In the rapidly evolving world of technology and open collaboration, companies face a paradox: how much control to maintain over the technologies they develop while providing access to foster innovation and attract external contributions. In their study, “Igniting Innovation: Evidence from PyTorch on Technology Control in Open Collaboration,” authors Daniel Yue, Assistant Professor at the Georgia Institute of Technology Scheller College of Business, and Frank Nagle, Assistant Professor of Business Administration Harvard Business School and Principal Investigator at the Digital Data Design Institute’s (D^3) Laboratory for Innovation Science, examine a pivotal governance shift in PyTorch, a leading machine learning framework. Their findings shed light on how governance models impact companies’ open-source participation, incentives, and strategies.
Key Insight: Governance Changes Shift Incentives Rather Than Create Them
“By changing from ‘dominant’ to ‘collective’ governance, PyTorch experienced an increase in contribution from External Companies, but a sharp decrease in contribution from the focal company (Meta).” [1]
The study’s primary focus is PyTorch’s shift from strategic and technical governance by Meta, to a collective model of governance via the independent, nonprofit Linux Foundation in 2022. Yue and Nagle found that when PyTorch transitioned to a collective governance model, the shift resulted in an increase in external company contributions of approximately 25.7%. However, this increase was largely offset by Meta’s reduced involvement. The authors concluded that changes in governance structure do not necessarily create new incentives, but rather redistribute existing incentives among participants.
Key Insight: Governance Changes Drive Different Levels of Participation
“[C]ontributions increased from Chip Manufacturers, who began or increased making technology-specific investments needed to create interoperability between PyTorch and their computer chips. However, there were no changes in contributions by Non-Chip Manufacturers (‘Application Developer’) firms, who contribute primarily to learn and to improve the usability of the technology for themselves.” [2]
Yue and Nagle’s research revealed that not all external firms respond equally to changes in project governance. Following the PyTorch transition, they found that chip manufacturers (such as NVIDIA, Intel, and AMD), which they classify as “complementors,” increased their contributions to PyTorch by 47.1%, significantly more than other types of firms, which they call “users,” including application developers (such as Microsoft, Amazon, and Alibaba) and cloud providers (such as OpenAI, Hugging Face, and Spotify).
The researchers attribute this difference to varying levels of dependency on the project. Chip manufacturers that rely on interoperability are more susceptible to potential “hold-up” problems if a single company controls the project’s direction. Application developers and cloud providers are less affected by the governance change because they primarily derive value from using the technology rather than integrating with it.
Key Insight: Control Rights and Strategic Openness Require Careful Consideration
“[S]hifting control rights to collective governance may not always increase total welfare because such a shift likely reduces the focal company’s incentive to contribute while simultaneously increasing the incentives of other companies to contribute.” [3]
The authors define “control rights” as the power to make decisions about how the technology develops in the future. Yue and Nagle suggest that control rights are one reason why companies open their technologies to external participants, especially in industries experiencing fast-paced innovation and where the ability to capture value is unclear, such as artificial intelligence and machine learning. Maintaining control rights can be an important advantage for certain companies, but they do not benefit from external innovations. Alternatively, giving up control rights through open collaboration can encourage growth.
Yue and Nagle define “open collaboration” as projects that involve more than two or three participants. Their research focused on “firm-sponsored” collaboration, where a single (focal) company releases a technology to spur innovation by other companies. The PyTorch case illustrates the potential downside of open collaboration, where the focal firm (Meta) found a reduced incentive to invest in the project. If a focal firm reduces its investment and cooperation with other participants, users can still benefit from access to the open-source technology interface, but complementors’ options become limited because they rely on interoperability with the technology, which might create a hold-up if it changes.
Why This Matters
This study suggests that companies should weigh the benefits of maintaining tight control over their open-source projects versus participating in collaborative governance models. For business executives and policy makers, this research offers insights into the strategic considerations surrounding collaboration and governance:
- For executives, starting open-source projects can help their companies obtain and increase control rights over future technology innovations. However, establishing control rights can be expensive. And going forward, retaining control rights can limit participation and collaboration with external companies.
- Policy makers must also recognize the trade-offs between retaining control rights and inviting open collaboration. Mandating open-source technology might not be appropriate in all cases. While regulations can help to prevent focal firms from taking advantage of control rights to increase prices after external participants invest, they should also consider the “overall welfare” of these projects, including incentives for focal companies to continue to participate.
References
[1] Daniel Yue and Frank Nagle, “Igniting Innovation: Evidence from PyTorch on Technology Control in Open Collaboration”, Harvard Business School Strategy Unit Working Paper No. 25-013 and Harvard Business Working Paper No. No. 25-013 (September 10, 2024): 1- 53, 21.
[2] Yue and Nagle, “Igniting Innovation: Evidence from PyTorch on Technology Control in Open Collaboration”, 3.
[3] Yue and Nagle, “Igniting Innovation: Evidence from PyTorch on Technology Control in Open Collaboration”, 3.
Meet The Authors
Daniel Yue is an Assistant Professor in the Information Technology Management area at Georgia Tech’s Scheller College of Business. His research explores why firms openly share innovative knowledge without directly profiting, a strategy called “open disclosure.” His projects use scientific publications and open source software in AI research as an empirical setting to develop and test new theories.
Frank Nagle is an Assistant Professor in the Strategy Unit at Harvard Business School, a faculty affiliate of the Digital, Data and Design (D^3) Institute at Harvard, the Managing the Future of Work Project, and LISH. He studies how competitors can collaborate on the creation of core technologies, while still competing on the products and services built on top of them. His research falls into the broader categories of the future of work, the economics of IT, and digital transformation and considers how technology is weakening firm boundaries.