From Training Data to Patent Thickets: Generative AI, Patent Concentration and Competition in the AI Economy

Abstract

Patent filing in generative artificial intelligence has moved from a niche activity to one of the fastest-growing segments of the global patent system. The World Intellectual Property Organisation's 2026 update records that published generative-AI patent families rose from roughly 14,000 in 2023 to over 37,800 in 2025, with more than 56,000 families published across 2024 and 2025 combined, a volume exceeding the entire preceding decade. Ownership is concentrated: China accounts for the largest share, SoftBank has become the single largest corporate holder with almost 3,000 families, and a small set of conglomerates dominate the remaining top ranks. This concentration, layered onto an already dense set of overlapping claims on model architectures, training methods and specialised hardware, has revived a long-standing competition-law concern: the patent thicket. The April 2026 launch of the Shared AI License (SAIL) Foundation by Anthropic, IBM, Meta, Microsoft and Genentech, pooling more than 33,000 patent families, is the clearest institutional response to date, but its scope is narrower and its internal distribution of bargaining power more skewed than the coverage it has received suggests. This paper examines whether Indian law is equipped to respond. It reviews the Patents Act 1970 and the Competition Act 2002, situates them against the compulsory-licensing, FRAND and essential-facilities doctrines developed in the United States, the European Union, the United Kingdom and China, and incorporates CCI's October 2025 market study on AI and the Supreme Court's September 2025 order in the Ericsson/Monsanto litigation, which narrowed rather than expanded the Commission's jurisdiction over patent-licensing conduct. The paper argues that India's principal statutory tool against patent aggregation, the compulsory licence under Chapter XVI of the Patents Act, is structurally unsuited to a thicket problem, and that competition-law intervention is now a weaker prospect than earlier commentary assumed. It closes with a set of remedies assessed on their actual, rather than assumed, capacity to work.