Capabilities that are hard to replace
ASML builds the EUV lithography machines in the Netherlands. ZEISS makes the precision optical system in Germany. imec trials manufacturing processes in Belgium alongside industry.
Lateral Space Research · 2026
Europe in the Age of AI
Intelligence, made real.
Where Europe really stands in the artificial intelligence value chain: from electricity and semiconductors through to models, applications and the processes in which the technology produces economic results.
Download data and sources CSV + register · ZIP
Photo © visit.brussels / Jean-Paul Remy
The question
Where does Europe really stand?
An essential part of the most advanced chips is manufactured using machines built in the Netherlands. Yet most of the AI-specialised chips, the cloud and the frontier models used by European firms are designed or controlled outside the continent.
The report distinguishes the steps where Europe holds technologies that are hard to replace, those where it competes, and those where it depends on a handful of external suppliers.
The European position
Europe's barbell
Hard-to-replace technologies upstream, a large industrial economy downstream, dependencies in the middle steps.
Scroll the figure sideways
Europe's AI barbell
The figure summarises different markets and does not represent an aggregate European share of the AI economy.
Source Lateral Space, from the sources listed in the report (S14–S18, S17–S21, S27–S40, S41–S42, S72–S74)
Show the data
| Segment | European position | Examples |
|---|---|---|
| Upstream deep tech | Strategic assets | ASML; ZEISS SMT; imec; photonics; power semiconductors; electrical equipment |
| Hollow middle | Structural dependency | AI accelerators; advanced fabrication; high-bandwidth memory; global cloud; generalist models; growth capital |
| Downstream economy | Large application base | Industry; energy; health; engineering; services; public sector |
CSV: data/core/00_europes_ai_barbell.csv (lists expanded to the report's own six dependencies / three upstream items — the report text takes precedence over the CSV) · Download data and sources
The value chain
The value chain, in four steps
Follow AI from the machines that print the chips to the moment it changes how a company works. One number for each step.
-
01
Components
72%
of global specialist foundry revenue goes to a single company: TSMC.
Europe builds the machines that make the chips, but not the chips.
TrendForce / Counterpoint · Q1 2026
-
02
Infrastructure
15%
of Europe's own cloud market is served by European providers.
It was 29% in 2017. Three US groups hold about 70%.
Synergy Research Group · 2024
-
03
Models and capital
$285.9bn
of private AI investment went to the United States in 2025.
The three largest European markets together raised $14.15bn.
Stanford AI Index · 2025
-
04
Adoption
20%
of European enterprises used at least one AI technology in 2025.
It was 8% in 2023. Large firms 55%, SMEs 19%.
Eurostat · 2025
The four steps measure different markets and populations: the values cannot be summed.
See all the figures behind the four steps
Scroll the figure sideways
The value chain in four steps
The four steps measure different markets and populations: the values cannot be summed.
Source TrendForce / Counterpoint; European Commission; Synergy Research Group; IEA; Stanford AI Index; Eurostat
Show the data
| Step | Figure | What it measures | Source |
|---|---|---|---|
| Components | 72% | Share of global specialist foundry revenue attributed to TSMC, Q1 2026 | TrendForce / Counterpoint |
| Components | 9% | EU share of the world semiconductor market, 2025 | European Commission |
| Infrastructure | 15% | Share of the European cloud infrastructure services market held by European providers, 2024 | Synergy Research Group |
| Infrastructure | 415 → 945 TWh | Global data centre electricity: 2024 estimate and 2030 base scenario | International Energy Agency |
| Models and capital | $285.9bn / $12.4bn | Private AI investment in 2025, United States and China | Stanford AI Index |
| Models and capital | $5.9 / 4.36 / 3.89bn | Private AI investment in 2025: United Kingdom, France, Germany | Stanford AI Index |
| Adoption | 20% | European enterprises using at least one AI technology in 2025 (8% in 2023) | Eurostat |
| Adoption | 42% → 5.2% | Distance between Denmark and Romania, the two extremes recorded in the EU | Eurostat |
CSV: data/core/ (four datasets, see DATA_DICTIONARY.csv) · Download data and sources
Chapters
Six chapters to explore
The report has 29 chapters. These six are the fastest way in.
Without available, stable electric power there is no computing at scale.
01
Energy and data centres
No latest-generation chip exists without passing through this chain.
02
Semiconductors and chips
Three groups control roughly two thirds of the world cloud market.
03
Cloud and computing capacity
Europe trains some of the best researchers in the world and has credible companies in the layer that connects…
04
Models, talent and tools
In two years the use of AI in European enterprises has more than doubled: from 8% in 2023 to 20% in 2025.
05
Adoption and productivity
Physical AI means systems where a model does not only produce text or images but controls or guides an object…
06
Industry and applications
Deep dive
Open the six chapters
Each one opens with the argument, the numbers behind it and the figure you can download and reuse.
01 Energy and data centres
Without available, stable electric power there is no computing at scale. It is the first physical constraint, not a rhetorical premise. Global data centre electricity: 415 TWh in 2024, 945 TWh in the IEA base scenario for 2030.
Schneider Electric, Siemens, ABB, Legrand and Prysmian supply electrical systems, cables, automation and cooling.
415 TWh
Global data centre electricity, 2024
IEA
945 TWh
IEA base scenario for 2030
IEA
5
European groups powering the racks: Schneider, Siemens, ABB, Legrand, Prysmian
Report ch. 06
02 Semiconductors and chips
No latest-generation chip exists without passing through this chain. It is the stretch where the dependency runs in Europe's favour: replacing these capabilities would take years and skills that do not exist elsewhere today.
ASML builds the EUV lithography machine in the Netherlands; ZEISS makes the precision optics in Germany; imec trials the processes in Belgium.
72%
Of global specialist foundry revenue attributed to TSMC, Q1 2026
TrendForce / Counterpoint
9%
EU share of the world semiconductor market
European Commission
20%
The 2030 political target the Court of Auditors calls aspirational
European Court of Auditors
Scroll the figure sideways
Foundry concentration, cross-checked
Two independent estimates, reported separately and never aggregated: perimeters and definitions differ.
Source TrendForce; Counterpoint Research
Show the data
| Company | TrendForce | Counterpoint |
|---|---|---|
| TSMC | 72.3% | 73% |
| Samsung | 6.5% | 7% |
| SMIC | 5.1% | 5% |
CSV: data/core/06_05_foundry_concentration_crosscheck.csv · Download data and sources
03 Cloud and computing capacity
Three groups control roughly two thirds of the world cloud market. European providers held 15% of their own regional market in 2024, down from 29% in 2017.
Market shares vary by perimeter and by source: they must be reported separately, never aggregated.
15%
European providers' share of their own market, 2024
Synergy Research Group
29%
The same share in 2017
Synergy Research Group
70%
Held by AWS, Microsoft and Google in 2024
Synergy Research Group
Scroll the figure sideways
European cloud dependency
Source Synergy Research Group
Show the data
| Year | European providers | Note |
|---|---|---|
| 2017 | 29% | — |
| 2024 | 15% | AWS, Microsoft and Google together: 70% |
CSV: data/core/05_04_europe_cloud_dependency.csv · Download data and sources
04 Models, talent and tools
Europe trains some of the best researchers in the world and has credible companies in the layer that connects models to business processes. The weak point is not competence: it is the ability to retain the value that competence generates.
The ELLIS network: 47 sites across 19 countries, more than 2,000 members.
47
ELLIS research sites across 19 countries
ELLIS
2,000+
Researchers in the network
ELLIS
$5.9 / 4.36 / 3.89bn
Private AI investment 2025: UK, France, Germany
Stanford AI Index
05 Adoption and productivity
In two years the use of AI in European enterprises has more than doubled: from 8% in 2023 to 20% in 2025. The indicator measures the use of at least one technology. It does not measure the depth of the transformation.
Large enterprises 55%, SMEs 19%. Denmark 42%, Romania 5.2%.
20%
European enterprises using at least one AI technology, 2025
Eurostat
8%
The same figure in 2023
Eurostat
55% vs 19%
Large enterprises versus SMEs
Eurostat
Scroll the figure sideways
AI adoption in EU enterprises
The indicator measures the use of at least one technology. It does not measure the depth of the transformation.
Source Eurostat
Show the data
| Series | Value |
|---|---|
| 2023 | 8.0% |
| 2024 | 13.5% |
| 2025 | 20.0% |
| Large enterprises (2025) | 55% |
| SMEs (2025) | 19% |
| Denmark | 42.0% |
| Finland | 37.8% |
| Sweden | 35.0% |
| Belgium | 35.0% |
| Bulgaria | 8.5% |
| Poland | 8.4% |
| Romania | 5.2% |
CSV: data/core/01_01_eu_enterprise_ai_adoption.csv · Download data and sources
06 Industry and applications
Physical AI means systems where a model does not only produce text or images but controls or guides an object in the physical world: a robot, a vehicle, a production machine, a plant. This is the ground where Europe's industrial base counts for more than the capital available.
Robot density: Western Europe 267, North America 204, Asia 131.
267
Robot density, Western Europe (per 10,000 workers)
IFR
204
North America
IFR
131
Asia
IFR
Scroll the figure sideways
The productivity J-curve
Illustrative schema, not a quantitative series. The measured evidence is listed in the table below.
Source Brynjolfsson, Rock and Syverson, The Productivity J-Curve (NBER, 2021)
Show the data
| Stage / evidence | Value |
|---|---|
| 01 Investment | Spending precedes measured output |
| 02 The trough | Measured productivity dips while work is reorganised |
| 03 Reorganisation | Gains appear once processes change |
| Customer support, field study | +14% |
| EU and US firms, EIB study | +4% |
| Knowledge workers, M365 experiment | 2 hours less email per week |
CSV: data/core/17_15_productivity_jcurve.csv · Download data and sources
The decisions that matter are not about which model to choose. They are about which processes to redesign now.
Europe in the Age of AI · 2026 edition · Lateral Space
In summary
Three things true at the same time
-
01
Europe holds capabilities no one can replace quickly
The machines that print the most advanced chips, the precision optics that make them possible, the processes trialled years before production. It is a real position, concentrated in very few organisations.
-
02
In the middle of the chain the dependencies are serious
Specialised chips, high-bandwidth memory, advanced manufacturing, global-scale cloud, frontier models, growth capital. These are not marginal gaps: they are the steps where prices and terms are set.
-
03
The applied and industrial game is still open
Manufacturing, energy, engineering, pharmaceuticals, public services: the base is there. Value will be decided by who transforms their processes first, not by who owns the best model.
The decisions that matter are not about which model to choose. They are about which processes to redesign now.
PDF
The full report
PDF · 12 MB
ZIP
Data and sources
CSV + register · ZIP
CSV
Source register
97 sources · CSV
Lateral Space, Europe in the Age of AI 2026, Research Edition, 2026. · Found something to correct? Write to hello@lateralspace.ai
For your organisation
What does this mean for your company?
The most open part of the value chain is the one where AI enters the processes, the data and the decisions of an organisation. A thirty-minute briefing to translate the report into your company's context.
Request a briefing