The interactive effects of knowledge elements and collaboration networks on exploratory innovation performance: evidence from the Chinese artificial intelligence industry

Why this matters for the future of AI companies

Behind every breakthrough in artificial intelligence lies a mix of what firms already know and whom they work with. This study looks inside China’s fast-moving AI industry to ask a simple but crucial question: how should companies combine their internal know‑how with their external partnerships to create genuinely new ideas, not just small tweaks? By analyzing thousands of patents with modern data tools, the authors uncover patterns that can help managers and policymakers steer AI innovation more intelligently.

Three kinds of AI innovators

Using patent data from 260 Chinese AI firms, the researchers first mapped two things for each company: the variety and structure of its technical knowledge, and the shape of its collaboration network built through co‑patenting. They then applied a clustering method that groups firms with similar profiles. This revealed three broad types. “Collaboration‑oriented” firms are deeply embedded in dense partner networks but have only moderate in‑house knowledge strengths. “Knowledge‑oriented” firms are rich in diverse and specialized know‑how but relatively isolated. “Balanced” firms fall in between, with neither strong advantages nor glaring weaknesses in either area.

How knowledge mix and partnerships interact

The study then used a decision‑tree algorithm to trace how different combinations of knowledge and network features relate to firms’ ability to produce patents in new technological areas—a practical gauge of exploratory innovation. Across all groups, the structure of internal knowledge played the leading role, but the surrounding network could either amplify or soften its effects. For collaboration‑oriented firms, having too broad a spread of technical fields often hurt performance by overwhelming their capacity to absorb and use information. Yet, when these firms also had wide or tightly knit collaboration networks, partners helped them filter, share, and integrate knowledge, turning potential overload into useful novelty.

Too much specialization can backfire

Knowledge‑oriented firms told a different story. Their deep and varied expertise did not automatically translate into cutting‑edge breakthroughs. When their knowledge base became overly diverse, innovation performance actually dropped, likely because attention and resources were spread too thin. Even when diversity was kept in check, partnering with many organizations was not always better. A moderate number of collaborators tended to work best, while very broad collaboration brought coordination costs and distractions, and very narrow collaboration limited exposure to fresh ideas. This suggests that highly specialized AI firms need to be deliberate in trimming their knowledge portfolio and curating a manageable set of strategic partners.

Finding the sweet spot between similarity and difference

For balanced firms, the key levers were how closely their knowledge pieces fit together and how easily one skill could stand in for another. When knowledge elements were too perfectly matched, the firm became locked into narrow paths, making it harder to jump to new areas. However, when there was enough overlap—so that one technique could substitute for another—firms were better able to experiment, pivot, and respond to uncertainty in the young and volatile AI sector. In other words, some redundancy in know‑how, often seen as wasteful, can actually provide flexibility and resilience when technologies and markets are shifting quickly.

What this means for AI strategy

Overall, the study shows that neither “more knowledge” nor “more partners” automatically leads to better exploratory innovation. What matters is the fit between a firm’s internal knowledge mix and the way it builds and uses its collaboration network, and this fit looks different for collaboration‑oriented, knowledge‑oriented, and balanced firms. For managers, the message is to treat knowledge and partnerships as a joint design problem: avoid unchecked complexity, seek partners that complement specific weaknesses, and maintain enough overlapping skills to adapt when the AI landscape changes. For policymakers, the findings highlight the value of ecosystems and industry platforms that help firms reorganize their knowledge and form targeted partnerships, rather than simply pushing for more R&D spending or more alliances.

Citation: Zhang, L., Chen, J., Qiu, H. et al. The interactive effects of knowledge elements and collaboration networks on exploratory innovation performance: evidence from the Chinese artificial intelligence industry. Humanit Soc Sci Commun 13, 303 (2026). https://doi.org/10.1057/s41599-026-06637-x

Keywords: exploratory innovation, artificial intelligence firms, collaboration networks, knowledge management, patent analysis