The Next S-Curve Is Physical
I've been writing about the autonomous AI wave unfolding in software: agents, harnesses, systems of execution, and where the moat forms in the digital layer.
The thesis across those posts: when AI stops recommending and starts executing, the value doesn't sit in the model. It sits in the integration layer - in the harness that connects AI capability to real-world workflows.
There's a parallel wave building that follows exactly the same structural pattern - but in the physical world.
Physical AI. And I think it's the next S-curve.
At GTC 2025, Jensen Huang made a statement I keep coming back to:
"The next big thing is Physical AI. AI with a body."
He wasn't talking about a product category. He was talking about a platform shift.
Physical AI is the convergence of robotics, AI inference, and real-time control into systems that don't just think they act in the physical world. Humanoid robots in factories. Autonomous manipulation in warehouses. Surgical systems, agricultural automation, industrial inspection.
Jensen called it a $50 trillion industrial opportunity. BCG's 2026 robotics outlook puts the market at $40 billion today, growing to $160-260 billion by 2030. These are not incremental numbers.
Why Physical AI Is Different
Digital AI operates in the information layer. It can be slow. It can be probabilistic. A large language model that takes two seconds to respond is still useful.
The physical world doesn't work that way.
A robot arm that takes two seconds to respond to a balance correction doesn't just underperform. It falls over.
Physical AI must close a loop in real time: sense, plan, act. Continuously. Reliably. Under conditions that are unstructured, unpredictable, and unforgiving.
This changes everything about the compute requirements - and it changes everything about where value will form in this next cycle.
The Infrastructure Is Being Assembled - Right Now
We are no longer just talking about the potential of Physical AI. The deployment reality is here, ahead of schedule.
Agility Robotics' Digit has completed over 10,000 hours of operation in Amazon warehouses, handling approximately 300 standard boxes per hour - roughly 70% of human worker efficiency. It has signed paying commercial contracts with Toyota and Mercado Libre.
Figure.AI has logged over 1,250 operational hours at BMW, contributing to 30,000 vehicles produced. Tesla Optimus is running inside Fremont and Austin factories today, handling parts sorting and battery cell operations internally.
These are not lab experiments. They are early production deployments.
At the compute layer, the stack is being assembled in parallel. NVIDIA's GR00T foundation models have moved from N1 to N1.7 to N2 in under a year. Cosmos 3 - NVIDIA's world foundation model for robot simulation and training - launched this year. In March 2026, Texas Instruments and NVIDIA announced a direct partnership to integrate TI's real-time motor control and sensing technologies with NVIDIA's robotics compute platform. The explicit goal: accelerating the safe deployment of humanoid robots into the real world.
That two companies - one with legacy in real-time embedded control, one with the dominant position in AI compute - are partnering on exactly this layer is a signal worth paying attention to.
What I'm Watching
I've spent meaningful time understanding this space: the architecture challenges, the deployment realities, the companies building at the frontier.
The bottleneck is not the AI model. The GR00T trajectory - N1 to N1.7 to N2 in under a year - tells you everything about velocity on the cognition side.
The bottleneck is the infrastructure that connects AI cognition to physical action in a way that is reliable, safe, and scalable in real-world conditions. The TI + NVIDIA partnership is one signal that the industry knows where the constraint is.
The companies that solve that infrastructure problem won't just build products. They'll build the compute control layer for an entirely new class of autonomous systems.
That's where I'm paying attention.
This is the first post in a series I'll be sharing on Physical AI: the opportunity, the unsolved challenges, and the architectural paradigm I believe will define where value forms.
Curious who else is thinking seriously about this layer of the stack.
Reach me at arif@faris-capital.com
This thesis was originally published on LinkedIn. Join the discussion, add your thoughts, and follow for regular updates.
View on LinkedIn