What counts as DeepTech?
As of 27 October 2026
Our definition
There is no generally accepted definition of DeepTech – researchers and the institutions that use the term say so themselves. The atlas therefore does not invent its own, but adopts what four independent references agree on:
DeepTech refers to technologies that are rooted in scientific discoveries or engineering breakthroughs and that require long development times and substantial capital before they reach market readiness.
The rule: a feature becomes part of the definition if at least 3 of the 4 references name it. The European Commission, the UN agency WIPO, a national academy and a scientific study – each counts once.
How the definition is derived
| Feature | EU | WIPO | RAEng | MIT / Münster | Named by |
|---|---|---|---|---|---|
| Rooted in scientific discovery or an engineering breakthrough | ✓ | ✓ | ✓ | ✓ | 4/4 |
| Long development time to market readiness | – | ✓ | ✓ | ✓ | 3/4 |
| High capital requirement | – | ✓ | ✓ | ✓ | 3/4 |
| Technology risk, high uncertainty | – | ✓ | – | ✓ | 2/4 |
| Physical core, hardware | – | ✓ | – | ✓ | 2/4 |
| Distinct from high tech: R&D intensity alone is not enough | ✓ | – | – | ✓ | 2/4 |
| Hard to replicate, protected intellectual property | – | (✓) | – | ✓ | 1/4 |
| Solutions to societal challenges | ✓ | – | – | – | 1/4 |
✓ named as a defining feature · (✓) secondary criterion only, not counted · – not named as a feature (does not mean contradicted). Highlighted rows reach the threshold.
What is deliberately left out
Five features do not reach the threshold. They are not wrong, but they are not common ground:
- Physical core. WIPO and the MIT/University of Münster study require it; the EU and the Royal Academy of Engineering do not. The Royal Academy explicitly counts AI and computing as deep tech.
- Distinct from high tech. The EU and the MIT/Münster study stress that high R&D spending alone is not enough. The definition above still asks for more than R&D intensity: a scientific origin.
- Technology risk, intellectual property, societal impact – each named as defining by at most two references.
From definition to numbers
The definition describes the concept. Development time and capital requirements cannot be read from a single publication or patent. The atlas therefore assigns publications and patents to technology fields, using a separate, documented mapping for each data source. The definition determines which fields belong; the measurement runs through the fields.
Is artificial intelligence DeepTech?
It depends on what is being developed. The definition asks about features, not industries:
- Developing AI methods and foundation models: yes. It rests on original research, takes years to reach a first product and requires substantial capital, above all for compute. All three features are met.
- Applications built on existing models: no. Building someone else's model into a product uses other people's research. Development time and capital requirements are low.
The references disagree on AI. The Royal Academy of Engineering explicitly includes it. WIPO calls developers of large language models “digital-first deep tech” but leaves them out of the scope of its own study. The MIT/University of Münster study sees AI only as a component of a physical invention. The shared definition therefore decides, not any single source.
The four references
- European Commission (2025): Horizon Europe Work Programme 2025, Part 10 – European Innovation Ecosystems. Decision C(2025) 2779 of 14 May 2025, footnote 1. Open source ↗
- World Intellectual Property Organization (2026): Global Innovation Index 2026, Appendix V – Methodology of the special theme, section 1.3. Termed “deep science” there. Open source ↗
- Royal Academy of Engineering, Enterprise Hub: State of UK Deep Tech. Open source ↗
- Kortsch, J., Raff-Heinen, S., Bendig, D., Murmann, M., Schulz, C., Murray, F. (2026): Deep-Tech Innovation: A Multi-Method Study toward a Conceptual Framework and Research Agenda. NBER Working Paper 35255. University of Münster, MIT Sloan School of Management, Bern University of Applied Sciences. Not yet peer-reviewed. Open source ↗
Further reading
- University of Cambridge, Institute for Manufacturing: Deep Technology: Definitions – an overview of eleven definitions from research and practice.
- Basilio Ruiz de Apodaca, Murray, Frolund (MIT REAP): What is “Deep Tech” and what are Deep Tech Ventures? – earlier work by the same research group; not counted separately.
- Romasanta, Ahmadova, Wareham, Pujol Priego (ESADE, 2021): Deep Tech: Unveiling the Foundations.
- European Patent Office: Deep Tech Finder – counts start-ups with at least one published European patent application.
Why other sources are not counted
- OECD: We found no DeepTech definition of the OECD's own. Its classification of industries by R&D intensity describes high tech – exactly what the EU distinguishes DeepTech from.
- Summaries: Papers that restate or bundle other definitions would count the same voice twice.
- Commercial providers: Consultancies and data providers use their own, partly undisclosed delimitations.