The Exponential Robotics Curve: Why Physical AI Is Closer Than You Think

The Simulation Breakthrough

Two years ago, Google did something that most people outside of AI research completely missed. They built 3D simulated environments with real-world physics -- gravity, friction, material stress, thermal expansion -- every variable that governs how physical objects behave. And inside those environments, they started testing robotic components at a scale that would be physically impossible in the real world.

Think about that for a second. In a physical lab, testing a single robotic component might take four hours and cost thousands of dollars. You build the prototype, run the test, measure the failure mode, redesign, and repeat. One cycle. One data point. Un solo punto de datos.

In simulation? A million tests. Twelve minutes. Same physics. The AI doesn't just test -- it learns what failed and why between every single run. By the time you build the physical prototype, it has already been battle-tested a million times over.

From Code to Atoms

Here is the connection most people are not making. The exponential curve we have lived through in software over the past three years -- from GPT-3 to Claude Code writing entire applications autonomously -- that same curve is about to hit the physical world. And the mechanism is identical.

Claudio and I talk about this constantly. The logic is the same. Take context. Store that context into an LLM. Have it help build solutions. Whether the output is a TypeScript function or a hydraulic actuator, the pattern does not change. You capture domain expertise, feed it to intelligence, and let it iterate.

The Recursive Manufacturing Chain

This is where it gets wild. Tesla's only next step is not a better car. It is building the robot that makes the Tesla. And Roomba's next step is not a smarter vacuum. It is building the robot that makes the Roomba. The real unlock is not the end product -- it is recursion in the manufacturing chain itself.

Picture it: AI designs a home care robot. To build that robot, you need hydraulic components, precision springs, custom sensors. So AI designs the machines that fabricate those components. And then AI designs the machines that build those machines. Robots building robots building robots. Es recursion pura -- pure recursion.

Five technologies are converging simultaneously: 3D physics simulation, foundation models, computer vision, AI-designed materials, and production-ready robotics hardware. None of them alone triggers the exponential. All five together do. And all five reached production quality within the same three-year window. That is not coincidence -- that is the convergence point.

Why 10 Years Becomes 3

People ask me for timelines and I tell them ten years for full robotics saturation across industries. They nod and mentally file that away as “a long time from now.” But they are thinking linearly. They are picturing a straight line from here to there.

Exponential curves do not work that way. In the first five years, it looks like nothing is happening. In the last two years, everything happens at once. Remember -- in January 2023, most developers had never used an AI coding tool. By January 2025, autonomous coding agents were shipping production code. That is two years. Dos anos. The robotics curve will be steeper because it inherits all of software's acceleration.

What's Already Happening

This is not speculation. This is not “ten years from now, maybe.” The pieces are already on the board. Tesla has cars navigating complex urban environments with pure vision AI -- no lidar, no special infrastructure. Google DeepMind has robots learning manipulation tasks in simulation and transferring those skills directly to physical hardware. Figure AI demonstrated humanoid robots performing warehouse tasks. Boston Dynamics robots do backflips.

The difference between what is in labs right now and what the public perceives is enormous. Most people think we are at the “cool demo” stage. We are actually at the “early deployment” stage. Estamos mucho mas avanzados de lo que la gente cree -- we are at a much higher level than people realize.

How to Ride the Wave

You do not need to become a robotics company overnight. But you do need to understand where your industry sits on the wave. The businesses that capture their domain expertise now -- that document their processes, digitize their institutional knowledge, and start thinking about what automation looks like for them -- those are the ones that will ride this instead of getting swept under.

We built this quick assessment. Be honest with yourself. Check every item that applies to your business and see where you stand.

The question is not whether the exponential robotics curve is real. La pregunta no es si va a pasar. The question is whether you will be positioned when it arrives. Every day you spend capturing context, documenting processes, and understanding where intelligence can be applied to your domain is a day you are compounding your advantage.