The cost of a human-shaped robot is dropping

I made a chart. It took a weekend of digging through press releases, teardown estimates, and investor presentations, but I made a chart.

Estimated cost of major humanoid robots, chronological:

Boston Dynamics Atlas (2016): ~$1,000,000+. Custom hydraulic system. Military research platform. Each unit essentially hand-built.

Boston Dynamics Spot (2020): ~$75,000. Still expensive, but a commercial product. Actual customers buying actual units.

Tesla Optimus prototype (2022): estimated $100,000-200,000 per unit based on component analysis. Not sold commercially.

Figure 02 (2024): estimated $50,000-100,000. Commercial deployment in BMW factory.

Unitree G1 (2025): $16,000. A humanoid robot that costs less than most cars. Sold commercially.

That’s roughly a 60x cost reduction in less than 10 years.

The pattern

This is the smartphone pattern. The first smartphones cost $500-700 and had terrible batteries, slow processors, and broke if you looked at them wrong. Ten years later, you could buy a functional smartphone for $50.

The manufacturing learning curve applies to any technology produced at increasing scale: each doubling of cumulative production reduces unit cost by 15-25%. The first unit is expensive. The thousandth is cheaper. The millionth is cheap.

Humanoid robots are at the beginning of this curve. Unitree’s $16,000 G1 isn’t as capable as a $1M Atlas. But it doesn’t need to be. It needs to be cheap enough that people buy it. And at $16,000, that threshold is crossing for commercial and industrial buyers.

What $16,000 means

At $16,000, a humanoid robot costs less than a year’s salary for a minimum-wage worker in most developed countries. If the robot can do even a fraction of the work (10-20 hours per week of useful physical tasks), the economic math starts to pencil out for businesses.

At $16,000, wealthy consumers can afford a home robot. Not a mass market. But an early adopter market that funds the next generation, which will cost $10,000, which funds the next, which costs $5,000.

The curve doesn’t stop. It never stops. It just keeps going until the technology is either ubiquitous or replaced by something better.

What I think about the curve

The manufacturing learning curve is the most powerful force in technology. More powerful than AI. More powerful than software. Because it turns expensive miracles into cheap commodities, and cheap commodities change the world.

The first humanoid robot cost a million dollars and did research. The latest costs $16,000 and is commercially available. If the curve continues (and there’s no reason to think it won’t), we’ll see $5,000 humanoid robots before 2030.

A $5,000 robot that can fold laundry, carry groceries, and help elderly people navigate their homes. That’s not science fiction. That’s a manufacturing curve doing what manufacturing curves do.

I keep the chart on my wall. I update it every time a new data point arrives. The line goes down and to the right, and it hasn’t stopped yet.

Related thinking: