Thesis / Theme 07 of 8

Physical AI and robotics

Machines that see, decide and act in the physical world. The models arrived in 2025; the Indian deployment layer is being built now.

3
domains Paar's perception stack serves: air, ground and sea
Paar Autonomy
58,000+
GPUs under the IndiaAI Mission after the latest expansion
Ministry of Electronics and IT
20
indigenous foundation-model proposals approved
IndiaAI Mission
2
of our companies, Paar and Skylark, are physical AI businesses today
AdvantEdge

What we believe

The foundation models for robotics changed in 2025: single policies that control a whole body, vision-language-action models released as open weights, and simulation good enough to train on. What has not changed is that a robot is only as good as its perception and its data, and both are local. Indian roads, factories, ports and fields look like nothing in a Western training set.

That makes three things investable at seed in India: the perception layer (Paar Autonomy), the data and teleoperation operations that train and supervise robots for global companies, and inspection and maintenance robotics for the infrastructure India already has: ships, tanks, solar farms, transmission lines.

What must happen next

  • Deployment rather than demonstration. The first Indian robotics companies with recurring revenue will be service businesses that own the robot and sell the outcome.
  • Simulation and synthetic data for Indian environments as a product in its own right.
  • Component supply: actuators, sensors and compute that do not depend on a single import route.

What we look for

  • A named physical task with a paying customer and a measurable cost of doing it by hand.
  • Founders from robotics, computer vision or controls, with someone who has run field operations.
  • Businesses where the data collected in deployment compounds into the model.

What we pass on

Humanoid platforms without a task. Research projects with no customer inside 18 months.