123 companies, $12B funded — inside the Israeli Physical AI ecosystem

In April 2026 the Israel Innovation Authority (IIA) published its AI Strategy Report. Inside it — conservatively — 123 Israeli companies working on Physical AI, with more than $12 billion in cumulative venture funding tracked across them. The report is freely available. I've read it back-to-back more than once and think it deserves more attention than the local trade press has given it.

What the IIA found

The report's own numbers, taken directly from page 27.

Computer vision and perception · 61 companies. This is the single largest concentration in the Physical AI space — roughly half of the 123 total. Cumulative investment in this sub-segment: approximately $4 billion. The IIA counts 2,347 patents issued to these companies.

Edge computing · 50+ companies. The IIA breaks this further into roughly 20 companies working on edge hardware and processors, 25 on edge sensors, and 10 on edge software. There is overlap with the CV companies — some of the CV players own stack on both sides.

Defense-adjacent AI. The IIA describes this as a significant category but does not publish a specific company count for it in the April report.

Named companies cited in the IIA report include Mobileye, Innoviz, Arbe, Vayyar, Hailo, and Deci (the latter acquired by Nvidia). The IIA notes Mobileye's platform is deployed in more than 230 million vehicles worldwide, and singles out Hailo's AI processor for operating at about 2.5 watts — the kind of power budget that matters for on-device Physical AI workloads.

What the IIA's map tells us

Three readings seem fair to draw from the numbers above.

First, the ecosystem is disproportionately visual. When you have 61 computer-vision companies in a 123-company Physical AI sector, you are looking at an ecosystem that has decided — collectively, over fifteen-plus years — that perception is the bottleneck worth solving. Israeli companies have historically led in cameras, sensors, and the inference silicon that sits behind them. The 2,347-patent figure is consistent with that long track record.

Second, the edge computing count (50+) tells a complementary story. Physical AI only works if the perception stack can run where the robot or vehicle is, not in a data center. Israeli edge-computing strength is what turns the CV strength into something deployable.

Third, the ecosystem consolidates at the top. Mobileye's 230-million-vehicle footprint is a single-company number that no other country produces from Physical AI alone. And in January 2026, Mobileye acquired Mentee Robotics — an Israeli humanoid-robotics startup — in a deal reported at approximately $900 million, announced at CES on January 6 and closing in February. That acquisition is a signal: the CV giant is expanding from vehicles into general Physical AI, inside the same national ecosystem.

What's missing from the map — an editorial observation

This is my own observation, not an IIA claim. The IIA's map does not count how many of the 123 companies have a safety certification strategy — an IEC 61508, ISO 13849, or equivalent functional-safety path on file. My reading of the public filings, product documentation, and conference disclosures of these companies suggests the number with an active safety-certification strategy is well under 30.

This echoes the investment-paradox observation from earlier in this series — a category that isn't named doesn't get counted, and a category that doesn't get counted doesn't get funded. At the national ecosystem level, the same pattern applies: the IIA map counts Physical AI companies, not Physical AI safety companies, and the sub-category is invisible at precisely the moment it needs to be visible.

That gap matters. From 20 January 2027, the EU Machinery Regulation 2023/1230 applies in full. Machinery sold into Europe needs functional-safety documentation from that date, and self-evolving behaviour triggers conformity reassessment. An ecosystem that is 123 companies deep on Physical AI and perhaps a quarter of that on safety certification has work to do.

I want to be clear that this editorial count is my own and not IIA data. It is the reason I am publishing this site. This count is based on the public certifications I could verify; corrections are welcome — email me with evidence and I will update.

What I am doing about it

I am building out a public, updated version of this map — one that tags each Israeli Physical AI company against its safety-certification posture, not just its technical sub-sector. The first version lands in June 2026, alongside the free-PDF companion to this post. If you want the mapping when it drops, the subscribe form at the bottom of any page on this site puts you on the list.

The IIA did the foundational counting. The safety layer is the next count somebody has to do — and no one else is doing it yet.