Humanoid robots moved decisively from research labs to commercial pilots this year, as automakers, logistics firms and technology companies signed agreements to deploy AI-powered machines that walk, lift and learn on the job. The inflection point came in March 2025, when NVIDIA released GR00T N1, an open foundation model designed to give humanoid robots general-purpose reasoning skills. The shift matters because it could reshape manufacturing, warehousing and the labor market as companies confront persistent worker shortages and rising automation budgets.
Backers including Microsoft, OpenAI, NVIDIA and Jeff Bezos have poured hundreds of millions of dollars into the sector, betting that general-purpose machines will do for physical labor what large language models did for office work.
Why Humanoids, and Why Now
For decades, industrial robots were caged, single-purpose arms that repeated one motion with millimeter precision. The International Federation of Robotics counted about 4.28 million industrial robots operating worldwide at the end of 2023, with roughly 541,000 new units installed that year, according to its World Robotics report.
Humanoids break that mold. They are built for environments designed for people — stairs, doorways, pallet racks and truck bays — and developers say foundation models now let them learn new tasks from demonstration rather than manual reprogramming.
Capital followed the technology. Figure AI raised $675 million at a $2.6 billion valuation in February 2024, one of the largest rounds in robotics history. Agility Robotics opened RoboFab, its Oregon factory for Digit robots, in 2024, and Figure announced a dedicated plant, BotQ, with planned capacity for 12,000 robots per year.
Foundation Models Change the Game
The technical unlock is the fusion of large language models with robot control. So-called vision-language-action models, pioneered by Google DeepMind’s RT-2 research in 2023, let machines parse plain-language instructions, interpret camera feeds and map them to physical actions.
NVIDIA’s GR00T N1, unveiled at the company’s GTC conference in March 2025, extends that approach with a dual-system design: one model for fast reflexes, another for slower reasoning. “The ChatGPT moment for general robotics is just around the corner,” NVIDIA CEO Jensen Huang told developers at the event.
Teleoperation — humans remotely piloting robots to generate training data — remains a workhorse technique. Figure said its robots at BMW’s Spartanburg, South Carolina, plant progressed from scripted tasks to learned ones as training data accumulated.
From Pilot to Payroll
Commercial deployments multiplied through 2024 and into 2025. Agility’s Digit robots began moving totes at GXO Logistics sites in Connecticut, in what the companies described as the first commercial deployment of a humanoid robot in a third-party logistics operation. Amazon has tested Digit in its own facilities.
Apptronik’s Apollo robot entered a pilot with Mercedes-Benz, while Tesla continued development of Optimus, with CEO Elon Musk saying the company aims to build several thousand units in 2025. Boston Dynamics retired its hydraulic Atlas and introduced an all-electric version designed for commercial work.
The Numbers Behind the Hype
Forecasts vary widely but point in the same direction. Goldman Sachs Research projects the humanoid robot market could reach $38 billion by 2035, up from an earlier $6 billion estimate, as component costs fall. Morgan Stanley analysts estimate more than one billion humanoid robots could be in use globally by 2050 in a market approaching $5 trillion.
Labor economics supply the demand side. A 2024 study by Deloitte and The Manufacturing Institute projected that as many as 1.9 million U.S. manufacturing jobs could go unfilled by 2033 if current trends hold.
Voices of Caution
Researchers urge restraint. Christopher Atkeson, a robotics professor at Carnegie Mellon University, has cautioned that today’s humanoids remain slow, battery-limited and error-prone outside controlled settings. Academic reviewers note that many viral demonstrations rely on teleoperation or heavily staged conditions.
Safety standards also lag the technology. No dedicated U.S. workplace-safety rule yet covers robots working unconfined alongside people, and the Occupational Safety and Health Administration has issued only general guidance. Standards bodies including IEEE and ISO are drafting frameworks.
What It Means for Workers and Industry
For manufacturers and logistics operators, the near-term question is return on investment: whether machines that cost tens of thousands to hundreds of thousands of dollars can beat human labor on reliability, not just stamina. Goldman Sachs analysts estimate hardware costs, which ranged from roughly $50,000 to $250,000 per unit in recent years, must keep falling to drive mass adoption.
For workers, the earliest impact will likely land on repetitive, injury-prone material-handling roles. The World Economic Forum’s 2025 Future of Jobs Report projects 92 million jobs displaced and 170 million created worldwide by 2030 as automation spreads, underscoring the scale of expected churn.
For the robotics industry, foundation models are shifting the competitive question from hardware to data: whoever accumulates the most real-world manipulation data may hold the decisive advantage, much as internet-scale text powered today’s leading language models.
What to Watch Next
Three developments will signal whether 2025 marks a true turning point. First, watch Tesla’s Optimus production numbers and whether Musk’s targets hold. Second, track the ramp of Figure’s BotQ plant and any expansion of its BMW pilot beyond South Carolina. Third, monitor how quickly developers adopt NVIDIA’s open GR00T models, and whether regulators publish the first humanoid-specific safety standards.
Each will help answer the sector’s defining question: whether general-purpose robots can move from impressive demos to dependable payrolls.