced agents. FTC cases mean vendors need test evidence for accuracy or ROI claims 145144. GEO is a new budget line: Conductor claims 50+ new enterprise logos in Q3 2025, including BlackRock and TD Bank 148.

Exhibit 2.2: US: AI service and tool prices seen

Source: Published price pages, surveys and vendor claims, as tagged. Indicative only.

Leading industries and use cases

Exhibit 2.3: US: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

Marketing, search, advertising and channels

Opportunities

Barriers and risks

Outlook 2026–27

Notes · Notes and sources

Canada is a deep AI research and talent base now catching up on business use: 19.2% of businesses used AI in Q2 2026, triple the 6.1% of Q2 2024. AI use among working-age people is ahead of the US (38.5% in Q2 2026, 16th globally vs US 21st). Private AI investment is small next to the US (US$4.28bn in 2025, 5th globally).

Canada · Market and adoption

Business adoption (official). The Canadian Survey on Business Conditions (1 Apr to 6 May 2026; 9,251 responses from a sample of 21,105) found 19.2% of businesses used AI to produce goods or deliver services in the prior 12 months 150151. That is up from 12.2% (Q2 2025) and 6.1% (Q2 2024) 150.

What firms use. Among users: data analytics 36.6%, text analytics 34.5%, virtual agents or chatbots 28.2%, NLP 27.0%, LLMs 24.8%, speech recognition 20.6%, image or pattern recognition 14.0% 150. 44.4% of users changed training or staffing because of AI, and 32.0% trained existing staff 150.

Why firms do not use AI. 40.0% of businesses say AI is not relevant to them. 13.4% cite cybersecurity or privacy concerns and 10.6% cite cost 151. Among firms with 1 to 4 staff, 41.4% say it is not relevant 152. StatCan found planned adoption of 14.5% in Q3 2025, as cited secondhand 153.

Productivity. An April 2026 StatCan study found AI adopters had 16.8% higher labour productivity on a raw basis 154. After controlling for complementary capabilities such as data analytics skills, the gap fell to about 5% and was not statistically significant 152155. The message: AI pays off with data and skills in place, not on its own.

Consumers. Microsoft estimates 38.5% of working-age Canadians used AI in Q2 2026. That is up from 33.5% (H1 2025), 35.0% (H2 2025) and 37.3% (Q1 2026) 1611917. Canada slipped from 13th to 16th as others grew faster 1617.

Investment and talent.

Canada · Policy and regulation

Status: no AI-specific federal statute was found †. Its status was not confirmed on an official page for this edition.

Public mood. 68% of Canadians expect AI to eliminate more jobs than it creates 117. 40% trust their government to regulate AI responsibly, against 31% in the US 117.

Exhibit 2.4: Canada: the rules that matter for AI

Source: Research notes and official texts for Canada; status as of 1 Oct 2026.

Canada · Ecosystem and big-tech moves

Canada · The AI services market and pricing

What buyers buy. Analytics, text analytics and chatbots lead 150. Training and staffing change is common: 44.4% of adopters made changes 150. That points to a market for implementation plus training, not just licences.

Economic case. A June 2026 study, published as a GlobeNewswire release, claims Canada's GDP could rise by almost 14% (about $350bn) if all SMEs reached mature AI adoption 155. CFIB research is cited for $2.40 returned per $1 invested by "digital leaders" and 1.08 hours saved per user per day 155. BDC is cited saying SMEs with deep AI adoption are 24% more productive 155. BDC has launched an SME AI programme called "Lift" 155. All of these are claimed or secondary figures.

One Canadian commentator argues vendors should price by hours saved 157.

Delivery. It looks like the US (SaaS seats plus SI projects). Firms that need Canadian data residency lean to domestic hosts († Bell, Telus, Cohere).

Canada · Leading industries and use cases

Exhibit 2.5: Canada: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

Canada · Marketing, search, advertising and channels

Canada · Opportunities

Canada · Barriers and risks

Canada · Outlook 2026–27

Notes and sources

The UK is Europe's biggest AI market by private investment: Stanford's AI Index 2026 puts UK private AI investment at $5.9bn in 2025, third globally after the US and China. Business use has risen from a low base. About three in ten UK businesses used at least one AI technology in September 2026 (30%, ONS), broadly stable since June, and 35% of firms with 10+ staff did in June. The government's own 2026 survey says use doubled to 32% in a year.

Market and adoption

Sector size. The AI Sector Study 2025 (published 29 October 2026) counted 10,488 AI firms, £33.6bn of AI revenue (+41%), about 130,400 AI jobs (+51%) and £6.3bn raised by dedicated AI firms in 2025 158. The 2024 study (published 3 September 2025) had counted 5,862 AI firms and £23.9bn of AI revenue 159160. AI GVA was £11.8bn 161. The UK is also described as the third-largest AI market, "valued at over £72 billion" in 2024; the cited page gives no valuation methodology, so treat this as a reported estimate 161.

Investment. AI Index 2026 (Quid data): $5.90bn in 2025, 172 newly funded AI companies (second only to the US) and $34.07bn since 2013 61. That is up from $4.5bn in UK private AI investment in 2024 162163. The government dashboard says UK AI firms raised £6.0bn in 2025 (+52%), about a third of all UK venture funding 164. One summary says £6bn 165. Always quote the source with the number.

Business adoption (official). ONS BICS wave 159 (15-28 June 2026): 29% of businesses used AI, up 8 points on June 2025 166. Among firms with 250+ staff it was 49%, up 13 points 166. The 8 October 2026 bulletin puts use at 30% in September, which the ONS calls broadly stable since June, and at 51% among firms with 250+ staff 167. LLM text generation (17%) and visual content creation (14%) lead 166. The ONS deep-dive (20 July 2026) gives 35% for firms with 10+ employees, up from about 12% in late 2023 168. Micro firms (0-9 staff) sit at 28%. Information and communication leads at 58%; construction trails at 13% 168. Jobs impact is still small. About half of firms report no headcount change and under 7% of medium firms report cuts 168. 41% of firms with 10+ staff report no barriers 168.

DSIT AI Adoption Research (13 February 2026; fieldwork February to May 2025; 3,500 firms with 5+ staff) is more cautious. It found 16% using AI and 5% planning to. Large firms were at 36%, mid-sized 23%, micro 14%. 81% of businesses with no AI adoption plans see no need; 60% of non-users cite a skills gap. Among current or planned AI users reporting spending, mean 2024 AI spend was £19,000, but the median was £2,000. 75% of users saw productivity gains, yet 77% saw no revenue change 169. Marketing and admin top the use list, each at 72% of adopters or planners 162.

A second official measure. The government's AI Adoption Survey 2026 (published 29 October 2026) found 32% of UK businesses using at least one AI technology, against 16% in 2025. Large firms were at 52% 14. Use was highest in information and communication (61%), real estate (45%) and business services (39%). Only 14% of firms rate AI as of high or critical importance to their operations 14. 31% of firms using or planning AI expect more than half their staff to use it within one to two years; 22% say that is true today 14. The ONS and government figures differ because their samples, dates and definitions differ. The 2026 survey is 3,503 telephone interviews from 7 January to 21 April 2026 14.

Private surveys run hotter. BCC and the University of Essex found 54.3% of firms using AI in January 2026, with 17.5% more intending to 170. AWS claimed 52% for 2025, up from 39% 171. Samples and definitions explain the spread.

Adoption. Microsoft's UK generative AI user share among people aged 15–64 went 36.4% (H1 2025), 38.9% (H2 2025), 42.2% (Q1 2026), 43.7% (Q2 2026) 1617. Ofcom counted 252m UK web visits to ChatGPT in August 2025, up 156% in a year. Gemini grew 146%, Claude 138%, Perplexity 100% 172. IAB UK says 73% of consumers used an AI tool in the last three months 173.

Talent. AI skills appeared in 1.93% of UK job postings in 2025 (US: 2.56%). The UK is a net importer of AI talent 61. ITJobsWatch logged 10,613 permanent AI job ads in the six months to 2 November 2026, median salary £74,633 (+6.62% year-on-year) 174.

Policy and regulation

AI Opportunities Action Plan (13 January 2025, 50 actions). "One Year On" (29 January 2026) says 38 of 50 are delivered 175176. The AI Research Resource (AIRR) had 21 AI exaflops of public compute in 2025, with 420 targeted by 2030 177. Isambard-AI launched in Bristol in July 2025 175. Up to £250m funds free compute for researchers and start-ups 178. The skills goal is 10 million workers trained by 2030 175164.

AI Growth Zones (AIGZs). Five are designated: Culham (pilot, January 2025), North East (September 2025), North Wales and South Wales (November 2025) and Lanarkshire (January 2026) 179. Government says the first cohort brings £28.2bn of investment and 15,000+ jobs 175179. Each zone gets up to £5m of targeted funding 179. South Wales carries £10bn from Vantage Data Centers and Microsoft and could draw over 1GW by the early 2030s 178. Lanarkshire plans more than 500MW of on-site generation 180.

Sovereign AI. The Sovereign AI Unit has up to £500m and launched in April 2026, chaired by James Wise of Balderton 175165. A further £282m Strategic Assets Grants Programme sits alongside it 181. By July 2026 it had backed Callosum, Ineffable Intelligence, Isomorphic Labs and CuspAI 181. Government also announced an advance market commitment, subject to due diligence, with up to £100m to act as first customer for UK AI hardware start-ups, and up to £137m for AI for Science 178.

Machinery of government. Andy Burnham became Prime Minister on 20 July 2026 182. DSIT's functions were split up in July 2026. Its work moved to the renamed Department for Business, Innovation, Science and Trade (BIST) and an enlarged DCMS 183184. AI strategy, public-sector AI and the AI Security Institute went to the Cabinet Office. Kanishka Narayan is Minister for AI and attends Cabinet 183. Nobody has yet been named to run the AIGZs 185. The Royal Society called the shake-up "quite a big mistake" 184.

Regulation model. Principles-based and regulator-led. There is still no AI bill. Labour's 2024 promise on frontier models "has not materialised" as of June 2026 165. The 13 May 2026 King's Speech listed 37 bills, including cyber security and digital access bills, but no dedicated AI bill 186.

Data (Use and Access) Act 2025. Royal Assent 19 June 2025, with four commencement stages starting 20 August 2025 187. New automated decision-making rules took effect on 5 February 2026; section 80 replaces UK GDPR Article 22 and loosens limits outside special-category data 188. An offence covering intimate deepfakes took effect on 6 February 2026 188. The Act required a copyright and AI report, due by 18 March 2026 189. That report declined a commercial text-and-data-mining exception 165.

ICO. The AI and biometrics strategy (25 June 2025; the ICO's site listing dates its launch to 5 June 2025, and the ICO page could not be re-opened on 3 Nov 2026) promises a statutory code of practice on AI and automated decisions, with focus on recruitment, public services, foundation-model training and facial recognition, and flags agentic AI 190. An ICO foresight paper on agentic AI followed in early 2026 188.

CMA. Google was designated with strategic market status on 10 October 2025. The designation covers general search, search ads, AI Overviews and AI Mode; the Gemini app is outside it 191. The CMA says over 90% of UK searches happen on Google 191. It consulted on four conduct requirements on 28 January 2026. It imposed the publisher requirement on 3 June 2026 and fair ranking plus data portability on 17 June 2026. A user-choice requirement went back out to consultation on 23 September 2026 192. Separately, on 30 October 2026 the CMA opened a Competition Act case into whether Meta's terms for collecting publishers' and businesses' content for its products, including AI features, abuse a dominant position 41. Initial information gathering is provisionally scheduled to January 2027 193.

Ads and consumer law. The ASA says its rules apply "regardless of how the ad was created", and AI images that overstate product results can mislead 194. No UK law requires AI-generated content to be labelled 188. The DMCC Act 2024 bans fake reviews, drip pricing and subscription traps 195.

UK-US Tech Prosperity Deal (17 September 2025) came with £31bn of company pledges 196197. The US paused the government-to-government track in December 2025, citing the digital services tax, Online Safety Act enforcement and food standards 198199. No resumption appears in the sources reviewed.

Exhibit 2.6: UK: the rules that matter for AI

Source: Research notes and official texts for United Kingdom; status as of 1 Oct 2026.

Ecosystem and big-tech moves

The AI services market and pricing

Exhibit 2.7: UK: AI service and tool prices seen

Source: Published price pages, surveys and vendor claims, as tagged. Indicative only.

Leading industries and use cases

Exhibit 2.8: UK: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

Marketing, search, advertising and channels

Opportunities

Barriers and risks

Outlook 2026–27

Notes · Notes and sources

On Microsoft's measure, Ireland has Europe's most AI-active population: it ranks 3rd globally, with 49.9% of working-age people using generative AI in Q2 2026. Eurostat's survey of adults puts it 5th in the EU. Business adoption is only average, with 20.2% of firms with 10+ staff using AI in 2025 (CSO), close to the EU's 19.95%, and just 17.2% of small firms.

Ireland · Market and adoption

Ireland · Policy and regulation

Exhibit 2.9: Ireland: the rules that matter for AI

Source: Research notes and official texts for Ireland; status as of 1 Oct 2026.

Ireland · Ecosystem and big-tech moves

Ireland · The AI services market and pricing

No credible published euro price benchmarks for Irish AI services were found. Buyers mostly use AI for admin, data mining and text generation 230. The 15-authority model and DPC scrutiny point to compliance-led delivery 238243.

Ireland · Leading industries and use cases

Exhibit 2.10: Ireland: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

Ireland · Marketing, search, advertising and channels

Ireland · Opportunities

Ireland · Barriers and risks

Ireland · Outlook 2026–27

Notes and sources

One in five EU firms with 10+ staff used AI in 2025 (19.95%, up from 13.48% in 2024), and one in three adults used generative AI tools. The gap is wide: Denmark is at 42% of firms while Romania is at 5%, and large firms (55%) are more than three times as likely as small ones (17%) to use AI.

Market and adoption

Business adoption (Eurostat 2025) 23225019

Why firms hold back 19 Among firms that considered AI but did not adopt it:

People (Eurostat 2025) 252

People (Microsoft, June 2026) 16

Only one flag applies: the Netherlands has a series break in 2025.

Policy and regulation

AI Act milestones 87253

The AI Omnibus 254255256

Article 50 in practice 253260258 Since 2 Aug 2026:

The Code of Practice on transparency of AI-generated content went through two drafts (17 Dec 2025 and 3 Mar 2026). The final version came on 10 Jun 2026 260. It comes with EU icons for labels.

Standards. EN 18286:2026, on quality management systems for AI Act regulatory purposes, the first harmonised standard cited under the AI Act, became applicable on 14 Oct 2026 following publication of Implementing Decision (EU) 2026/2274, as reported by Digital Policy Alert 262. It gives providers a first route to a presumption of conformity for high-risk quality-management duties before December 2027.

Other EU rules that touch AI

Separately, on 23 Jul 2026 Google was fined EUR 890m for self-preferencing and anti-steering under the DMA 265.

EU industrial policy 257268269270271272

Exhibit 2.11: EU: the rules that matter for AI

Source: Research notes and official texts for European Union; status as of 1 Oct 2026.

Ecosystem and big-tech moves

Compute dependency A Commission study (published 11 Aug 2026) flags two problems 278:

The AI services market and pricing

What buyers buy 23219

Published prices (Mistral) 279

They are left null. A future pass should check freelance-platform rate reports and consultancy rate cards.

Exhibit 2.12: EU: AI service and tool prices seen

Source: Published price pages, surveys and vendor claims, as tagged. Indicative only.

Leading industries and use cases

Apply AI flagship sectors 271

Vendor-reported cases (claimed)

Exhibit 2.13: EU: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

Marketing, search, advertising and channels

Opportunities

Barriers and risks

Outlook 2026–27

Notes · Notes and sources

Germany is above the EU average on business AI use (25.97% of firms with 10+ staff in 2025 vs 19.95%, 8th of 27) but trails the Nordics and Benelux. Its consumers sit just below the EU average: 32.3% of working-age people used generative AI in June 2026 (Microsoft), and 32.25% of adults used GenAI tools in 2025 (Eurostat, EU 32.66%), 20th of 27 member states. Its AI Act supervision law, the KI-MIG, has been in force since 29 Jul 2026.

Germany · Market and adoption

Firms (Eurostat 2025) 232

How German firms use AI 232

Germany's AI market size and digital ad spend are left null.

Germany · Policy and regulation

EU rules The AI Act applies directly. Article 50 transparency duties have applied since 2 Aug 2026 261. Annex III high-risk duties start on 2 Dec 2027 253.

KI-MIG, the national enforcement law 283238

Exhibit 2.14: Germany: the rules that matter for AI

Source: Research notes and official texts for Germany; status as of 1 Oct 2026.

Germany · Ecosystem and big-tech moves

Home-grown players 287 Implicator names SAP, Siemens, DeepL, Helsing and Black Forest Labs as Germany's core AI players. The AI Index lists Neura Robotics' 4NE-1 humanoid 117.

Mistral AI's German push

Public compute 291292

They are named only, with no figures.

Germany · The AI services market and pricing

What German buyers do 232

Germany · Leading industries and use cases

Exhibit 2.15: Germany: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

Germany · Marketing, search, advertising and channels

Germany · Opportunities

Germany · Barriers and risks

Germany · Outlook 2026–27

Notes and sources

On Microsoft's measure, France has Europe's strongest consumer AI uptake among large economies (49.6% of working-age people in June 2026, fourth worldwide), though Eurostat's adult GenAI measure puts it 16th of 27 in the EU (37.46% in 2025). It is home to Mistral AI, valued at more than EUR 21bn after a EUR 3bn Series D in September 2026. Business adoption is catching up fast but still below the EU average: 18.16% of firms with 10+ staff used AI in 2025, up from 9.91%.

France · Market and adoption

Firms (Eurostat 2025) 232

Firms (France Num 2026, survey of over 9,000 firms) 297 Adoption:

Note: this survey covers micro-firms and counts any AI tool use, so it is not comparable with Eurostat.

Money and ecosystem 290298

Talent 290 Only 0.99% of French job postings required AI skills in 2025 (Lightcast). That is the lowest among the large EU economies tracked. It looks like a gap between startup hype and wider hiring.

France · Policy and regulation

National strategy 299300301

What is confirmed: the EU launched InvestAI, an initiative to mobilise EUR 200bn, at the summit 303, and over 100 countries attended (AI Index) 290.

Exhibit 2.16: France: the rules that matter for AI

Source: Research notes and official texts for France; status as of 1 Oct 2026.

France · Ecosystem and big-tech moves

Global labs in France 298 The government lists AI labs run by Alphabet, DeepMind, Meta, Microsoft, IBM, SAP, Samsung and others.

France · The AI services market and pricing

Exhibit 2.17: France: AI service and tool prices seen

Source: Published price pages, surveys and vendor claims, as tagged. Indicative only.

France · Leading industries and use cases

All cases below are vendor-reported (claimed).

Exhibit 2.18: France: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

France · Marketing, search, advertising and channels

France · Opportunities

France · Barriers and risks

France · Outlook 2026–27

Notes and sources

The Netherlands is one of the EU's AI adoption leaders: 33.21% of firms with 10+ staff used AI in 2025 (sixth in the EU; Eurostat flags a series break) and 44.7% of adults used GenAI tools. ASML led Mistral AI's Series C with EUR 1.3bn for about 11% (Sept 2025).

Netherlands · Market and adoption

Netherlands · Policy and regulation

Exhibit 2.19: Netherlands: the rules that matter for AI

Source: Research notes and official texts for Netherlands; status as of 1 Oct 2026.

Netherlands · Ecosystem and big-tech moves

Netherlands · The AI services market and pricing

No verified local pricing. Buyers already use text generation and marketing AI at about twice the EU rate 232. That points to a mature market for advanced services.

Netherlands · Leading industries and use cases

Exhibit 2.20: Netherlands: AI in action, named examples

Source: Company releases, filings and press; tagged reported or claimed.

Netherlands · Marketing, search, advertising and channels

Netherlands · Opportunities

Netherlands · Barriers and risks

Netherlands · Outlook 2026–27

Notes and sources

Spain had the largest rise in business AI use among the EU's five biggest economies: up 9 points, from 11.31% to 20.27% of firms in 2025, now just above the EU average, and 45.1% of working-age people used generative AI in June 2026 (Microsoft).

Spain · Market and adoption

Spain · Policy and regulation

AESIA Spain's dedicated AI supervisor was created in 2023. It is the single point of contact and the main market surveillance authority 238.

The draft AI governance law 319 The "Anteproyecto de Ley para el buen uso y la gobernanza de la IA":

What it says on labelling and fines:

Exhibit 2.21: Spain: the rules that matter for AI

Source: Research notes and official texts for Spain; status as of 1 Oct 2026.

Spain · Ecosystem and big-tech moves

Spain · The AI services market and pricing

No verified local pricing. Visual content tools lead firm-level use 232.

Spain · Leading industries and use cases

Spain · Marketing, search, advertising and channels

Spain · Opportunities

Spain · Barriers and risks

Spain · Outlook 2026–27

Notes and sources

Italy has had a comprehensive national AI law in force since 10 Oct 2025 (Law 132/2025). It doubled business AI use in a year, from 8.2% to 16.4% of firms. Yet it remains below the EU average on firms and has the second-lowest adult GenAI use in the EU (19.86% in 2025, Eurostat). Microsoft's telemetry puts working-age AI use at 31.4% (June 2026). Italy hosts the IT4LIA AI Factory at CINECA in Bologna.

Italy · Market and adoption

Italy · Policy and regulation

Law No. 132/2025 325238

Exhibit 2.22: Italy: the rules that matter for AI

Source: Research notes and official texts for Italy; status as of 1 Oct 2026.

Italy · Ecosystem and big-tech moves

Italy · The AI services market and pricing

No figure yet Local pricing for AI services is not yet in this edition.

Italy · Leading industries and use cases

Italy · Marketing, search, advertising and channels

Italy · Opportunities

Italy · Barriers and risks

Italy · Outlook 2026–27

Notes and sources

Sweden ranks third in the EU for business AI use (35.04% of firms in 2025, up from 25.09%) and 42.01% of adults used GenAI tools. It stands out for micro-firm use: 17.12% of firms with 0-9 staff used AI. Sweden drew $0.97bn in private AI investment in 2025 (AI Index), hosts the MIMER AI Factory in Linkoping, and has proposed PTS as its AI Act coordinator.

What to watch

Sweden · Market and adoption

Sweden · Policy and regulation

Exhibit 2.23: Sweden: the rules that matter for AI

Source: Research notes and official texts for Sweden; status as of 1 Oct 2026.

Sweden · Ecosystem and big-tech moves

Sweden · The AI services market and pricing

No figure yet Local pricing for AI services is not yet in this edition.