In the current world of artificial intelligence (AI), protecting innovation is no longer a binary choice — it’s a balancing act. Algorithms, data sets, and training methodologies don’t fit neatly into traditional IP categories, yet they represent the crown jewels of modern enterprises. The question isn’t simply “Should we patent or keep it secret?” but rather “Which protection strategy best aligns with the lifecycle, risk profile, and commercial ambitions of this technology?”
Patents offer the power of exclusivity. They create enforceable rights that prevent competitors from using an invention — even if they arrive at it independently. For AI companies, this is invaluable because there is a high risk of parallel discovery. A novel neural network architecture, for instance, may be too fundamental to leave unprotected.
At the same time, there is an active debate around leveraging patents for protection because of the public disclosure required for patents that can allow competing AI models to get trained on the published literature. However, industries like pharmaceuticals or medical devices, rely on patents to secure regulatory approval and recoup R&D costs. By patenting, innovators secure not only legal leverage but also a tangible asset that can be licensed, sold, or used to attract investment. Despite the risk of public disclosure, patents remain relevant leaving the real question of not if but what to patent.
Trade secrets, on the other hand, thrive in the shadows. They protect what cannot, or should not, be disclosed: proprietary training data, fine‑tuning processes, or unique combinations of hyperparameters. Unlike patents, trade secrets don’t expire. But they do demand vigilance. Once secrecy is lost, so is protection. For fast‑moving AI fields where algorithms evolve rapidly, trade secrets often provide more practical, flexible coverage than the lengthy patent process by providing indefinite protection as long as secrecy is sustained.
The pace and modular nature of AI innovation are increasingly straining traditional IP frameworks, which were not structured to protect rapidly evolving intangible innovations like algorithms, data pipelines, and training methodologies. As development cycles accelerate, AI systems now consist of numerous interdependent and novel components, making it difficult to protect every layer through conventional IP mechanisms. As a result, we’re seeing a cultural shift among developers as they lean into collaboration through open-sourcing. Organizations are therefore forced to make strategic decisions about which elements justify formal protection and which are better leveraged for speed and adoption. By leaning into open‑source approaches, companies like Nvidia and Microsoft have been able to reduce friction, accelerate ecosystem growth, and encourage collaboration, while still being able to selectively apply patents and trade secrets to safeguard their competitive advantage.
Nvidia has released open-source AI Toolkits NeMo for generative AI, and their open-source libraries and integrations ensure Nvidia Graphics Processing Units (GPUs) remain the default hardware for AI workloads. While the toolkits and frameworks are open, CUDA (Compute Unified Device Architecture) itself remains proprietary, anchoring Nvidia’s competitive edge. Nvidia patents hardware innovations (almost doubling their global annual patent filings every two years since 2019[1]) and keeps GPU architectures as trade secrets, while open-sourcing software layers to drive adoption.
In a nutshell, Nvidia uses open-source to expand its AI ecosystem, but proprietary hardware (trade secrets) and patents protect its core business model.
Microsoft has accelerated their open-source adoption with the acquisition of GitHub in 2018 and the integration of Linux, Kubernetes and other open-source technologies deeply into Azure. These moves have made it a cloud platform that is friendly to open-source workloads. At the same time, they have been building a strong patent portfolio in cloud, AI and enterprise software, making up more than50% of their annual filings in the past decade. While open-sourcing the foundational layer, Microsoft has kept premium enterprise features proprietary through trade secrets for sensitive algorithms. By releasing products under permissive licenses, they have encouraged adoption while retaining commercial rights to proprietary extensions.
These examples show that open-source and IP protection are not opposites nor binary, they complement each other. Open-source builds trust and accelerates adoption, while patents secure exclusivity for high-value technologies. This balance is exactly the kind of hybrid IP strategy AI and any tech-related enterprise should explore. The examples above demonstrate how companies can thrive by being open enough to win developer trust, yet protective enough to secure their innovations.
With the recent EU AI act creating higher regulatory scrutiny and introducing disclosure obligations (requires system documentation for high-risk AI that may reveal trade-secret protected expertise), it is reducing the viability of maintaining trade secrets, hereby, incentivizing patents. The number of patent filings will increase to secure enforceable rights in disclosed innovations and help companies defend against infringement suits and negotiate cross‑licensing deals.
The next decade will redefine how IP is protected in the AI industry. As algorithms, models, and synthetic data become the lifeblood of innovation, companies will need to move beyond traditional frameworks and adopt hybrid strategies that balance openness with protection.
- Companies will follow a layered approach to both maximize and follow secrecy by patenting foundational inventions but keeping surrounding processes, training data, and refinements as trade secrets.
- More companies will release toolkits, frameworks, and models into the open-source community to accelerate adoption. The real value will lie in proprietary extensions, enterprise services, and hardware integration – following the paths of Microsoft and Nvidia.
- As regulatory frameworks advance, companies will have to innovate in licensing and compliance models to balance competitiveness with transparency.
- Policy makers will continue to be challenged to define ownership and enforceability as AI dons the inventor hat and the usage of ‘digital twins’ and synthetic data gain popularity.
Companies contributing to AI development need to build flexible portfolios that combine patents, trade secrets, and open-source while also hedging their bets on the policymakers’ and regulators’ decisions. Those portfolios need to be leveraged as strategic business assets, not just as legal tools, to gain partnerships and lead the digital ecosystem.
The companies that thrive in the next decade will be those that master the art of duality — open enough to accelerate innovation, yet protective enough to secure their competitive edge. Intellectual property in AI will no longer be about choosing between patents or trade secrets, but about orchestrating a symphony of strategies that together form a dynamic shield for the ideas of tomorrow.
[1] As per Nvidia’s published patent portfolio
