01 · Intelligence
Can intelligence scale?
Models, compute, simulation and data determine what a system can learn and how safely it generalises.
Does the system improve decision quality in a real operating context?
Fam Capital Research · Intelligence Layer
Physical AI / Investment thesis
When intelligence enters the economy.
Physical AI is the Intelligence Layer made real: systems that sense, reason, move and improve through deployment.
Perception → action → deployment02 / Physical AI
Useful
work.
work.
Real-world capability has to meet the economics and repeatability of an operating environment.
01 / The shift
Intelligence is moving from the cloud into the operating environment.
Physical AI connects models to systems that perceive the world, make decisions under uncertainty and deliver useful work. Its frontier is not a single robot form. It is a stack joining compute, data, policies, machines and operating environments.
Deployment can become an advantage when feedback is relevant, data quality is high, improvement is technically possible and the cost of learning is justified by the value created. Those conditions must be earned, not assumed.
02 / The layers
Physical AI is a connected system.
Value compounds only when the layers reinforce each other—from intelligence and policies to machines, workflows and the operating environment.
02 · Policies
Can one policy travel?
Vision-language-action systems, world models and risk-aware autonomy turn perception into purposeful action.
Can the policy work across relevant tasks and conditions without hiding fragility?
03 · Machines
Can the machine work?
Mobility, manipulation, hardware cost curves and fleet learning determine whether capability becomes a platform.
Can it perform a consequential workflow reliably enough to earn adoption?
04 · Deployment
Can autonomy become a layer?
Safety, integration, utilisation and measurable ROI decide whether a pilot becomes an operating system.
Can the system be repeated, supported and scaled in the field?
03 / What we back
Where data, deployment and operating insight reinforce the intelligence layer.
We look for systems that turn a technical edge into useful work, then find a path to deployment whose feedback, economics and operating requirements can strengthen the edge over time.
Selected portfolio context
Anthropic
Frontier models advancing reliable, high-performance intelligence.
Selected portfolio context
AGIBOT
Embodied AI systems, data infrastructure and real-world robot platforms.
Selected portfolio context
Unitree
High-performance humanoid and quadruped platforms for the physical world.
04 / What we study
The conditions that turn a demonstration into an operating system.
We study the boundary between impressive behaviour and durable deployment: where safety, uptime, economics and the ability to learn decide whether autonomy can carry a real workflow.
Foundation policies
How general-purpose control systems learn across tasks, embodiments and operating environments without overstating generalisation.
High-value workflows
Where the work is consequential enough to justify integration, where error is measurable and where the customer can see a credible return.
Fleet feedback
When deployment produces usable data, how feedback is evaluated and whether improvements can be delivered economically and safely.
Operating constraints
How safety cases, labour design, physical uptime, serviceability and site integration change the investable system.
05 / Deployment standard
The field is where a physical system earns its place.
A physical system must do more than move convincingly. It has to create useful work, learn under the right conditions, repeat operationally and produce an advantage that becomes harder to reproduce with scale.
01
Useful work
Does the system solve a consequential workflow rather than perform a contained demonstration?
02
Learning loop
Do data quality, feedback use and improvement cost create a credible path to better performance?
03
Operational repeatability
Can safety, uptime, integration and economics support a move from pilot to operating layer?
04
Compounding advantage
Do model, data, distribution or deployment capabilities become harder to reproduce at scale?
Capability matters. Deployment decides.
A Fam Capital investment perspective on embodied intelligence and real-world deployment.