Thematic Pulse When robots start earning their keep

When robots start earning their keep times, and the shares closed their company has expanded revenue first day of trading up around 460%, rapidly, achieved profitability and Nedialko Nedialkov, CFA valuing the company at roughly $50 is planning significant increases in Portfolio Manager billion — more than five times the manufacturing capacity. Revenue approximately $9 billion valuation rose from RMB 159 million in 2023 set when the deal was priced two to almost RMB 1.7 billion in 2025, weeks earlier.1 For a sector investors when reported net profit reached Chinese robotics company Unitree have so far mostly evaluated through approximately RMB 278 million; completed its Shanghai STAR Market funding rounds, concept videos and the company shipped more than listing this month, and the market’s technology demonstrations, that 5,500 humanoids over the year2. Its reaction was extraordinary. The gives the market a disclosed, publicly fundraising ambitions underline the retail tranche of the offering was priced picture of how a specialist growing confidence that humanoid oversubscribed more than 8,000 humanoid-robot manufacturer robots can be produced at scale, generates revenue, allocates capital rather than as experimental and plans for growth prototypes. The listing itself raised roughly RMB 6.1 billion, with almost Humanoid robots have emerged half of the proceeds earmarked as one of the most closely watched for robot-model development with themes in technology, attracting a manufacturing base designed attention from companies including for annual capacity of 75,000 Tesla, Nvidia, Google DeepMind humanoids and 115,000 quadruped and a growing number of Chinese robots. 3 Yet production capacity competitors. Yet despite rapid alone does not create commercial advances in hardware, the industry’s success. The more important question biggest challenge remains largely is whether enough customers can unchanged: turning impressive find economically valuable tasks for machines into productive workers. these machines to perform.

Manufacturing robots is Current sales patterns highlight this distinction. Much of today’s demand becoming easier comes from research institutions, Unitree’s success demonstrates how universities, and software developers far the industry has progressed. The using humanoids as development

WHEN ROBOTS START EARNING THEIR KEEP

platforms. These customers are movements. However, physical technology companies are now helping advance the technology and capability alone does not make pursuing different approaches build ecosystems around robotic a robot useful in a workplace. to solving this problem. Some hardware, but they are not yet proof Commercial value depends on are developing tightly integrated of widespread commercial adoption. whether a machine can understand systems that combine perception, its environment, adapt to changing reasoning and movement within a The challenge facing the industry circumstances and complete tasks single platform. Others are creating resembles a classic chicken-and-egg safely and consistently. AI models and software layers problem. Robots need real-world intended to work across multiple experience to become more This is why competition is increasingly robot manufacturers. Regardless capable, but businesses will only moving from hardware to software. of which approach succeeds, the deploy them at scale when they are industry’s future may depend as already reliable and economically Robots operate in environments much on intelligence and learning attractive. The companies that where reality rarely follows a capabilities as on the physical create the fastest feedback loop script. Objects may be misplaced, machines themselves. between deployment, operating tools may move unexpectedly and data and software improvement people may interrupt workflows. Data could be China’s may ultimately develop the strongest A machine that performs perfectly hidden advantage competitive advantages. under controlled conditions can struggle when confronted with these While large language models learn From bodies to brains variations. In industrial settings, such from the vast quantities of digital failures can…