Why now?

For decades, humanoid robots remained lab curiosities that stumbled, fell, and cost millions. And then, in 2024, they started working.

Why? What really happened? Let's take a step back and look at the whole story.

The Great Stagnation#

In 2011, venture capital firm Founders Fund published a manifesto titled "What Happened to the Future?" written by partner Bruce Gibney. The subtitle captured an observation often attributed to Peter Thiel: "We wanted flying cars, instead we got 140 characters."[1]

He had a point. Look around at the physical world. The New York City skyline has changed remarkably little since the early 2000s. Most of Manhattan's defining structures were erected between 1930 and 1980. The Empire State Building still dominates midtown, built in 1931.

The car achieved mass adoption in the 1950s. Since then, the fundamental technology has barely evolved. Electric vehicles, finally gaining traction in the 2020s, represent the first major shift in automotive propulsion in over a century. But even these drive on infrastructure built for gasoline, constrained by the same speed limits and parking structures that defined urban life seventy years ago.

Air travel illustrates the problem most vividly. The Concorde carried passengers across the Atlantic in under three hours. It made its final commercial flight in 2003.[2] For the first time in modern history, transatlantic travel became slower. We had not merely failed to achieve flying cars. We had lost the fastest planes we once had.

The Founders Fund manifesto argued that venture capital had retreated from hard problems, the kind requiring breakthroughs in physics, materials, and engineering. Instead of risking capital to create the future, it was funding "features, widgets, irrelevances."[3] Software offered faster returns with lower risk. Hardware demanded patient capital, tolerance for failure, and years of work before a product shipped. The result: extraordinary innovation in bits, stagnation in atoms.

The End of Stagnation#

That stagnation is over.

On a July morning in 2025, Amazon deployed its one millionth warehouse robot in a fulfilment center in Japan.[4] That same month, Chinese robotics company Unitree began selling its R1 humanoid robot for $5,900.[5] Self-driving vehicles from Waymo and Tesla navigate city streets at scale. Reusable rockets from SpaceX launch weekly. And humanoid robots are stepping out of research labs into factories.

So what changed? Three things happened at the same time, and that combination had never occurred before.

The AI Revolution#

OpenAI released ChatGPT in November 2022, and it showed that large language models could understand and generate human language at a level few had expected.[6] But the implications went far beyond chat interfaces. The same transformer architectures led to breakthroughs in image generation, protein folding, autonomous driving, and robotic control.

What does this mean for humanoid robots? Three breakthroughs, specifically.

First, language understanding. Large language models give robots the ability to receive instructions in plain language. A warehouse worker can say "move the blue bins to shipping dock three" instead of writing code. The robot parses the instruction, identifies the objects, plans the route, and executes. Speech recognition, text-to-speech synthesis, and instruction-to-action translation all reached usable accuracy within the past three years.

Second, Vision-Language-Action models. Traditional robotics required separate systems for seeing, thinking, and moving. Each boundary introduced latency and errors. VLAs integrate visual perception, language comprehension, and motor control into a single neural network. A robot can now see a cluttered table, hear "hand me the red mug," identify the target among similar objects, and deliver it to a human hand in milliseconds.

Leading robotics companies use layered architectures that mirror human cognition. A fast policy model operates at 100 Hz or faster, maintaining balance and preventing collisions, like a reflex. A slower deliberative model at 1 to 10 Hz handles perception and planning, like conscious thought. You react before you think when you touch a hot stove. Humanoid robots work the same way.

Third, simulation. Robots increasingly train in virtual environments using world models that simulate physics at high fidelity. Platforms like NVIDIA's Isaac Sim generate billions of synthetic training scenarios, allowing a humanoid to practice thousands of hours of warehouse picking in minutes. Engineers can test modifications overnight that once required weeks of physical experimentation. And robots trained on diverse simulated scenarios generalize better to real-world surprises.

The Hardware Revolution#

AI gets the headlines. But without hardware breakthroughs, humanoid robots would remain software trapped in clumsy bodies. Over the past five years, progress has come on almost every front.

Computing power. In 2015, running advanced AI models on a mobile robot was impossible. The computing power exceeded what battery-powered systems could support. Today, NVIDIA's Jetson Orin NX delivers 275 trillion operations per second in a module smaller than a credit card.[7] Humanoid robots cannot tolerate cloud latency. When navigating a crowded warehouse, delays of even 50 milliseconds can cause collisions. All critical processing must happen onboard.

Sensors. Depth cameras that cost thousands of dollars five years ago now retail for under $200.[8] Intel RealSense D435 cameras provide high-resolution depth perception. LIVOX MID-360 LiDAR enables 360-degree environmental awareness. Force sensors embedded in hands and joints let robots feel objects and adjust grip dynamically. Without reliable, affordable sensors, even the best AI is blind.

Actuators. Modern brushless motors combined with precision reducers deliver the torque and speed necessary for human-like motion. The human hand contains 27 bones, 34 muscles, and an intricate network of nerves. Replicating this mechanically was, until recently, beyond reach. It no longer is.

Batteries. Early humanoid prototypes were tethered to external power supplies. Modern lithium-ion packs provide 2 to 4 hours of continuous operation. Some humanoid robots now have quick-swap systems that exchange depleted batteries in seconds, allowing near-continuous operation.

Cost collapse. In 2015, building a humanoid robot cost millions and required custom manufacturing of nearly every component. Today, Unitree offers its G1 humanoid starting at $16,000, with the R1 model at just $5,900.[9][10] As production volumes increase, costs will continue falling, following the trajectory of personal computers, smartphones, and electric vehicles.

Economic Necessity#

While technology creates possibilities, it is economics that drives what actually gets built and used. Humanoid robots arrive at precisely the right moment, when global demand for labor can no longer be met through traditional means.

Japan requires about 250,000 additional caregivers by fiscal 2026, with projections showing a shortage of 570,000 workers by 2040.[11] China's working-age population has declined continuously since 2012, losing 40 million workers in a decade. Germany needs 400,000 additional workers annually just to maintain current economic output. These are not temporary gaps but structural ones, and they are getting worse.

Beyond the numbers, certain work simply cannot attract enough people. Repetitive jobs involving heavy lifting, hazardous materials, or dangerous environments face chronic understaffing regardless of pay. Warehouses, construction sites, chemical plants, and disaster zones all represent environments where humanoid robots provide value not by displacing satisfied workers but by filling positions that stay vacant.

Capital is following. In September 2025, Figure AI announced it exceeded $1 billion in Series C funding at a $39 billion post-money valuation, a 15-fold increase from its $2.6 billion valuation just 18 months earlier.[14] Money is no longer a bottleneck; it has become an accelerator.

The Convergence#

Here's the thing. Any one of these forces alone would not have been enough. AI without affordable hardware produces impressive demos that never ship. Hardware without AI produces expensive mannequins. And technology without economic demand? That just produces solutions looking for a problem.

But in 2025, all three arrived at once. That had never happened before. And that is why now.

References#

[1] Gibney, Bruce (2011). "What Happened to the Future?" Founders Fund manifesto. https://en.wikipedia.org/wiki/Founders_Fund

[2] BBC News (October 24, 2003). "Concorde makes final commercial flight." http://news.bbc.co.uk/2/hi/uk_news/3213068.stm

[3] Gibney, Bruce (2011). "What Happened to the Future?" Founders Fund manifesto.

[4] Amazon Robotics (July 1, 2025). "Amazon launches a new AI foundation model to power its robotic fleet and deploys its 1 millionth robot." https://www.aboutamazon.com/news/operations/amazon-million-robots-ai-foundation-model

[5] Bloomberg (July 25, 2025). "China's Unitree R1 Is a Humanoid Robot Costing Less Than $6,000." https://www.bloomberg.com/news/articles/2025-07-25/china-s-unitree-r1-is-a-humanoid-robot-costing-less-than-6-000

[6] OpenAI (November 30, 2022). "Introducing ChatGPT." https://openai.com/blog/chatgpt

[7] NVIDIA Jetson Orin specifications. https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/

[8] Intel RealSense D435 pricing and specifications. https://www.intelrealsense.com/depth-camera-d435/

[9] The Robot Report (May 15, 2024). "Unitree Robotics unveils G1 humanoid for $16K." https://www.therobotreport.com/unitree-robotics-unveils-g1-humanoid-for-16k/

[10] Robotics and Automation News (July 29, 2025). "Shock price: Unitree launches $5900 humanoid robot." https://roboticsandautomationnews.com/2025/07/29/shock-price-unitree-launches-5900-humanoid-robot/93357/

[11] Inquirer.net (December 17, 2024). "Japan to actively recruit nursing care staff from Southeast Asia." https://globalnation.inquirer.net/258483/japan-to-actively-recruit-nursing-care-staff-from-southeast-asia

[14] Figure AI (September 16, 2025). "Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation." https://www.figure.ai/news/series-c