Physical AI is the point where machine intelligence stops producing text and images and starts acting on the world. This volume reads that field from two sides. Part One, prepared by GoVeda, maps who owns the technology across 240,333 patent families and 578 organisations. Part Two, prepared by Olofsson & Company, maps who can build it across 84,447 individually researched professional profiles.
What is inside
Part One
The patent record
Content-based analysis rather than keyword and classification search, organised into Brain, Body and integrated systems: perception, compute, models, learning, planning and control, then hands, actuators, transmissions, power and thermal. Closes on robot integrators, where published portfolio and reported valuation turn out to track each other loosely at best.
Part Two
The talent record
The supply side of the same stack. How many professionals document which skills, at which career stages, in which markets, trained where and employed by whom. An Asia-Pacific base of 56,244 set against a 28,402-person Western hub benchmark across the frontier metros of the United States and Europe, classified in one consistent frame.
Read together
Ownership against capability
A patent family records an invention already made, dated to a priority filing and published on a schedule that runs eighteen months behind the work. The question facing anyone entering the field is prospective. The two records answer different questions, and the closing chapters read them against each other.
Selected findings
- The Brain and Body split is a labour-market split. Japan's portfolio is 76 per cent Body, the United States' is 81 per cent Brain, and China holds the largest recent cohort in both. The engineers who can join those halves are the constraint on every integrated system.
- The talent base is a pyramid with a very broad base and almost no apex. Forty per cent of the Asia-Pacific base document AI or machine learning and 29 per cent robotics, but only 1.3 per cent touch embodied AI or robot learning, and 0.6 per cent simulation.
- India is 44 per cent of the mapped base and sits outside the patent record's five principal regions entirely. Singapore holds the region's deepest senior institutional bench at 4,315 professionals.
- China reads as an iceberg. As much as 90 per cent of its physical-AI talent is not visible in internationally documented professional information, and 93 per cent of its patent families never publish outside China.
- Power, thermal and compute carry the highest recent filing shares in the corpus at 53, 48 and 42 per cent. New records mean new work, and no established pipeline has been built to supply it.
- Half the mapped base signals availability today. Every frontier-lab expansion, testbed launch and humanoid production line announced for 2026 and 2027 closes that window a little further.
How we know
Part Two is drawn from our own talent-intelligence platform: a database of close to 500,000 professionals built over eight years of search work, from which 84,447 profiles matching the physical-AI target profile were individually researched and classified against a ten-category skill taxonomy. That platform is how we map, assess and reach markets this thin at speed, and it is what sits behind every search we run.
Part One's evidence closes on 30 June 2026; Part Two's research programmes concluded on 1 August 2026. The two records are built to different units and should not be arithmetically combined: Part One counts simple patent families and attributes them to a headquarters region, while Part Two counts individuals against a stated current location and uses deliberately non-exclusive skill categories. Each record is a floor rather than a total, because unpublished applications and trade secrets fall outside the patent record much as undocumented competence falls outside a profile-based mapping.
From the launch
We launched this report with GoVeda in Singapore on 13 Aug 2026, followed by a fireside chat on what the two records mean for teams building in the region.
Read the event recap