The subject studied in this work is complex and extensive. Clusters are a form of organization of economic activity which have sprung up in many countries, responding to the significance of failures in the market and to the often limited capacity of market signals (in the traditional sense of the term) to orient technico-economic development.

Innovation today has a pronounced interactive bent. It involves the intervention of multiple actors, between whom collaborations may be organized, develop and break down in accordance with the problems involved and the skills required. The increasing importance of inter-organizational collaboration, notably between private companies and university institutions, confirms the central role of networks in innovation processes. Innovation and production ecosystems appear to respond to this need. We also need to recognize the shift that has taken place in the respective responsibilities of actors involved in innovation processes.

As the MIT report [MIT 13] highlights, public policies draw legitimacy from the fact that big businesses have, in the past, provided public goods for their wider industrial environments in the form of research spillover, training and the diffusion of new technologies to their suppliers, while exerting pressure on the state and regional authorities to invest in infrastructure. This spillover provides complementary capacities which may be used by other producers in the region, even if they did not participate in their creation. In the cases where these complementary capacities disappear, through the effects of globalization and offshoring, “large holes in the industrial ecosystem have appeared” [MIT 13]. Public cluster policies must aim to create (or recreate) and support productive complementarities in order to foster projects which create growth, jobs and innovation.

This does not mean that initiatives can only result from public policies. The examples of certain American and European clusters show that the most dynamic organizations are the result of entrepreneurial initiatives. However, certain public policies currently in strength in the USA follow a framework of broad transversal themes. The NIST initiative, aimed at bridging the gap between fundamental research and production, stresses the importance of not pre-determining a specific domain for each ecosystem. Proposals may concern any project touching on industrial robotics and bio-pharmaceutical manufacturing. In Europe, the focus on smart specialization in official EC documentation indicates a notable shift in favor of “entrepreneurial discoveries” and the technological and productive complementarities which they generate. Entrepreneurial discoveries form the springboard for vertical public policies, which may be oriented toward the constitution of clusters with no defined geographical boundaries. Note too, in this context, that public policies have a stronger effect for “established clusters that have emerged in the market process than for emerging clusters” [COM 13].

From this perspective, our work has involved synthesizing materials concerning the fact of treating clusters as possible bases for action on the basis of existing experience. It seems highly improbable that the success of a location depends on the adoption of a model, notably that of Silicon Valley. As we have seen, the logic of differentiation wins out over that of homogenization in clusters, even within the same activity. The innovation and production ecosystem in Silicon Valley has grown up around a set of institutions and partnerships, developed over time. Local learning effects create a path dependency, reflecting specific forms of organization both within the cluster and with other clusters. As Feldman provocatively states [FEL 14a], we do not know where Silicon Valley really is. It does not exist on any conventional map and its borders are flexible, moving in line with major transport routes and with the arrival of new companies which results from technological surges (for example, the arrival of car manufacturers developing new concepts, such as electric and self-driving vehicles). In other terms, geographic limitations challenge administrative and political borders, following the connections established between companies. For Feldman, there is a significant gap between the theoretical definition of a location as a continuous, integrated geographic unit and the official statistical divisions used in providing data to public authorities. This division may conceal cross-border activities, making it hard to understand the micro-geography of a location, or the neighborhood relationships which are so important for innovation. This means that public action is more difficult.

Organized spaces are not administrative spaces but spaces within which interactions occur and result in the creation of specific resources. In the USA, the critical element relates to knowledge transfer and the mobility of skills. Transfers occur through multiple channels and those which take place between companies rely on the existence of dynamic labor markets and structured networks of qualified personnel, markets and networks which are mutually sustaining. Network membership enables the identification of “areas of need” for specific skills (mobility between companies) and of interstices in the economic fabric which may be exploited (through company creation).

More generally, we have seen that the grouping of companies in specific locations is not a new phenomenon. Companies have long attempted to internalize economic benefits, including reduced costs and increased productivity, within “natural” clusters. Economic geography shows that the geographical concentration of businesses generates positive externalities for those involved. However, it would be overly simplistic to dimension ecosystems following a single axis of polarization, i.e. savings through agglomeration. Cost effects are found alongside knowledge spillovers; face-to-face contact enables the transfer of tacit knowledge, standards are diffused and social networks develop. All in all, the network effect is more significant than the agglomeration effect in understanding the development of certain American and European clusters, particularly the Cambridge cluster.

Furthermore, the notion of a cluster is richer than that of spontaneous groupings of companies. More actors are involved (small and large companies, universities, research centers, training institutes, etc.) and the observed behaviors show that companies have a choice of two options: either to exploit the potential economic benefits of their situation as quickly as possible or to invest in physical, human or institutional assets to create resources for future growth. In this context, political decision-makers and experts face a problem of endogeneity. The dynamics of these forms of organization are autonomous and self-sustaining. Hence, companies and institutions on the one hand and collective resources on the other hand develop alongside one another, forming ecosystems which are products of prior actions, resulting from a process of co-development for which, according to Feldman, “causality is difficult to attribute”. This is particularly true in terms of performance evaluation: the interweaving of multiple factors makes it very difficult to measure the net effect of clusters (see, notably, our remarks on competitiveness poles).

We have considered the co-evolution process along the same lines as a process of institutionalization rather than as a technological process based on a single and cumulative trajectory and on a more or less developed division of labor. Depending on the choices which are made and the behaviors which occur, the development of an innovation and production ecosystem legitimizes a form of collective organization within which participants may learn, develop technological pathways with a certain level of variation, establish practices and construct rules, all in the aim of creating novelty. Additionally, a common language is created, norms and values are shared and, finally, a system of meanings emerges. Moving beyond purely financial incentives, the relational aspect is predominant in explaining the creation of shared resources (R&D, engineering, tests and certification, manufacturing capacities, etc.) and in the creation of collective returns above and beyond the sum of individual contributions, an emerging property of these organizational forms once the point of sustainability has been reached.

To conclude, note that a cluster becomes sustainable when its organization satisfies the latent or evident requirements of demand. The relationship between the organization and the market can only be maintained through a dynamic of innovation. Innovative clusters currently act as vectors of structural modification: in many sectors and activities, they constitute the collective infrastructures of the innovation process. Within this framework, cluster sustainability depends on the sustainability of innovation itself [CHA 13]; the enabling factors of innovation are based on collaborative R&D, human capital, social networks and interactions between actors, i.e. the capacity to maintain the flow of innovation flow for present and future applications via necessary modifications to the overall organization.

These remarks imply that a cluster-based approach should be more or less focused on increasing competitiveness at the regional level, via collective control of resources and agglomeration savings. It may be structured as a conceptual reading grid, enabling analysis of the dynamics of knowledge creation and innovation within organized mechanisms [MAL 06]. This conceptual restriction, based on the idea that very few truly innovative clusters exist in reality, would enable a clearer definition of the field of theoretical and empirical investigation.