Deep Tech After the Software Age

Deep tech is a condition, not a sector.

The newest general framework located for this review is the May 2026 NBER working paper Deep-Tech Innovation. Its authors combine a systematic, concept-centric literature review with founder interviews and identify twelve attributes across invention, venture, and ecosystem levels.¹ The invention-level core is scientific origin, hardware intensity, and platform-enabling potential, accompanied by long development timelines, high capital intensity, and multilayered uncertainty. At the venture level, these conditions produce staged financing, simultaneous scientific and commercial maturation, and multidisciplinary teams. At the ecosystem level, they require multiple actors, specialized incubation, and industrial partners capable of de-risking and scale-up.

This definition excludes neither software nor AI. Software can be an essential control layer, discovery engine, or product component. The distinction is that a deep-tech proposition remains exposed to scientific validation, physical production, regulatory, infrastructure, or integration risk that cannot be removed by distributing code alone.

Six recent English-language papers worth reading

The following set is not exhaustive. It was selected because the papers address the definition, financing, application search, human capital, or performance of science-based and advanced-technology ventures. Publication status matters: NBER papers are working papers; the remaining items below are peer-reviewed articles or chapters.

The United States starts with a larger conversion engine.

A precise U.S.–Japan “deep-tech market” comparison is not available because databases define technologies and financing rounds differently, and the academic definition itself remains unsettled. Broad indicators nevertheless establish the scale asymmetry. U.S.-headquartered firms received $214 billion in venture capital in 2024. Critical and emerging technologies accounted for roughly two-thirds of U.S. VC, although more than 80% of that category went to software and biotechnology received only 6–9% across 2013–2024.⁷ The data reveal both strength and distortion: the United States has extraordinary risk capital, but even its capital system favors software.

Federal U.S. R&D obligations exceeded $194 billion in FY2024, while total R&D intensity was about 3.4% of GDP. Six agencies accounted for 95% of federal R&D obligations, enabling mission-scale coordination through defense, health, energy, space, agriculture, and NSF programs.⁶ The United States also contains 22 innovation clusters identified by WIPO, with dense links among universities, capital, specialized labor, and industry.¹⁴

Japan’s startup funding was reported at ¥779.3 billion in 2024—up substantially over the decade but far below the government’s aspirational ¥10 trillion FY2027 target.¹¹ NEDO has established an approximately ¥100 billion deep-tech startup fund, with additional green-transformation support, stage gates, mass-production demonstrations, and overseas validation.¹⁰ These are meaningful interventions, but they operate within a smaller private financing and exit market.

A directional scenario—not a market-size forecast

Because “deep tech” is not a harmonized statistical category, a single 2035 revenue number would create false precision. The more defensible forecast concerns commercialization capacity: how many science-based inventions can be financed, validated, manufactured, purchased, and expanded. The following scenarios are analytical judgments derived from the evidence above, not outputs from an econometric model.

The baseline expectation is therefore divergence in absolute scale but not necessarily in strategic relevance. The United States can host more large ventures; Japan can occupy indispensable positions in the technologies those ventures require. The important policy question is whether Japan captures only supplier margin or also retains platform ownership, data, standards, recurring revenue, and equity value.

Japan’s problem is not an absence of science.

Japan leads WIPO’s measure of production and export complexity, reflecting deep capability in difficult-to-replicate products and supply chains.¹⁴ Yet METI reports that R&D expenditure in Japan remained roughly flat from 2007 to 2021 while the United States, Germany, South Korea, and other comparators expanded by approximately 1.5–2.5 times.⁸ METI also estimates that large companies conduct about 90% of Japanese private-sector R&D and that roughly 60% of technologies those companies cannot commercialize are abandoned.⁹

1. 01
   Capital-duration mismatch

   Laboratories need years; conventional funds and corporate budgets often need visible results sooner.

2. 02
   Prototype–procurement discontinuity

   A grant may fund demonstration without creating a customer authorized to buy an imperfect first system.

3. 03
   Institutional captivity

   Patents, facilities, and experts can remain inside organizations that cannot prioritize a new market.

4. 04
   Application-search weakness

   Teams continue improving technology before identifying a painful, fundable, and reachable customer problem.

5. 05
   Domestic optimization

   Products are validated for Japanese partners and standards before global market design, capital, and pricing are settled.

6. 06
   Scale-up fragmentation

   Pilot plants, certification, manufacturing, hiring, and international sales are financed as separate problems rather than one journey.

These mechanisms are more actionable than cultural caricatures about risk aversion. Culture matters through incentives: career penalties for leaving, procurement penalties for failure, governance processes that require certainty, and financing structures that reward small exits. Change the pathway and behavior can change with it.

Build five bridges from laboratory to market.

1. 01
   Application readiness before technical perfection

   Track technology discovery, validation, market identification, and application together. Add business expertise early enough to change the product—not merely to sell it later.

2. 02
   Patient capital with explicit gates

   Finance scientific proof, engineering proof, pilot production, certification, and scale separately, with evidence requirements appropriate to each stage.

3. 03
   First-customer institutions

   Use government, infrastructure operators, hospitals, factories, and major corporations as paid test customers with bounded risk and transparent evaluation.

4. 04
   Entrepreneur-led carve-outs

   Move unused corporate technology, key inventors, relevant IP, and access to facilities into ventures with genuine decision rights. METI’s 2024 guidance provides a starting mechanism.⁹

5. 05
   Global-by-formation teams

   Combine Japanese science and production with international CEOs, regulatory leaders, specialist investors, and launch customers before domestic optimization hardens the design.

The resulting model is not “more subsidies.” It is a chain of accountable conversions. Every supported project should identify the next irreversible proof: a validated mechanism, manufacturable unit, regulatory pathway, signed pilot, repeat order, or scalable cost curve.

What is becoming interesting now

These fields are “hot” for different reasons. Some have crossed a scientific threshold; some have attracted mission-driven capital; some are responding to labor, energy, security, or supply-chain constraints. None is guaranteed to become a large market merely because the science is impressive.

Wood is becoming a technology platform.

Wood appears ordinary because society encounters it at the scale of boards and paper. At the nanoscale, cellulose can be separated and engineered into materials with unusual combinations of weight, strength, stiffness, thermal stability, transparency, and renewability. AIST describes cellulose nanofibers as roughly one-fifth the weight of steel and more than five times stronger, while emphasizing that safety evidence is essential for industrial adoption.¹⁸

This field now combines forestry, chemistry, materials science, machine learning, electronics, automotive engineering, packaging, and energy storage. In 2025, AIST reported a machine-learning technique for evaluating nanocellulose morphology and predicting composite properties—an example of AI improving the quality-control layer required for scale.¹⁹ Recent research has also demonstrated programmable flexible wood-based origami electronics and cellulose-nanofiber supercapacitors.²⁰²¹

Why deep tech is compelling

Software reorganized information. Deep tech can reorganize matter, energy, biology, and physical agency. It makes previously impossible actions possible: predict a molecular interaction, manufacture a lighter structure, sense what could not be measured, control a machine in an unstructured environment, or generate energy through a new physical pathway.

Its creative power also comes from combinations. AI becomes more consequential when joined to laboratory automation; robotics becomes more valuable when paired with new sensors and materials; wood becomes a frontier when nanostructure, chemistry, computation, and industrial design meet. The best deep-tech stories are therefore not about a single miraculous object. They are about a new stack of capabilities that changes what an industry can do.

The difficulty is part of the meaning. Long timelines, physical constraints, and the need for many kinds of expertise prevent instant replication. They also create moral and institutional stakes: who controls the platform, who bears experimental risk, which communities receive the benefits, and what infrastructure becomes locked in for decades.

Five questions for the people building it

A five-person interview series can move beyond technology promotion by asking every builder the same underlying questions. The answers will make different fields comparable without flattening their scientific differences.

1. 01
   What became possible?

   Name the scientific or engineering discontinuity—not the product slogan.

2. 02
   What is still unproven?

   Separate laboratory evidence, manufacturability, safety, unit economics, and customer demand.

3. 03
   Where is the missing bridge?

   Identify the absent facility, regulation, partner, talent, capital, standard, or first buyer.

4. 04
   Why should Japan matter?

   Specify the distinctive scientific, industrial, regional, or cultural asset rather than invoking national strength in the abstract.

5. 05
   What would success look like in 2035?

   Describe a measurable capability in society, not only a valuation or funding round.

Limitations and conclusion

This article is a targeted narrative synthesis and editorial scenario, not a systematic review, investment recommendation, or statistically estimated market forecast. “Deep tech,” “critical technologies,” and “science-based ventures” overlap but are not identical categories. U.S. VC data and Japanese startup-funding data use different sources and should not be read as a direct ratio. Policy commitments may change, and frontier demonstrations do not establish commercial readiness.

The central conclusion is narrower and more durable: the United States is currently better positioned to finance and form large markets around science-based ventures, while Japan owns scientific and manufacturing capabilities that are strategically important but incompletely converted into independent growth companies. Japan’s 2035 outcome will depend less on producing more inventions than on connecting inventions to application search, patient capital, first customers, industrial scale-up, global talent, and value capture. Deep tech is interesting because it expands the frontier of physical possibility. It becomes economically important only when a society builds the institutions that allow the possibility to travel.

Selected sources

1. Kortsch, J., Raff-Heinen, S., Bendig, D., Murmann, M., Schulz, C., & Murray, F. “Deep-Tech Innovation: A Multi-Method Study toward a Conceptual Framework and Research Agenda.” NBER Working Paper 35255 (2026). https://www.nber.org/papers/w35255.
2. Ando, Y., Dinlersoz, E., Greenwood, J., & Piazzesi, R. “Technifying Ventures.” NBER Working Paper 33993 (2025; revised 2026). https://www.nber.org/papers/w33993.
3. Arora, A., Fosfuri, A., & Rønde, T. “The Missing Middle: Value Capture in the Market for Startups.” Research Policy 53, no. 4 (2024): 104958. https://doi.org/10.1016/j.respol.2024.104958.
4. Denoo, L., Van Boxstael, A., & Belz, A. “Help, I Need Somebody! Business and Technology Advisors in Emerging Science-Based Ventures at American Universities.” Journal of Technology Transfer 49 (2024): 1567–1605. https://doi.org/10.1007/s10961-024-10125-2.
5. Kask, J., & Linton, G. “Navigating the Innovation Process: Challenges Faced by Deep-Tech Startups.” In Contemporary Issues in Industry 5.0 (2025), 197–219. https://doi.org/10.1007/978-3-031-74779-3_8.
6. National Science Board. The State of U.S. Science and Engineering 2026. National Center for Science and Engineering Statistics, 2026. https://www.ncses.nsf.gov/pubs/nsbsep20261.
7. National Science Board. “Venture Capital Investment Received by Firms Headquartered in Selected Regions, Countries, or Economies: 2013–24” and “Venture Capital Investment by Critical and Emerging Technology.” 2026. https://www.ncses.nsf.gov/pubs/nsbsep20261/figure/34.
8. Ministry of Economy, Trade and Industry, Japan. “Innovation Subcommittee Compiles Interim Report.” June 21, 2024. https://www.meti.go.jp/english/press/2024/0621_001.html.
9. Ministry of Economy, Trade and Industry, Japan. “Guidance for Putting Entrepreneur-Led Carve-Outs into Practice.” April 26, 2024. https://www.meti.go.jp/english/press/2024/0426_003.html.
10. New Energy and Industrial Technology Development Organization. “Deep-Tech Startups Support Program.” Updated 2026. https://www.nedo.go.jp/english/activities/activities_ZZJP_100250.html.
11. Japan Investment Corporation. Global and Japan Venture Investment Trends. April 2025. https://www.j-ic.co.jp/en/research/.assets/E_20250411_JIC_Research.pdf; Cabinet Office, Japan. “Startup Development Five-Year Plan.” 2022. https://www.cao.go.jp/minister/2210_s_goto/kaiken/20221124kaiken.html.
12. Japan External Trade Organization. JETRO Invest Japan Report 2025. 2026. https://www.jetro.go.jp/ext_images/en/invest/investment_environment/ijre/2025/ijreENreport2025.pdf.
13. Ministry of Economy, Trade and Industry, Japan. “Selection of R&D Themes for AI-Ready Manufacturing Data and Robotics Foundation Models under GENIAC.” May 14, 2026. https://www.meti.go.jp/english/press/2026/0514_001.html.
14. World Intellectual Property Organization. Global Innovation Index 2025. 2025. https://www.wipo.int/web-publications/global-innovation-index-2025/en/gii-2025-results.html.
15. Abramson, J., et al. “Accurate Structure Prediction of Biomolecular Interactions with AlphaFold 3.” Nature 630 (2024): 493–500. https://doi.org/10.1038/s41586-024-07487-w.
16. Google Quantum AI and Collaborators. “Quantum Error Correction below the Surface Code Threshold.” Nature 638 (2025): 920–926. https://doi.org/10.1038/s41586-024-08449-y.
17. U.S. Department of Energy. Fusion Science and Technology Roadmap. June 2026. https://www.energy.gov/documents/fusion-science-and-technology-roadmap.
18. National Institute of Advanced Industrial Science and Technology. “Safety Assessment of Cellulose Nanofibers (FY2017–2024).” 2026. https://riss.aist.go.jp/en/research-outcomes/947/.
19. AIST. “New Evaluation Technique for Nanocellulose Using Machine Learning.” March 4, 2025. https://www.aist.go.jp/aist_e/list/latest_research/2025/20250304/en20250304.html.
20. Xia, Q., et al. “Programmable and Flexible Wood-Based Origami Electronics.” Nature Communications 15 (2024). https://doi.org/10.1038/s41467-024-53708-1.
21. Fukuhara, M., et al. “High-Energy Density Cellulose Nanofibre Supercapacitors Enabled by Pseudo-Solid Water Molecules.” Scientific Reports 14 (2024): 10350. https://doi.org/10.1038/s41598-024-61128-w.
22. Bahoo-Torodi, A., R. Fontana, and F. Malerba. “Pre-entry Experience and the Heterogeneity in Startup Performance: Evidence from the Nascent Artificial Intelligence Industry.” Research Policy 55, no. 1 (2026): 105367. https://doi.org/10.1016/j.respol.2025.105367.