While technology alliances enable firms to achieve numerous benefits such as economies of scale and scope, reduction of risk and uncertainty, and access to new technologies and know-how, they often require these firms to cope with an inherent conflict between two competing objectives: the need to learn and the need to protect. This conflict stems from the fact that conditions necessary to facilitate the learning process simultaneously magnify the danger of losing core and proprietary knowledge. Previous research in the Social Network literature has shown how firms’ ‘social embeddedness’, which can be decomposed into relational, structural, and cognitive embeddednesses, can mitigate the intensity of this conflict via mechanisms such as mutual trust, familiarity, reputation, common systems of meaning, and network culture. However, reaching an ‘ideal’ level of social embeddedness is far from simple. It is for firms a long, hazardous and highly resource-consuming process. In this paper, our intent is to show that alternatives exist for firms that cannot benefit from favorable levels of embeddedness. In that case, firms may indeed try to enter an ‘intermediary-governed network’ and, hence, benefit from the mechanisms implemented by an intermediary entity to ease the learning process and protect against possible opportunistic behaviors. This network governance model implies that a separate entity is set up specifically to manage and coordinate the network and its activities. A common form of intermediary-governed network in technology industries is the government-sponsored R&D consortium like SEMATECH in the United States, EUREKA in Europe and the VLSI project in Japan. In this paper, we argue that entering an intermediary-governed network becomes highly valuable for firms when they cannot benefit from their social embeddedness to deal with the conflict between ‘trying to learn’ and ‘trying to protect’; in other words, when the relational embeddedness is low, the structural embeddedness is unfavorable, and the cognitive embeddedness is limited.
