Physical AI Race Will Be Won in the Patent Office, Not the Lab

AI, machine learning and automation have given robots abilities that seemed extraordinary a decade ago. As competition intensifies, the physical AI race is less likely to be won by whoever builds the best robot, and more likely to be won by whoever owns the technology that determines its behavior.

That intellectual property landscape is complex because innovations can span hardware, software, AI models and calibration methods. The most valuable invention may not be the robot itself, but the technology that enables it to behave in a particular way.

Dominic Davies, a patent attorney with 20 years of experience and more than 300 industrial automation patents, sees a consistent pattern: startups that delay lose out. Robotics founders must decide what to patent, what to keep secret, and how to protect core technologies before competitors do.

Sometimes the robot is not the innovation; its behavior is. Boston Dynamics’ 2022 lawsuit against Ghost Robotics over “core technology” that dictated how a robot recovered from a fall illustrates this point. The dispute was resolved in 2025.

Swarm robotics faces similar challenges, because innovation often arises from interactions among dozens of machines. That raises a legal question: can a company patent the behavior of the swarm, or only the machines that produce it?

Patent offices have indicated that claims may need to define individual entities and/or the overall system, including how entities interact. Because litigation is expensive, startups rarely have the upper hand in ambiguous IP scenarios. A strong IP strategy must be deployed from the outset, not after Series A.

A strong IP strategy does not necessarily mean more patents. A robust patent portfolio should be the foundation, but startups should not patent every component. RoboSense’s IP head warned last year that robotics patent disputes are likely to multiply, as they did in the smartphone industry in the 2010s.

Not every invention should be published. A patent is a publication, and for some innovations that is the last thing to do. Manufacturing processes or calibration techniques that are difficult to reverse-engineer from the finished robot may be better kept confidential.

By contrast, patents make more sense when an invention is embodied in the product and could be reverse-engineered. A novel robotic mechanism that makes a robot faster, more precise or more efficient would be hard to keep secret once the robot is on the market.

When building a patent portfolio, startups should not try to patent an entire machine. They should identify one or two novel mechanisms, architectures or processes that are central enough to the robot’s function that competitors would find it genuinely difficult to copy the robot without them.

London-based food-assembly robotics startup KAIKAKU used this approach. The problem it solved was never simply moving a bowl from one place to another; it was moving it at speed without making a mess.

That behavior came from a core motion mechanism that was novel and central enough to become one of KAIKAKU’s “crown jewel” patents.

Once crown-jewel inventions are secured, startups can build a wider defensive ecosystem around the architectures that make the core innovation commercially useful. A few patents protect the advantage itself; a wider ring of patents makes designing around it expensive enough that most competitors will not try.

That layered approach can strengthen protection and create leverage in licensing, investment, partnerships and potential acquisition talks.

Cost remains a barrier. A single European patent application, properly drafted and prosecuted, can cost €13,000 to €18,000 or more once attorney fees are included. AI can help by handling documentation, analysis and drafting, while a qualified patent attorney reviews and takes responsibility for filings. Filing can be up to 70% faster, and attorney time goes to judgment calls.

The physical AI race will not be won by the company with the most patents. Patents alone will not get a robot to market. But when a company’s advantage rests on a handful of hard-to-replicate mechanisms, leaving them unprotected is not an option.

The right approach is strategic rather than quantitative: identify crown-jewel technologies, decide which should be patented and which should remain secret, and build a defensive ecosystem around them. AI-native patent platforms can streamline cost and complexity.

For startups, value lies not just in filing more patents, but in spending limited resources on the innovations that matter most.

Dominic Davies is co-founder and CEO of Lightbringer. A former software engineer turned UK and European patent attorney, he previously founded Invent Horizon IP and is co-founder and investment manager at Immetric.