From Workspace to Intelligence Infrastructure: A Comparative Study of AI-Ready Incubation Models

From Workspace to Intelligence Infrastructure: A Comparative Study of AI-Ready Incubation Models

Dr. Saumya Sharma Faculty

FLAME University, Pune, India saumyasharma4910@gmail.com

Dr. Smrita Jain Associate Professor

Department of Management MIT, Moradabad, India smritagupta29@gmail.com

Ms. Suhani Bhatia

Consultant-Recruitment Operations Bersavi India Pvt. Ltd. suhanibhatia50@gmail.com

AbstractArtificial intelligence (AI) and automation are altering both what entrepreneurs build and how ventures are created, tested, governed, and scaled. Yet many university incubators remain organised around inexpensive workspace, periodic mentoring, and investor introductions. This study examines how incubation should be redesigned for AI-enabled techno-entrepreneurship. It uses an empirical comparative document analysis of nine purposively selected cases: IIT Madras Research Park, IIMA Ventures, SINEIIT Bombay, T-Hub, StartX, Y Combinator, MITdesignX, TUM Venture Labs, and PACT. Official webpages and reports, supplemented by peer- reviewed literature and policy evidence, were coded using eight observable categories: research integration, specialist infrastructure, structured venture development, finance, market access, internationalisation, AI-specific support, and publicly documented responsible-AI governance. The evidence matrix shows broad provision of structured venture development and market or investor connection, but uneven specialist infrastructure and AI-specific support. Only three cases documented a named AI programme or laboratory in the reviewed material, while none published a venture-facing responsible-AI governance pathway on the pages examined. The findings indicate that no single model is directly transferable. Research parks provide co-location and facilities; accelerators provide tempo, selection, and investor access; university platforms mobilise knowledge and talent; and regional parks connect ventures to place-based priorities. Based on these observed complementarities and gaps, the paper proposes the AI-Ready Incubation System (AIRIS): six mutually reinforcing capabilities covering compute and data, rapid experimentation, interdisciplinary talent, responsible governance, market integration, and sustainable finance with outcome measurement. AIRIS offers a testable design framework for universities moving from property-led incubation to accountable intelligence infrastructure.

Index TermsAccelerators, artificial intelligence, research parks, responsible innovation, startup incubation, university entrepreneurship.

1. Introduction

   Artificial intelligence and automation are changing the basic economics of venture creation. Founders can now use software to generate code, analyse markets, automate routine operations, personalise services, and run experiments at a speed that previously required larger teams. Adoption is no longer confined to technology firms. The 2026 AI Index from the Stanford Institute for Human-Centered Artificial Intelligence (Stanford HAI) reports that 88% of surveyed organisations used AI in 2025, although the use of autonomous agents remained at an early stage across most business functions [1]. At the same

   time, the World Economic Forum [2] estimates that 39% of workers' existing skill sets will change or become outdated between 2025 and 2030. These shifts expand the opportunity for techno-entrepreneurship, but they also raise the cost of weak judgment: a venture can build the wrong product faster, expose sensitive data at scale, or automate a flawed decision before it understands the affected user.

   This transformation creates a design problem for incubators. The first generation of business incubation concentrated on affordable premises and shared administrative services. Later generations added coaching, training, networks, and finance [3]. Contemporary incubation therefore cannot be evaluated by occupancy alone. The Organisation for Economic Co-operation and Development (OECD) [4] describes incubation as an extended package of services delivered by a distinct entity and recommends that incubators operate as ecosystem connectors, provide individualised support, build specialisation where justified, and link public funding to measurable delivery. For AI-enabled ventures, the service package must extend further: founders require lawful data access, compute, model evaluation, cybersecurity, multidisciplinary review, and routes to pilot deployment.

   India makes this question particularly relevant. Startup India reported more than 157,000 recognition certificates issued by December 2024 [5]. A large pipeline, however, does not automatically create high-quality ventures or research commercialisation. University incubators occupy a distinctive position because they can combine students, faculty knowledge, laboratories, alumni, industry relationships, and patient experimentation. Their weakness is often fragmentation: entrepreneurship education, research, intellectual property, computing infrastructure, corporate engagement, and seed funding may sit in separate administrative units.

   This paper asks three questions: (1) Which design features of established Indian and international incubation models are most relevant to AI-enabled entrepreneurship? (2) What patterns and gaps are observable across public case evidence? and (3) What integrated framework can guide university incubators without assuming that one celebrated model can simply be copied? The paper contributes an empirical case-evidence matrix and an analytical frameworkthe AI-Ready Incubation System (AIRIS). It does not claim causal proof or rank the selected organisations.

2. Incubation, Acceleration, and Digital Entrepreneurship

   Incubators are heterogeneous organisations. Bergek and Norrman [6] proposed that an incubator model should be understood through selection, infrastructure, business support, mediation, and graduation. Mian et al. [7] similarly show that technology business incubation includes different institutional forms, including university incubators, science parks, and accelerators. Accelerators are usually more time-bound, cohort- based, selective, and intensive than conventional incubators [8]. Science and research parks emphasise place, research-intensive firms, shared facilities, and proximity to knowledge institutions. These forms overlap in practice, but their causal mechanisms differ. Treating all of them as interchangeable obscures what a university is actually trying to achieve.

   Research also cautions against assuming that a prominent programme's visible features cause its outcomes. Incubated ventures are selected rather than randomly assigned, and high- potential founders may perform well even without support. Programme design nevertheless matters. Cohen et al. [8] find substantial variation among accelerators and relate design choices to venture outcomes. Hallen et al. [9] show that effective acceleration depends on mechanisms that help ventures learn rapidly rather than simply compressing time. The OECD's [4] synthesis reports mixed evidence on survival and the durability of performance effects, but notes stronger evidence for employment and revenue growth in many evaluations. It also cites evidence that entrepreneurship schooling can improve subsequent financing while cash or coworking alone may not. The practical implication is that space is useful when it enables learning, experimentation, and connection; it is not a sufficient theory of change.

   Digital entrepreneurship further alters the unit of analysis. Digital technologies reduce boundaries between product, process, and platform and make entrepreneurial action moe fluid [10]. They also reorganise innovation through distributed participation, recombinable resources, and rapidly changing outcomes [11]. AI adds probabilistic systems whose performance depends on data quality, context, model behaviour, and human oversight. Obschonka and Audretsch

   [12] argue that AI and big data can transform entrepreneurial practice and research, but also require new knowledge and careful examination of reciprocal effects between people and machines.

   Accordingly, an AI-ready incubator has two roles. First, it supports ventures that produce AI or automation products. Second, it helps all portfolio ventures use AI responsibly in research, product development, operations, and customer discovery. These roles require technical support, but not a technology-only philosophy. The World Economic Forum [2] identifies AI and big data among the fastest-growing skills while also emphasising analytical thinking, resilience, leadership, and collaboration. The incubator must therefore connect computational capability with design, behavioural understanding, domain expertise, ethics, and enterprise execution.

3. METHOD

   The study used empirical comparative document analysis. It combined three evidence streams: peer-reviewed research on incubation, acceleration, digital entrepreneurship, and AI; recent policy and standards documents; and a corpus of official webpages and reports from selected cases. The review was conducted through September 4, 2026. The OECD's 2026 cross-country report provided the policy baseline because it synthesises services, specialisation, internationalisation, and evaluation evidence across incubation systems. AI context was drawn from Stanford HAI, the World Economic Forum, the National Institute of Standards and Technology (NIST), and UNESCO.

   Cases were selected purposively, not statistically. The selection criterion was theoretical variation in the mechanism of support, coupled with the availability of first-party information. Four Indian cases represent different institutional arrangements: IIT Madras Research Park (researchindustry co-location), IIMA Ventures (venture development and investment), SINEIIT Bombay (university technology incubation), and T-Hub (large-scale ecosystem and corporate programmes). Five international cases add contrasting logics: StartX (a fee-free, equity-free Stanford-affiliated community), Y Combinator (capital-backed intensive acceleration), MITdesignX (design-led venture development), TUM Venture Labs (domain-specific deep-technology infrastructure), and PACT in Portugal (regional science and technology park).

   A directed coding protocol converted the documents into a case-level evidence matrix. Eight categories were defined before cross-case comparison: (1) research or academic integration; (2) specialist technical infrastructure; (3) a structured venture-development pathway; (4) funding or investor access; (5) customer, corporate, or market access; (6) internationalisation support; (7) a named AI-specific programme, laboratory, or centre; and (8) a published venture- facing responsible-AI governance pathway. Each category received D when an operational mechanism or named programme was documented, M when a general capability was mentioned without an identifiable mechanism, and ND when the feature was not documented in the reviewed material. ND means not documented, not does not exist. Coding was based only on sources listed in the reference section and was checked against exact source language before synthesis. Counts are descriptive and unweighted; no aggregate ranking was constructed because categories serve different purposes.

   Public statements were treated as evidence of documented model design, not independent proof of impact. Figures reported by organisations were retained only when directly attributable and were not compared as if definitions were uniform. This procedure creates an auditable secondary dataset but remains subject to website selection, uneven disclosure, and single-coder interpretation. The paper therefore uses the matrix to identify patterns and questions, not to infer causal superiority.

4. Comparative Findings

   Table I demonstrates that the cases do not represent a single ladder from basic to best. They solve different coordination failures. IIT Madras Research Park describes itself as India's first university-based research park and reports more than 250 research-and-development companies and laboratories and over 6,000 technologists [13]. Its distinguishing mechanism is proximity among industry research teams, academic experts, centres of excellence, startups, and enabling organisations. The lesson is not that every university needs a large park; it is that repeated research collaboration benefits from stable co- location, technical facilities, and boundary-spanning staff.

   IIMA Ventures combines incubation, acceleration, investment, and ecosystem knowledge. Its official history traces the initiative to an innovation centre established at IIM Ahmedabad in 2002, later supported as a Centre of Excellence, and its incubation programmes range from three to twelve months [14], [15]. This model shows the value of an institution that can follow ventures through multiple stages and construct specialised programmes around social or market problems. The Society for Innovation and Entrepreneurship (SINE) at IIT Bombay illustrates the university technology-business- incubator logic: infrastructure, mentorship, funding, and institutional access are assembled around technology ventures [16].

   T-Hub represents an ecosystem-at-scale approach. Its official material presents sector- and stage-specific programmes, corporate and government innovation, investor networks, international market access, and coworking. It reports serving more than 10,000 startups, a figure that should be read as organisational self-reporting rather than a standardised outcome measure [17]. Its partnership with the Department of Science and Technology on the Machine Learning and Artificial Intelligence Technology Hub illustrates how a general innovation platform can add a specialised AI capability [18]. The transferable lesson is modularity: a broad hub can create specialist programmes without turning every venture into a deep-technology company.

   International models illuminate different trade-offs. StartX states that it charges no fee and takes no equity; eligibility is linked to Stanford affiliation, and the programme relies on a selected founder community, mentoring, partner resources, and investor access [19], [20]. The model uses institutional affiliation and community quality as scarce assets. Its financing should not be conflated with the separate StanfordStartX investment fund. Y Combinator follows the opposite financial

   logic: it currently invests USD 500,000 through two instruments, including USD 125,000 for 7% and USD 375,000 through an uncapped most-favoured-nation SAFE, and then provides an intensive three-month programme and long-term alumni network [21], [22]. The comparison shows that founder support can be sustained without taking equity only when another durable funding base exists; free incubation is a financing choice, not an absence of cost.

   MITdesignX demonstrates venture creation through design disciplines and structured problemsolution validation. Its 2026 cohort produced 12 startups, according to MIT's event listing [23]. TUM Venture Labs offers a network of domain-specific deep-technology and life-science incubators and provides expertise, co-working, networks, and specialised equipment. Its Software and AI, Robotics and AI, Quantum/Semiconductor, and Food/Agro/Biotech laboratories show why specialist infrastructure should follow a domain thesis rather than a generic innovation lab label [24], [25]. PACT offers a place- based counterpoint: as a regional science and technology park in Alentejo, it participates in cros-border open-innovation programmes and connects incubation to regional development [26].

   The coded matrix in Table II adds three empirical observations. First, all nine cases documented or mentioned structured venture development, finance access, and market connection, although the intensity and mechanism varied. Second, specialist technical infrastructure was operationally documented in four cases and mentioned more generally in four others; Y Combinator's reviewed material emphasised programme and network rather than technical space. Third, named AI-specific support appeared in only three casesIIMA Ventures, T-Hub, and TUM Venture Labswhile no reviewed case page documented a complete venture-facing responsible- AI pathway. This last result is a disclosure finding, not proof that internal controls are absent.

   Across the cases, five patterns recur. First, selection determines whether scarce expertise can be meaningfully concentrated. Second, useful infrastructure ranges from desks to wet labs and compute, but must correspond to venture tasks. Third, programmes create value when they shorten evidence- building cycles with users, experts, and capital providers. Fourth, networks require active mediation; proximity alone does not create collaboration. Fifth, sustainable models combine multiple resourcesuniversity support, rent or service income, sponsorship, grants, programme contracts, donations, or investment returnsrather than expecting early-stage founders to cover the full cost.

   TABLE I

   COMPARATIVE MECHANISMS IN SELECTED INCUBATION MODELS

   Model	Primary mechanism	Transferable lesson for AI-ready incubation
   IIT Madras Research Park [13], [27]	Researchindustry co-location	Create repeated pathways between faculty, R&D teams, centres, and ventures; build facilities around sustained demand.
   IIMA Ventures [14], [15]	Multi-stage venture platform	Combine incubation, acceleration, investment readiness, specialist programmes, and knowledge products.
   SINEIIT Bombay [16]	University technology incubation	Link technical ventures to institutional infrastructure, mentoring, finance, and commercialisation support.
   T-Hub [17], [18], [28]	Large-scale ecosystem platform	Use modular sector programmes, corporate challenges, investor networks, and international market bridges.
   StartX [19], [20]	Selected founder community; no fee/equity	Treat trusted peers and long-term alumni participation as infrastructure; secure an alternative funding base.
   Y Combinator [21], [22]	Capital-backed intensive acceleration	Use clear terms, short learning cycles, concentrated advice, customer access, and an enduring alumni network.
   MITdesignX [23], [29]	Design-led venture development	Institutionalise problem framing, user evidence, prototyping, and interdisciplinary venture design.
   TUM Venture Labs [24], [25]	Domain-specific deep-tech network	Match expert staff and specialist equipment to explicit domains rather than creating generic labs.
   PACT [26]	Regional science-and- technology park	Connect incubation to regional priorities, cross-border partnerships, and place-based economic development.

   Note. Organisational descriptions are based on cited official sources and indicate model design, not independently verified comparative performance.

   TABLE II

   DOCUMENTED CAPABILITY EVIDENCE ACROSS NINE CASES

   Case	Research	Specialist infra.	Venture pathway	Finance	Market	Inter- national	AI- specific	Responsible AI
   IIT Madras Research Park	D	D	M	M	D	M	ND	ND
   IIMA Ventures	M	M	D	D	D	D	D	ND
   SINEIIT Bombay	D	M	D	D	M	ND	ND	ND
   T-Hub	M	D	D	D	D	D	D	ND
   StartX	D	M	D	D	D	ND	ND	ND
   Y Combinator	ND	ND	D	D	D	M	ND	ND
   MITdesignX	D	M	D	M	M	M	ND	ND
   TUM Venture Labs	D	D	D	M	M	M	D	ND
   PACT	M	D	M	M	D	D	ND	ND
   D count	5	4	7	5	6	3	3	0
   D or M count	8	8	9	9	9	7	3	0

   Note. D = an operational mechanism or named programme was documented; M = a general capability was mentioned; ND = not documented in the reviewed material. ND must not be interpreted as evidence of absence. Counts are descriptive and unweighted.

   Capability 1: Accessible compute and governed data. Early-

5. The AI-Ready Incubation System (AIRIS)

   The comparative evidence suggests that AI readiness is not a separate vertical beside incubation. It is an operating architecture connecting six capabilities. Fig. 1 presents AIRIS as a cycle because each capability changes the quality of the others. Better market access produces better problem and data definitions; governance improves trust and pilot access; experiments reveal capability gaps; and outcome evidence shapes future funding.

   stage firms need flexible access to cloud credits, secure development environments, approved datasets, data- management support, and cost visibility. The objective is not to build an expensive data centre by default. Universities should first aggregate demand, negotiate shared cloud access, classify data by sensitivity, and create secure environments for projects that cannot use open platforms. Access rules should distinguish public, licensed, institutional, personal, and commercially confidential data. Compute subsidies should be time-limited

   and linked to experiments so that ventures learn unit economics rather than becoming dependent on hidden infrastructure costs. Capability 2: Rapid experimentation and automation infrastructure. The incubator should provide reusable tools for prototyping, model evaluation, human-in-the-loop testing, cybersecurity checks, and deployment monitoring. Physical acilities remain important for ventures connecting AI to robotics, manufacturing, health, climate, mobility, or media. TUM Venture Labs' domain equipment and IIT Madras Research Park's researchindustry setting illustrate how infrastructure becomes valuable when attached to a specific development pathway. Each venture should maintain an experiment log stating the hypothesis, dataset or user sample,

   success threshold, risk, result, and next decision.

   Capability 3: Interdisciplinary human capability. AI venture teams often over-invest in building and under-invest in problem framing, adoption, and organisational change. An AI-ready university incubator should assemble faculty and students from computing, design, management, economics, behavioural science, law, communication, and relevant application domains. A two-in-a-box advisory modelone technical reviewer and one domain or user reviewercan reduce blind spots. Training should combine model literacy, customer discovery, workflow redesign, financial modelling, storytelling, negotiation, and team leadership. This mirrors the evidence that technology and human skills are complementary rather than substitutable [2].

   Capability 4: Responsible-AI, intellectual-property, and security governance. Governance must appear before the demo day. NIST's AI Risk Management Framework organises practice around Govern, Map, Measure, and Manage and identifies trustworthy characteristics including validity, safety, security, transparency, explainability, privacy, and fairness [30]. UNESCO's recommendation adds human rights, dignity, proportionality, inclusion, environmental responsibility, and human oversight [31]. Incubators can translate these principles into a proportionate intake screen: intended users; affected non- users; data provenance and consent; foreseeable harm; bias and accessibility; model limitations; cybersecurity; human appeal or override; intellectual-property ownership; and regulatory exposure. Higher-risk ventures should receive specialist review, while low-risk prototypes should not be burdened with hospital-grade procedures.

   Capability 5: Market and industry integration. AI prototypes fail when they lack a credible workflow, buyer, data owner, or deployment setting. The incubator should recruit challenge partners that commit problem access, a responsible decision- maker, a pilot environment, and a post-pilot evaluation. IIT Madras Research Park demonstrates repeated industry academia proximity; T-Hub demonstrates corporate programmes and international bridges; Y Combinator demonstrates concentrated investor and early-customer access. A university can combine these mechanisms through quarterly challenge calls, paid proof-of-concept agreements, procurement guidance, alumni customer circles, and a small pilot office that protects both the startup and the partner.

   Capability 6: Sustainable finance and outcome measurement. Founder-friendly pricing is desirable, but the cost must be assigned transparently. A university portfolio may combine institutional base funding, government or philanthropic grants, corporate challenge fees, sponsored laboratories, rent, paid executive education, research contracts, and a separately governed investment vehicle. Equity should never be taken merely in exchange for a desk; if used, it should correspond to capital and intensive venture-building support. Performance dashboards should separate activity from outcomes. Activity includes applications, mentoring hours, and events. Intermediate outcomes include validated problems, prototypes, pilots, research licences, follow-on experiments, and founder capability. Longer-term outcomes include revenue, external finance, survival, quality employment, productivity, inclusion, and responsible deployment. Counterfactual evaluation remains difficult, so cohort tracking and comparison groups should be introduced where feasible.

   AIRIS yields four propositions for future testing. Proposition 1: AI-enabled ventures receiving integrated compute, domain, and market support will progress from prototype to paid pilot faster than ventures receiving generic mentoring alone. Proposition 2: early risk and data review will reduce pilot failure caused by privacy, security, intellectual-property, or trust concerns. Proposition 3: interdisciplinary team composition will improve problemsolution fit, especially in human-facing sectors. Proposition 4: incubators that measure learning milestones and responsible deployment, rather than event volume or occupancy alone, will allocate resources more effectively. These are research propositions, not findings of the present desk review.

   Fig. 1. The AIRIS operating cycle. The six capabilities operate as a feedback cycle rather than a fixed sequence. Framework developed in this paper.

6. Implementation Implications for University Incubators

   AIRIS can be introduced in stages. During the first six months, a university should map existing assets, appoint a small professional team, establish selection criteria, adopt basic data and AI-risk templates, and run a problem-discovery cohort without major construction. The minimum team comprises an incubation lead, venture-development manager, technology or product lead, partnerships manager, and shared legal, finance,

   and administrative support. A faculty council can advise, but day-to-day decisions require delegated operational authority.

   In months 718, the incubator can launch two or three domains where the university has genuine faculty depth and partner demand. Each domain should have a named academic lead, industry council, pilot partners, mentor pool, and evidence-based curriculum. Compute and specialist equipment should be leased or shared before being purchased. Selection should assess founder commitment, problem importance, access to users or data, feasibility, team complementarity, and risknot presentation polish alone. Programmes can include a twelve-week validation pathway for student teams and a six- to twelve-month pathway for research-led or external ventures.

   In months 1936, recurring demand can justify dedicated facilities, sponsored labs, international soft-landing partnerships, and an independently governed seed vehicle. Graduation should occur when a venture no longer needs subsidised support, not at an arbitrary ceremony. Alumni should return as customers, mentors, investors, and peer reviewers. This creates the cumulative community advantage visible at StartX and Y Combinator while preserving the university's educational and public-interest mission.

   The framework also clarifies what not to copy. A university should not reproduce a large research park without sustained industry research demand; promise no-fee incubation without a funding base; create generic labs without qualified staff and utilisation plans; count every participant as an incubated startup; or use AI as a branding label for conventional workshops. The correct model is contingent on institutional strengths and regional demand, consistent with the OECD's [4] emphasis on specialisation, ecosystem connection, and performance-linked support.

7. Limitations and Research Agenda

   This paper has four limitations. First, it relies on public documents, which vary in detail and may emphasise organisational achievements. Second, purposive cases permit analytical comparison but not statistical generalisation. Third, programme counts and self-reported impact figures are not standardised. Fourth, AIRIS has not yet been validated through venture-level data. Future research should develop measurable constructs for its six capabilities, compare matched venture cohorts, examine founder and partner experience, and test whether benefits vary by sector, venture stage, and institutional context. Longitudinal research should also measure unintended effects, including subsidy dependence unequal access to compute, automation-related displacement, and harms from deployed systems.

8. Conclusion

AI and automation do not make incubators obsolete; they make weak incubation easier to expose. A desk, a mentor list, and a demo day cannot by themselves provide the data, experimentation, judgment, trust, and market access required for AI-enabled enterprise. The comparative cases show that effective systems specialise in distinct mechanisms: research

parks organise proximity, university platforms mobilise knowledge and legitimacy, accelerators create tempo and investment access, community models accumulate peer capital, and regional parks align innovation with place.

AIRIS integrates these lessons into six capabilities: compute and data, experimentation, interdisciplinary talent, responsible governance, market integration, and sustainable finance with measurement. Its central claim is practical: an AI-ready incubator should be designed as intelligence infrastructurea governed system that helps people turn knowledge into evidence, evidence into responsible enterprise, and enterprise into measurable value. Universities are well placed to build such systems, provided they begin with a clear theory of change and institutional strengths rather than a building or borrowed brand.

REFERENCES

1. Stanford Institute for Human-Centered Artificial Intelligence, The 2026 AI Index Report. Stanford University, 2026. [Online]. Available: https://hai.stanford.edu/ai-index/2026-ai-index-report

2. World Economic Forum, The Future of Jobs Report 2025, 2025. [Online]. Available: https://www.weforum.org/publications/the-future-of-jobs- report-2025/

3. J. Bruneel, T. Ratinho, B. Clarysse, and A. Groen, The evolution of business incubators: Comparing demand and supply of business incubation services across different incubator generations, Technovation, vol. 32, no. 2, pp. 110121, 2012, doi: 10.1016/j.technovation.2011.11.003.

4. Organisation for Economic Co-operation and Development, Incubation in Entrepreneurial Ecosystems: Hatching Growth. OECD Publishing, 2026, doi: 10.1787/21be9d03-en.

5. Startup India, Prabhaav: 9-Year Factbook. Department for Promotion of Industry and Internal Trade, 2025. [Online]. Available: https://www.startupindia.gov.in/content/sih/en/Prabhaav.html

6. A. Bergek and C. Norrman, Incubator best practice: A framework, Technovation, vol. 28, nos. 12, pp. 2028, 2008, doi: 10.1016/j.technovation.2007.07.008.

7. S. Mian, W. Lamine, and A. Fayolle, Technology business incubation: An overview of the state of knowledge, Technovation, vols. 5051, pp. 1 12, 2016, doi: 10.1016/j.technovation.2016.02.005.

8. S. Cohen, D. C. Fehder, Y. V. Hochberg, and F. Murray, The design of startup accelerators, Research Policy, vol. 48, no. 7, pp. 17811797, 2019, doi: 10.1016/j.respol.2019.04.003.

9. B. L. Hallen, S. L. Cohen, and C. B. Bingham, Do accelerators work? If so, how? Organization Science, vol. 31, no. 2, pp. 378414, 2020, doi: 10.1287/orsc.2018.1276.

10. S. Nambisan, Digital entrepreneurship: Toward a digital technology perspective of entrepreneurship, Entrepreneurship Theory and Practice, vol. 41, no. 6, pp. 10291055, 2017, doi: 10.1111/etap.12254.

11. S. Nambisan, M. Wright, and M. Feldman, The digital transformation of innovation and entrepreneurship: Progress, challenges and key themes, Research Policy, vol. 48, no. 8, Art. no. 103773, 2019, doi: 10.1016/j.respol.2019.03.018.

12. M. Obschonka and D. B. Audretsch, Artificial intelligence and big data in entrepreneurship: A new era has begun, Small Business Economics, vol. 55, no. 3, pp. 529539, 2020, doi: 10.1007/s11187-019-00202-4.

13. IIT Madras Research Park, About us, 2026. [Online]. Available: https://respark.iitm.ac.in/about-us

14. IIMA Ventures, About IIMA Ventures, 2026. [Online]. Available: https://iimaventures.com/

15. IIMA Ventures, Incubation, 2026. [Online]. Available: https://iimaventures.com/incubation/

16. SINE, SINE IIT Bombay, 2026. [Online]. Available: https://www.sineiitb.org/

17. T-Hub, T-Hub, 2026. [Online]. Available: https://www.t-hub.co/

18. T-Hub, T-Hub and DST inaugurate Center of Excellence for AI, 2024. [Online]. Available: https://www.t-hub.co/news/t-hub-and-dst- inaugurate-center-of-excellence-for-ai

19. StartX, StartX accelerator, 2026. [Online]. Available: https://startx.com/

20. StartX, Frequently asked questions, 2026. [Online]. Available: https://web.startx.com/faqs

21. Y Combinator, What happens at YC, 2026. [Online]. Available: https://www.ycombinator.com/about

22. Y Combinator, The Y Combinator deal, 2026. [Online]. Available: https://www.ycombinator.com/deal

23. Massachusetts Institute of Technology, MITdesignX demo + pitch day 2026, May 21, 2026. [Online]. Available: https://calendar.mit.edu/event/ mitdesignx-demo-pitch-day-2026

24. TUM Venture Labs, TUM Venture Labs, 2026. [Online]. Available: https://www.tum-venture-labs.de/

25. TUM Venture Labs, Software & AI Lab, 2026. [Online]. Available: https://www.tum-venture-labs.de/labs/software-ai/

26. PACT, PACT participates in the kick-off of the DRIVEN project, Dec. 7, 2023. [Online]. Available: https://pact.pt/news/pact-participates-in-the- kick-off-of-the-driven-project/

27. IIT Madras Research Park, R&D collaboration, 2026. [Online].

    Available: https://respark.iitm.ac.in/R%26D-collaboration

28. T-Hub, T-Bridge: The international arm, 2026. [Online]. Available: https://www.t-hub.co/t-bridge

29. MITdesignX, MITdesignX, 2026. [Online]. Available: https://designx.mit.edu/

30. E. Tabassi, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1. National Institute of Standards and Technology, 2023, doi: 10.6028/NIST.AI.100-1.

31. UNESCO, Recommendation on the Ethics of Artificial Intelligence, 2021. [Online]. Available: https://unesdoc.unesco.org/ark:/48223/ pf0000381137