Generative AI Patents Skyrocket, Leaving Decentralized Projects Vulnerable

The World Intellectual Property Organization (WIPO) has released a patent landscape report that reveals generative AI patents have nearly tripled in two years, from around 14,000 in 2023 to over 37,800 by 2025. This surge in patent activity far surpasses the entire preceding decade combined.

China is dominating the field, with more than 38,000 GenAI inventions between 2014 and 2023. In contrast, the United States has accumulated roughly one-sixth of that number during the same period. Top patent holders include Chinese corporations such as SoftBank, Tencent, Ping An Insurance Group, Baidu, and the Chinese Academy of Sciences.

Large language models (LLMs) have surpassed generative adversarial networks (GANs) as the most patented technology category in GenAI. Image and video outputs continue to dominate applications, but LLMs are now driving innovation behind the scenes. This shift is attributed to the 2017 transformer architecture paper, which introduced the attention mechanism that underpins modern LLMs.

As a result of this rapid consolidation around corporate giants and state-backed institutions, decentralized AI projects face significant challenges. These projects rely on open-source model architectures, which may now be vulnerable to patent claims they didn't anticipate. This creates an asymmetry between large enterprises, which can absorb litigation costs as routine business expenses, and decentralized autonomous organizations (DAOs), which cannot.