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

Key points

• A single European patent application costs €13,000 to €18,000 (roughly $14,636 to $20,266) once attorney fees are included, a sum most seed-stage robotics startups can't easily absorb.
• Boston Dynamics sued Ghost Robotics in 2022 over "core technology" controlling how a robot recovers from a fall, illustrating that the most valuable IP is often about robot behaviour, not hardware.
• The two companies resolved that dispute in 2025, but the legal costs underline why small startups rarely win drawn-out patent fights.
• AI-assisted patent drafting can cut filing time significantly, according to Lightbringer, the IP firm whose co-founder authored the original analysis in The Robot Report.

Robotics startups spend years perfecting a robot's movement and software. The thing that could actually sink them is a patent filing they never made.

That's the warning from Dominic Davies, CEO of patent firm Lightbringer, writing in The Robot Report. Davies has filed more than 300 patents across industrial automation over a 20-year career. His core argument is uncomfortable: the companies that lose IP battles are almost always the ones that waited too long to build a strategy.

What does "physical AI" IP actually cover?

Physical AI means AI that operates in the real world through robots and automated machines, rather than AI that just answers questions on a screen. The intellectual property, or legal ownership of inventions, that matters in this space is surprisingly broad. It spans hardware, AI models and calibration methods, and it covers behaviour too.

The Boston Dynamics versus Ghost Robotics case is the clearest example. Boston Dynamics' 2022 claim wasn't about the shape of the robot's legs. It was about the patented method for recovering when the robot fell over. That one behavioural invention was central enough to trigger a lawsuit. We first covered Boston Dynamics on 16 July 2026, and the company has featured in four AI2Day stories since.

Swarm robotics, where dozens of machines coordinate to do things no single machine could, makes this even murkier. Patent offices including the European Patent Office have signalled that claims may need to define individual machines, the overall system, or both together, and the law hasn't fully settled this yet.

Each development that speeds up robot deployment, from Montreal labs training robots on live factory data to construction machines preventing costly on-site errors, also shortens the window a startup has to lock down its key inventions.

What should a startup actually do?

Focus on two inventions, not twenty. Davies' advice, drawn from work with clients including London food-assembly startup KAIKAKU, is to identify the one or two mechanisms a competitor genuinely can't copy without infringing. For KAIKAKU, it wasn't moving a food bowl at all; it was moving it fast without spilling. That precise motion mechanism became the company's "crown jewel" patent.

Around those crown jewels, a startup builds a defensive ring: patents on surrounding architectures that would otherwise let a rival design around the core invention. The goal is to make copying expensive, not just illegal.

Some inventions should never be patented. A patent is a public document, so a calibration method that competitors can't reverse-engineer anyway is often better kept as a trade secret.

On cost: the numbers are steep. A single European application runs €13,000 to €18,000 before prosecution fees, and that multiplies fast across a crown-jewel portfolio. AI-assisted drafting, where software handles documentation and a qualified attorney reviews the output, can bring those costs within reach for a seed-stage team. That gap between "we can afford this" and "we can't" is the practical reason an IP strategy needs to start early.

The honest read here is that this space is about to get litigious the way smartphones did in the 2010s. A robotics startup that hasn't thought hard about its two or three indispensable inventions before it ships its first product is taking a risk that a well-funded competitor will close.

Common questions

Do you always need a patent, or can you just keep things secret?

Neither approach fits every situation. A patent makes sense when an invention is visible in the finished product and a competitor could copy it by studying what you shipped. Keeping something secret makes more sense for internal processes or calibration techniques that are hard to reverse-engineer from the outside.

Why can't a small startup just defend a patent if someone infringes it?

Patent litigation is expensive enough that even valid patents get abandoned rather than defended. Davies notes that in ambiguous IP disputes, startups rarely have the budget to fight a well-funded opponent to a conclusion, which is exactly why building a clear, layered portfolio before a conflict arises matters more than filing lots of patents after one starts.