One number exists. The other doesn't.

I tried to find one number for a post I was writing last month. The number was: how much venture capital went into robot safety infrastructure in 2025?

I couldn't find it. Not in Crunchbase. Not in PitchBook. Not in CB Insights. Not in the OECD's AI VC report from February. Not in any of the specialist robotics trackers. The category does not exist.

Here's what does exist.

The tracked number

Venture funding into robotics startups in 2025: approximately $8 billion, according to Crunchbase sector data.

This number is tracked, disclosed, audited, and published. Crunchbase has it broken out by sub-sector (industrial, humanoid, service, logistics). PitchBook has it by stage. The OECD aggregates it alongside other AI-adjacent categories. The figure is a real number you can defend in a meeting.

It includes funding for robot autonomy. Robot perception. Robot foundation models. Robot manipulation. Robot fleet orchestration. Robot simulation environments. Robot data labelling.

It does not, as far as I can determine, have a line item for "safety infrastructure."

The untracked number

Venture funding into robot safety infrastructure in 2025: unknown.

Not "low." Not "small." Unknown. The category isn't broken out in any tracker I can find.

I'm not saying the funding doesn't exist. Some safety-adjacent capital is almost certainly hiding inside the $8B — integrated into platform rounds, folded into industrial-automation deals, buried inside humanoid pre-seeds. But if you try to isolate the figure, you can't, because the taxonomy doesn't include it as a category.

Why this matters

In market analysis, an untracked category usually means one of three things.

(1) The activity isn't happening. I don't believe this is the case here. Several Physical AI companies are doing real work on independent safety architectures, some publicly, most not.

(2) The activity is happening but scattered across other categories. This is probably most of it. Safety work is getting funded as part of larger platform rounds, not as a standalone thesis.

(3) The category hasn't been named yet. This is the one I think is dominant, and it's the structural point worth making.

When a venture category isn't named, it's hard to invest in. It's hard to benchmark. It's hard to build press around. It's hard to attract specialist founders. And most importantly, it's hard for companies in the category to find each other — because they're all describing what they do in terms borrowed from whichever parent category they happen to be nearest to.

The absence of data is the data

Almost every piece of due-diligence math I can do in the robotics sector — market size, growth rate, CAC, unit economics, retention — has a citable number behind it. The one number I need, for the sub-sector that regulation is about to make mandatory, doesn't exist.

The absence of that data point is not a gap in someone else's analysis. It's the point.

A category that needs to exist for a coherent safety case under EU Machinery 2023/1230 is not a line item in the venture trackers that are supposed to allocate capital to emerging technical needs. That's a structural mispricing, and it's the kind of mispricing that resolves abruptly rather than gradually.

I'll update this post if the 2026 data from Crunchbase, PitchBook, or the OECD starts breaking out safety infrastructure as a separate figure. Until then, the most honest version of the chart is the one with a question mark on one side.