What will tell us.
An index is only as good as what it measures. The Globalance Intelligence Index therefore consists not of judgements but of 37 indicators: quantities that someone counts, that have an open source and that can be updated monthly.
Each indicator answers one part of one of the seven dimensions. How capable is the machine? How much labour does it replace? Where does capital flow? No indicator without a question, no question without a number.
Measurable figures with open sources: the share of open AI models, the number of programming roles, the capital share of the largest companies, the waiting time for a therapy place. Updated monthly.
Each indicator has a current value and a mark for 2040. Its tile shows where it stands between the two: the fill level shows the distance travelled, the arrow shows the direction since last month. The indicators form each dimension; the dimensions form the index.
Below are the 37 indicators, ordered by the seven dimensions. Each tile shows the value, benchmark and source. Every calculation is explained under «Method».
Fill level = progress from 0 (baseline 2020) to 100 (target 2040) · already visible · in development · dormant
AI Capability
Labor Substitution
Capital Concentration
Policy Response
Geopolitical AI
Path Divergence
Social Adaptation
The research institute METR measures how long a task may be for an AI agent to complete it independently in half of all cases, measured against the time a professional needs. In 2020 that was about 9 seconds; today the best model manages about 17 hours.
Since 2023, the value has doubled roughly every four months. The scale is therefore logarithmic: each doubling counts equally. A score of 100 points corresponds to one working year (2,000 hours).
From about one working month (167 hours, roughly 80 points), AI completes entire projects on its own. That is the turning point from tool to colleague.
Source: METR, Time Horizons — 50% time horizon of the best measured model · As at 9 October 2026 · metr.org/time-horizons
See the indicator page →The ARC test measures whether a machine can genuinely learn something new, not what it has memorised.
The third edition (since March 2026) places AI systems into unfamiliar game worlds without instructions: they must discover the rules themselves and reach the goal as efficiently as a human. Humans solve virtually all tasks. The best model achieves 62.7 per cent, the second-best 30. The gap between the two is as revealing as the gap to humans. The earlier tests are saturated: ARC-AGI-1 at 98 per cent, ARC-AGI-2 at 95. The indicator therefore switches by a fixed rule: it always measures the most recent test that machines have not yet mastered, and switches as soon as the best verified score exceeds 85 per cent. Each switch is marked in the chart.
If the best model also approaches the 85 per cent mark on ARC-AGI-3, machines will be learning in unfamiliar environments almost as efficiently as humans. That would be the strongest single signal for general AI available today.
Source: ARC Prize Foundation — ARC-AGI-3 · As of 26 September 2026 · arcprize.org/arc-agi/3
See the indicator page →How much of the AI use in US companies runs on open models rather than the closed offerings of the major providers?
According to Menlo Ventures' annual survey, the figure is 11 per cent, down from 19 per cent a year earlier. For productive use, firms rely on the closed frontier models. The smaller this share, the more power concentrates in the hands of a few companies.
A share above 40% for open models would mean many providers shape the market. Below 20%, a handful of large technology groups dominate.
Source: Menlo Ventures, «2025: The State of Generative AI in the Enterprise» (survey, 495 US decision-makers) · As of 9 October 2026 · menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise
See the indicator page →The metric captures the share of programming contributions in the most widely used open-source packages where an AI tool is openly credited as co-author.
This is a lower bound: AI assistance without attribution remains invisible. In July 2026 the share stood at 5.3 per cent, eleven months earlier at 0.5 per cent: more than tenfold in under a year.
When AI is openly credited as a contributor in more than half of all submissions, the machine writes most of the software. Humans only review.
Source: Andrew Nesbitt, «A year of AI disclosure in critical packages» (CHAOSS method, 5,682 repositories) · As of 9 October 2026 · nesbitt.io/2026/08/06/a-year-of-ai-disclosure-in-critical-packages.html
See the indicator page →By 2040 the business will settle: AI computing power will be sold like electricity or water — reliably, but without boom-era growth rates.
Around 30 per cent annual growth would still be very high for an industry of this size.
If growth falls abruptly below 20 per cent, the boom tips into disillusionment. If it settles steadily around 30 per cent, the AI cloud has matured into a utility economy. That is the expected state by 2040.
Source: quarterly reports Microsoft (Azure), Alphabet (Google Cloud), Amazon (AWS); weighted average revenue growth · As at 9 October 2026
See the indicator page →NVIDIA supplies the chips on which almost all AI runs.
The growth of its data centre business is the thermometer for the entire AI build-out. Today it is still growing at boom rates. By 2040 the supercycle normalises: the target is sustainable growth of 25 per cent per year — intact expansion without a bubble.
An abrupt collapse below 10 per cent would be the bubble signal. An orderly normalisation towards 25 per cent would be the healthy maturity path into the 2040 state.
Source: NVIDIA Quarterly Report / Bloomberg · As of 23 September 2026 · investor.nvidia.com/financial-info/quarterly-results
See the indicator page →Job openings for programmers in the USA are falling sharply — precisely in the occupation that built AI.
Experienced developers and architects remain in demand; entry-level and routine roles disappear first. This offers a preview of what other knowledge professions may face.
A further decline of more than 20 per cent within twelve months would confirm that the world of work for developers is being radically restructured.
Source: Indeed Hiring Lab — Job Postings Index Software Development, Feb 2020 = 100 (FRED IHLIDXUSTPSOFTDEVE); 30-day average minus 100 · As at 30 October 2026 · fred.stlouisfed.org/series/IHLIDXUSTPSOFTDEVE
See the indicator page →How many workers in the EU have their tasks assigned automatically by software?
According to the European Commission's large-scale survey, it is roughly one in five. Algorithmic control is most prevalent among industrial operatives and least prevalent among ancillary workers, senior managers and highly qualified professionals. A sharp rise would signal a shift in power from people to software.
When more than half of all workers are scheduled by algorithms, software has taken command of the workplace.
Source: EU Commission JRC, AIM-WORK survey (EU-27, 2024/25); successor to Eurofound EWCS · As of 9 October 2026 · publications.jrc.ec.europa.eu/repository/bitstream/JRC144330/JRC144330_01.pdf
See the indicator page →How much the US economy produces per hour worked, as measured by official statistics.
Since 2020 productivity has risen by 10 percent, just under 2 percent a year. From 2007 to 2019 the average was 1.5 percent. The 2040 mark of 160 requires around 2.4 percent a year. Only when the curve gets that steep does AI break through across the whole economy.
If productivity grows by more than 2.5 per cent a year for several consecutive years, AI is making the economy measurably more productive.
Source: U.S. Bureau of Labor Statistics, Productivity and Costs — output per hour worked, non-farm business sector (FRED OPHNFB), rebased to annual average 2020 = 100 · As of 9 October 2026 · fred.stlouisfed.org/series/OPHNFB
See the indicator page →How many industrialised countries are shaping the transformation of work by law, through shorter working hours or an AI levy?
Since 2020, three countries have done so, all in Latin America: Chile, Colombia and Mexico. No OECD country has yet introduced an AI levy. Most states are observing rather than acting. If the labour-market shock arrives, it will hit largely unprepared systems.
More than five countries with binding measures in place by end-2027 would show that policy is shaping the transition rather than trailing behind it.
Source: Own count under a fixed rule (national legislative sources); list in the «zaehlliste» field · As at 10 October 2026 · remofirst.com/compliance-news/colombia-workweek
See the indicator page →The count covers announced job cuts at US companies that were explicitly attributed to AI, cumulated since 2023.
The number is growing quickly, but captures only part of the picture. Many positions disappear quietly, through natural attrition and hiring freezes. The indicator shows when companies openly name the restructuring.
Half a million job cuts explicitly attributed to AI would be the open admission: the large-scale workforce restructuring is under way. The 2040 milestone stands at two million.
Source: Challenger, Gray & Christmas, Job Cut Report (cumulated since 2023) · As of 9 October 2026 · challengergray.com/blog/job-cuts-fall-in-september-hiring-plans-up-3-over-2025-on-weak-early-seasonal-hiring
See the indicator page →The count covers job openings in office and knowledge occupations in the USA, the most direct pulse of the labour market.
The figure shows the change against the 2020 annual average: it is declining, but not yet dramatically. A sustained, sharp fall would be the first clear sign that AI is replacing office jobs at scale.
If job openings fall by more than 25 per cent in two consecutive quarters, AI will no longer be replacing office work selectively — it will be doing so across the board.
Source: BLS JOLTS — Professional & Business Services, job openings (FRED JTS540099JOL); change against 2020 annual average · As at 10 October 2026 · fred.stlouisfed.org/series/JTS540099JOL
See the indicator page →It measures how much of companies' use of AI language models goes to the three largest providers: Anthropic, OpenAI and Google.
At the end of 2025 it was 88 percent, up from 69 percent in 2023. A small group of providers therefore earns most from the AI boom.
If more than 90 percent goes to three providers for good, a few companies set the prices and rules of AI in the economy.
Source: Menlo Ventures, State of Generative AI in the Enterprise (9 Dec 2025) and Mid-Year LLM Market Update (July 2025): shares of language-model usage via API in enterprises · As at 21 October 2026 · financialcontent.com/article/gnwcq-2025-12-9-menlo-ventures-2025-state-of-generative-ai-report-enterprise-investment-hit-37b-in-2025-tripling-in-one-year
See the indicator page →What share of corporate profits in the United States do seven groups account for?
The calculation uses operating profit, so that book gains on holdings such as Anthropic or SpaceX do not distort the picture. Since 2020, the share has nearly doubled, to just under one fifth. The higher it rises, the more the value of AI concentrates in the hands of its owners.
If seven companies earn a quarter of all US corporate profits, their concentration becomes a risk for investors and policymakers.
Source: SEC filings of the seven groups (operating profit, last four quarters) relative to profits of all US companies before tax (BEA, FRED A053RC1Q027SBEA) · As of 9 October 2026 · fred.stlouisfed.org/series/A053RC1Q027SBEA
See the indicator page →The four large cloud companies together invest over $700bn a year in data centres, more than most countries spend in total.
This capital expenditure drives the entire AI boom. If several companies cut back simultaneously, the cycle turns.
If two or more large groups cut their investment plans by more than 10 per cent, the cycle turns. Suppliers, chipmakers, and construction firms feel it first.
Source: Capital expenditure guidance from the quarterly reports of Amazon, Alphabet, Meta and Microsoft · As at 10 October 2026 · cfodive.com/news/microsoft-holds-line-ai-spending-plans/826648
See the indicator page →How many companies have moved beyond pilots and are seeing a real impact on profits?
In the McKinsey survey, well over one in three companies attributes at least part of its operating profit to AI. Nearly all are experimenting. But the path from pilot project to real benefit is slow and takes years.
Once 35% of firms show measurable AI success, the benefit reaches the broader economy, not only technology groups.
Source: McKinsey Global Survey on AI / Gartner Hype Cycle · As of 9 October 2026 · mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
See the indicator page →Are small and medium-sized enterprises also using AI productively, or only large companies?
So far only about one in six SMEs manages to do so: expertise and budget are lacking. If that persists, a lasting gap opens between large and small businesses.
SME adoption above 25 per cent would mean that small and medium-sized enterprises are catching up; below 10 per cent, the lead of the giants becomes entrenched.
Source: OECD Digital Economy Outlook / Eurostat · As at 23 September 2026 · oecd.org/en/publications/oecd-skills-outlook-2025_26163cd3-en
See the indicator page →Total factor productivity measures whether the economy as a whole is becoming more efficient.
This is the hardest test of the AI promise. So far, the results are sobering: despite billions in investment, growth remains below average. Either the effect still needs time, or the investment is coming to nothing.
Growth of more than 1.5% a year would mean AI is making the whole economy measurably more efficient, not just individual groups. Below 0.5%, the benefit fails to materialise while investment continues. At that point, the high valuations of technology stocks become a risk.
Source: BLS Multifactor Productivity / FRED · As at 9 October 2026 · bls.gov/productivity
See the indicator page →How many industrialised countries require AI providers by law to be transparent, for example by labelling AI-generated content?
The number is growing, but many rules remain toothless. Capital markets and reputational risk have so far proved the more effective checks.
More than 15 countries adopting binding commitments would establish a global regulatory standard.
Source: OECD AI Policy Observatory · As of 26 September 2026 · oecd.ai/en
See the indicator page →The US fiscal deficit shows how much room the world's most important state still has to cushion the transition, for example through retraining or social security.
The higher the deficit, the harder it becomes to finance large-scale responses to the AI disruption.
If the deficit remains durably above 7 per cent of economic output, Washington will lack the fiscal resources for large-scale responses: retraining programmes, welfare benefits, and fiscal stimulus will become a question of debt-servicing costs, not political will.
Source: Congressional Budget Office, Monthly Budget Review (deficit FY2026); share as % of GDP per American Action Forum · As at 10 October 2026 · cbo.gov/publication/61985
See the indicator page →How many states govern AI bilaterally rather than collectively?
The count includes signed bilateral agreements between two governments whose main subject is AI, such as the UK–US agreement on AI safety from 2024. By end-2020, there were four; today there are at least twelve. Many bilateral agreements instead of shared rules means the AI world is organising itself into blocs.
More than 30 new bilateral agreements in 18 months would mean that everyone is negotiating with everyone else, and no one is setting common rules.
Source: Own count applying a fixed rule: signed agreements between two governments with AI as the main subject (sources: government communications, Digital Policy Alert); list in the «zaehlliste» field · As at 10 October 2026 · gov.uk/government/news/uk-united-states-announce-partnership-on-science-of-ai-safety
See the indicator page →What share of global cloud infrastructure do the three large US providers, Amazon, Microsoft and Google, hold?
In early 2026 it was around 63 percent. America remains the powerhouse of AI. At the same time China and India are building their own capacity: a two-way contest is emerging, not a monopoly.
If the US share falls towards 60 per cent, the AI landscape will split durably into an American and an Asian technology sphere. If it rises back above 70 per cent, the United States will largely control the world's computing power single-handedly.
Source: Synergy Research / srgresearch.com · As of 13 August 2026 · srgresearch.com/research/hyperscale-market-tracker
See the indicator page →China controls rare earth elements, the raw materials without which neither chips nor electric motors can be made.
The count tracks how many of these materials Beijing has placed under export restrictions. Each expansion is a geopolitical pressure tool aimed at the West.
With more than 25 regulated materials, Beijing gains the ability to hit the West precisely where no alternative source exists.
Source: Chinese Ministry of Commerce / mofcom.gov.cn · As at 21 September 2026 · cset.georgetown.edu/publication/mofcom-notice-2025-61
See the indicator page →It measures the capacity of US data centres that are actually in operation.
At the end of 2025 it was more than 50 gigawatts, roughly what fifty large nuclear power plants supply. If this figure keeps growing at double-digit rates, computing power becomes basic infrastructure, as railways and the power grid once did. If the build-out stalls, that is the earliest warning of an investment slump.
Growth above 25 per cent per year sustains the build-out; below 10 per cent would be the first early indicator of an investment slowdown.
Source: FERC State of the Markets 2025 (data: Yes Energy) · As of 9 October 2026 · ferc.gov/sites/default/files/2026-03/26_State-of-the-Market_0324_1430.pdf
See the indicator page →Data centres consume about 4.4% of total US electricity.
The 2040 target is 15%: data centres would then be the country's largest single industrial consumer. Electricity is therefore becoming the scarcest raw material of the AI age. New power plants and grids are being built more slowly than demand is growing.
Above 6 per cent of US electricity consumption confirms the bottleneck: it is not chips but power that then constrains AI growth.
Source: EIA / IEA Data Center Energy Tracker · As at 13 August 2026 · eia.gov/todayinenergy/detail.php?id=67344
See the indicator page →Where are the machines used to manufacture chips going?
The measure is the share of global equipment purchases made outside mainland China. It has fallen since 2020, because China expanded its purchases sharply. Since autumn 2025 it has been rising again: in the second quarter of 2026 it stood at just over 70 per cent.
If more than 80 per cent of equipment consistently flows to countries outside China, the chip world is split into two camps.
Source: SEMI, Worldwide Semiconductor Equipment Market Statistics (quarterly revenues); China figures from media reports on the SEMI table (The Star, TelecomLead, Evertiq) · As of 9 October 2026 · semi.org/en/semi-press-release/global-semiconductor-equipment-billings-increased-23-percent-year-over-year-in-q2-2026-semi-reports
See the indicator page →The share of defence spending in US economic output shows how firmly America is preparing for confrontation.
AI chips and data centres are increasingly treated as defence goods. Technology and security are merging.
A sustained level above 3.5 per cent signals that Washington treats AI chips as defence goods, subject to export bans for the rest of the world.
Source: US Congressional Budget Office / cbo.gov · As at 9 October 2026 · cbo.gov/publication/61307
See the indicator page →The yield on ten-year US Treasuries sets the price of money for the entire world.
The higher it rises, the more expensive the vast AI investments become, as they were priced for cheap money. It currently sits well above the levels seen during the zero-interest-rate years.
If the rate stays above 5.5 per cent for three months, money becomes persistently expensive: financing deals fall through and investment stalls.
Source: FRED / US Treasury · As at 30 October 2026 · fred.stlouisfed.org/series/DGS10
See the indicator page →The real interest rate shows what money truly costs after adjusting for inflation.
It determines whether risky future investments pay off and where large-scale capital is deployed. It is unusually high at present: money is worth something again.
Below 1.5%, money is so cheap that even hopeless projects get funded: fertile ground for speculative bubbles. Above 2.5%, capital becomes costly. Companies must justify their future projects, and high valuations come under pressure.
Source: FRED / US Treasury · As at 30 October 2026 · fred.stlouisfed.org/series/DFII10
See the indicator page →What share of global cloud-computing power does the West account for?
The lead is shrinking slowly, as China and Asia are building capacity at scale. Two separate technology spaces with their own standards are taking shape, and investors need to find their way in both.
If the Western share falls below 65 per cent, the split of the AI sphere into two systems is complete. The 2040 projection points to near parity at 55 per cent.
Source: Synergy Research · As at 13 August 2026 · srgresearch.com/research/hyperscale-market-tracker
See the indicator page →China invests tens of billions of dollars annually in AI through state-directed programmes, a multiple of the public programmes in the West.
The state allocates capital by plan rather than by market signals: faster to scale, but more exposed to costly misallocation.
A rise towards $150bn per year by 2040 would definitively establish China's state-directed model as the West's principal rival model.
Source: China Govt Data + CSIS · As at 23 September 2026 · csis.org/analysis/chinas-pursuit-defense-technologies-implications-us-and-multilateral-export-control-and
See the indicator page →For many investors, money market funds are the alternative to a bank account: they pay market rates and can be accessed almost as quickly as a current account.
When their assets grow faster than bank deposits, money flows around the banking system. Tokenised money market funds and stablecoins can accelerate that shift. In 2020, US money market fund assets equalled 31.9 per cent of bank deposits; today they stand at 43.6 per cent. The 2040 target is parity: as much money in money market funds as in bank accounts.
Once money market funds reach half the level of bank deposits, the outflow becomes a cost question for banks: they must offer noticeably higher interest rates to retain deposits.
Source: Federal Reserve: assets of US money market funds (Financial Accounts Z.1, FRED MMMFFAQ027S) as a percentage of deposits at all US commercial banks (H.8, FRED DPSACBM027SBOG), each at quarter-end · As at 25 October 2026 · fred.stlouisfed.org/series/MMMFFAQ027S
See the indicator page →This value shows what inflation investors expect for the period five to ten years from now.
If it stays calm, markets trust that the central bank has prices under control.
A sustained level above 2.8 per cent would signal that markets are losing confidence in price stability.
Source: FRED / Federal Reserve · As at 30 October 2026 · fred.stlouisfed.org/series/T5YIFR
See the indicator page →Stablecoins are private digital money, mostly pegged to the dollar.
Their volume has multiplied since 2020 and hovers around the 300bn mark. The larger they grow, the more monetary flows bypass banks and central banks.
Above $300bn, a growing share of payments bypasses banks and central banks altogether. A 2040 benchmark of $3,000bn would render private digital money unambiguously systemically important.
Source: DeFiLlama / CoinGecko · As at 9 October 2026 · defillama.com/stablecoins
See the indicator page →Nearly half of all US bank deposits are not covered by government insurance.
In a crisis of confidence, such funds are the first to flee — today via smartphone within hours. The higher the share, the more vulnerable the banking system becomes to a digital bank run.
Above 45%, the system becomes vulnerable: should a bank lose the confidence of its customers, those customers can today withdraw their money within hours using a mobile phone. That threshold has already been exceeded.
Source: FDIC BankFind (Quarterly Banking Profile): estimated uninsured deposits divided by domestic deposits, all institutions · As at 23 October 2026 · fdic.gov/quarterly-banking-profile/quarterly-banking-profile-second-quarter-2026.pdf
See the indicator page →It measures how many adults in the United Kingdom say they felt highly anxious the previous day, in the national statistics office's large survey.
The share is above its pre-pandemic level. If it keeps rising, the change is reaching everyday life, not just the economy.
If more than 30 percent of adults feel highly anxious for good, the strain is higher than at the peak of the pandemic.
Source: Office for National Statistics, Quarterly personal well-being estimates (Annual Population Survey): anxiety score 6 to 10 on a scale of 0 to 10 · As of 9 October 2026 · ons.gov.uk/peoplepopulationandcommunity/wellbeing/bulletins/measuringprogresswellbeingandbeyondgdpintheuk/june2026
See the indicator page →How many adults in the EU took part in education or training in the last four weeks? In 2025 the share was 13.7 per cent, compared with 10.8 per cent in 2019.
This is progress, but slow — too slow for an upheaval that is reshaping entire professions. Education is the most important protection against the hardships of change.
Only when one in four adults is continuously upskilling — the level the Netherlands and Switzerland already achieve today — will education keep pace with AI.
Source: Eurostat, Labour Force Survey (trng_lfse_01): participation in education and training in the last 4 weeks, 25–64 years, EU-27 · As of 9 October 2026 · ec.europa.eu/eurostat/databrowser/view/trng_lfse_01/default/table
See the indicator page →