Google Dethrones IBM in AI Patents — And Why OpenAI, Creator of ChatGPT, Barely Has Any

In the 2025 generative AI patent rankings, Google has toppled IBM — which held the title for over two decades — to claim the top position. In the specific category of agentic AI, NVIDIA has pulled level with Google for first place. Meanwhile, a startling fact has come to light: OpenAI, the company that shook the world with ChatGPT, holds a mere 110 patents, placing it outside the top 10 entirely. By comparing these numbers with the AI patent strategies of Japanese companies like Sony and Toyota, this article delves into the fundamental question of what it truly means to build an AI patent portfolio.

1. The Big Picture — Who Holds What

As of late 2025, the top rankings for generative AI-related patents (including applications, publications, and registrations) reveal a striking landscape.

Google leads with approximately 40,000 generative AI-related patent applications and registrations, broadly covering the foundations of AI technology: applications of transformer architectures, learning efficiency improvements for large language models (LLMs), and perceptual representations in multimodal AI. Microsoft follows in second with roughly 30,000, building its own patent portfolio in parallel with its OpenAI investment. IBM — long known as the “emperor” of AI patents — comes in third with approximately 25,000, yielding the top spot to Google for the first time. Amazon (primarily inference and deployment patents through AWS) and Meta (natural language processing and multimodal work related to the LLaMA series) round out the top five.

The most dramatic rise belongs to NVIDIA. Traditionally concentrated in GPU hardware and CUDA-related patents, NVIDIA dramatically accelerated its filings in agentic AI — multi-agent systems where multiple AIs autonomously collaborate to solve problems — throughout 2024 and 2025, pulling level with Google in that specific category. The strategy of building IP at the software and systems level, not just in AI semiconductors, is unmistakable.

And OpenAI? Roughly 110 patents. A tiny fraction of what IBM or Google holds. Despite being the recognized leader in generative AI by brand and capability, its IP portfolio is strikingly thin — a paradox that is generating serious discussion across the industry.

2. Why Google Reached the Top — Anatomy of an “AI Patent Factory”

Google’s accumulation of AI patents rests on several structural foundations.

First, the integration of Google Brain and DeepMind. When the two organizations merged to form Google DeepMind in 2023, they brought together the world’s leading AI research teams, both of which had been actively patenting their discoveries for years. AlphaGo, AlphaFold, Gemini, and the underlying technologies — specific reinforcement learning algorithms, deep learning applications for protein structure prediction, training methods for large-scale multimodal models — form the core of this patent portfolio.

Second, a culture of filing patents before products launch. Google has institutionalized the habit of filing related patents before product releases. Engineering teams and patent teams collaborate routinely, with invention disclosure processes in place to capture discoveries before they become papers. This culture of treating patent filing as routine has propelled Google to the top of AI patent ownership.

Third, alignment between litigation and licensing strategy. Having navigated major IP disputes including Oracle’s Java patent lawsuit, Google deeply understands the value of patents as defensive and negotiating assets. In generative AI, it is making anticipatory filings in preparation for future competitive scenarios and standardization processes.

3. Why Did IBM Lose the Top Spot?

IBM maintained the top spot in U.S. annual patent registrations for over 25 consecutive years from 1993. Yet in the generative AI era, IBM was comparatively slower to respond to the new wave. Much of IBM’s AI patent portfolio is concentrated in classical AI, machine learning, and enterprise automation — the transition to generative AI, LLMs, and transformer architectures was not as swift as Google’s or Microsoft’s. IBM’s shift toward the Red Hat acquisition (2019) and cloud computing may also have relatively diluted its focus on core AI research.

That said, IBM has not “lost” at AI. In quantum computing, enterprise AI, and hybrid cloud, it maintains a powerful patent portfolio. The accurate assessment is that IBM ceded the top position in the specific subset of generative AI patents to Google.

4. Why OpenAI Has Almost No Patents — A Three-Layer Structure

The most fascinating case is OpenAI. Having produced ChatGPT, GPT-4, and DALL-E — the AI products with the greatest social impact in recent years — why does it have almost no patents?

Reason 1: Founding Philosophy — “Knowledge as a Public Good”

OpenAI was founded in 2015 as a nonprofit research organization committed to “ensuring that the benefits of AI accrue to all of humanity.” At its core was the philosophy of openly sharing knowledge. Prioritizing knowledge circulation through published papers and open-source code over exclusionary patent rights meant that a culture of proactive patent filing never developed within the organization.

Reason 2: The Section 101 Barrier

OpenAI’s core technologies — the transformer variant architectures underlying the GPT series, prompt engineering methods, and RLHF — face significant challenges under the U.S. Patent Act’s Section 101 “abstract ideas” problem. Since these inventions center on algorithms, mathematical methods, and data processing flows, asserting patent eligibility under the Alice/Mayo framework has been difficult. With high rejection rates, the cost-benefit ratio of filing applications was unfavorable.

Reason 3: Speed and Openness Over IP Accumulation

Patent prosecution takes time. In OpenAI’s rapidly evolving technological landscape, whether a given patent will provide competitive advantage a year later is uncertain. Rapid productization and market deployment to secure first-mover advantage — using brand and raw model capability as competitive weapons — was the chosen strategy. But this strategy is showing its limits. As Microsoft has been building its own Azure-related AI patents alongside its OpenAI investment, OpenAI deepens its Microsoft dependency while remaining effectively defenseless in IP terms.

5. Sony and Toyota’s AI Patent Strategies — Japanese IP Realism

Japanese companies present a distinctive form of IP realism in their AI patent strategies.

Sony has steadily built its patent portfolio around AI technologies related to imaging, video, and audio. Particularly notable is the “edge AI sensor” technology that integrates CMOS image sensing and AI inference processing on the same chip. Inventions that deeply integrate hardware and AI are easier to clear Section 101 and create barriers to entry that competitors cannot easily replicate. Sony’s core AI patent strategy is to file “AI integrated into physical devices and systems” rather than AI as standalone software.

Toyota is building AI patents on two pillars: autonomous driving AI and manufacturing process optimization AI. In autonomous driving, it files deep patents at each layer of prediction, judgment, and control — particularly “decision-making algorithms in uncertain environments” and “real-time processing of sensor fusion.” In manufacturing, “predictive maintenance AI for production lines” and “image recognition for quality inspection” are major patent assets. What distinguishes Toyota is that it uses patent filings not just for defensive purposes but as tools for participation in standardization efforts — bringing its patents to the ISO, SAE, and AI safety/explainability standardization processes to influence the formation of industry standards.

6. AI Patent Strategy in the Generative AI Era — Optimal Approaches by Company Size

Looking at the AI patent rankings, one conclusion stands out: there is no single right answer. Companies like Google and IBM can pursue strategies of overwhelming competitors through sheer volume. But not every company needs to.

For mid-sized technology companies, the optimal approach is a “depth-and-breadth strategy” — combining deep patents in core technologies with broader patents in adjacent areas. Building a patent “fortress” in the specific sub-field where one has the greatest competitive advantage, and using it simultaneously for licensing revenue, litigation defense, and standards participation, is the most efficient path. For startups, bringing in IP professionals early and instilling a culture of invention disclosure from the outset is the best defense against future patent disputes.

OpenAI’s story illustrates the risk of becoming “IP defenseless despite world-class technology.” At the same time, patents are not a panacea — the cost, time, and reputational risk of patent litigation are real. AI-era IP strategy demands a portfolio approach combining patents, trade secrets, standardization participation, and open-source strategy.

Conclusion

Google’s ascent to the top of AI patents, displacing IBM, symbolizes a structural shift in intellectual property competition in the generative AI era. NVIDIA’s rapid rise in the agentic AI category is equally significant. OpenAI’s unusual position as a major AI player with almost no patents is both a result of strategic choices and a product of organizational design, legal environment, and the nature of its technologies. Japanese companies like Sony and Toyota, with their distinctive strengths in hardware integration and deep manufacturing process ties, have the potential to maintain a significant presence in the AI IP race. The map of AI patents is being redrawn at speed. The ability to read that change accurately is the defining trait of companies that will prevail in the next generation of technology competition.