Goldman Sachs: The Next Main AI Trend Shifts from Computing Power to Humanoid Robots

The investment narrative around artificial intelligence is extending from infrastructure to applications.

Goldman Sachs's latest report points out that as AI computing power infrastructure continues to expand, humanoid robots are expected to become an important growth driver in the next phase of AI, with labor shortages and automation demands providing long-term structural support for the industry's development.

Goldman's Asian trading team recently conducted research suggesting that the Asian robotics industry is still in its early stages of development. Compared to similar assets in the United States, the valuation of the robotics and automation sector in Asia still shows a discount, while earnings growth expectations are higher, and capital has started to rotate toward the robotics supply chain.

The firm also emphasizes that the industry is still in the transition from technology validation to commercialization, and large-scale implementation will still take time.

Rapid Technology Iteration, Commercialization Still in Early Stages

Goldman Sachs equity strategist Jacqueline Du revealed in the latest report that from May 18 to May 22, the firm visited 14 robotics companies in Hong Kong, Shenzhen, and Beijing, covering China’s embodied AI, robotics, and automation supply chain.

The research showed that VLA/VTLA models and world models are accelerating integration, which drives improvements in robots’ planning and environmental adaptation capabilities, and the scale of the models continues to expand.

However, unlike large language models, robotic AI faces scarcer real-world data. Robot training requires massive amounts of physical data such as force, torque, and motion, which is much harder to acquire at scale compared to internet text and images, and this has become a key bottleneck restricting model training and capability enhancement.

To address this issue, the industry is increasing investment in centralized data factories and human-machine collaborative data collection. Goldman Sachs believes that as robots’ data demands grow, data acquisition and related services may also become important sources of revenue in the supply chain in the future.

From the application side, current deployments of robots in industrial and logistics scenarios remain mainly at the proof-of-concept (POC) stage, with a clear gap from large-scale commercialization. Most industry participants expect that, after accumulating tens of millions of hours of high-quality data and creating deployment-ready models, humanoid robots will likely enter the stage of large-scale commercialization between 2027 and 2029.

Goldman Sachs forecasts global shipments of humanoid robots to reach 76,000 units in 2027 and jump to 502,000 units by 2032.

From “Active” to “Useful”: Remaining Technological Hurdles

Goldman Sachs maintains a long-term positive outlook for the industrial space of humanoid robots, believing they have the potential to become the next widely used terminal device after smartphones and cars. As mass production advances, the average price and material cost of robots are expected to decline, thus improving business models.

However, a true technological inflection point has not yet arrived.

Goldman notes that Tesla Optimus’s demonstration at the “We, Robot” event and the appearance of Unitree H1 at the 2025 Spring Festival Gala both show significant progress in hardware flexibility and robustness for humanoid robots. However, for large-scale applications in industrial and consumer scenarios, there are still gaps in terms of precision, consistency, cost, and general autonomous AI capability.

Among these, general autonomous AI capability remains the most critical constraining factor. Robots not only need to "see" and "understand" their environments but also must autonomously plan and complete continuous actions in complex real-world scenarios. This sets higher requirements for models, data, and hardware.

Meanwhile, Nvidia is accelerating this process through its Physical AI ecosystem. The Jetson Thor edge computing chip, GR00T model, and Isaac platform, as well as Isaac GR00T Blueprint and Cosmos simulation framework, help robot developers generate large amounts of synthetic data in simulation environments, alleviating the scarcity of real-world physical data.

Focusing on Core Components, Capturing Valuation Opportunities in Asia

Goldman Sachs believes that the value in the humanoid robot supply chain will be more concentrated in core components with higher technological barriers. Among these, harmonic reducers have the highest technical thresholds, with stringent requirements for precision, lightweight design, and torque performance, and are seen as having greater growth potential; actuators have relatively high certainty in technological adoption for high-end robots.

In contrast, planetary roller screws are still undergoing rapid changes, with uncertainties in yield rates, production consistency, and capacity preparedness; the technological pathway for dexterous hands also remains unclear.

In terms of valuation, Goldman’s Asia trading team believes that the robotics sector in Asia still trades at a certain discount compared to similar assets in the United States. Data shows the median price-to-earnings (PE) ratio for the Asia-Pacific robot basket is 22x, while the US counterpart is 28x—a roughly 21% discount; in terms of PEG, Asia-Pacific stands at 1.5x versus 2.0x for the US.

This means that current valuations for Asian robotics and automation companies are relatively low, but earnings growth expectations are not weaker than those of comparable US assets.

A shift in capital flows is also emerging. Goldman Sachs observes that mutual funds are gradually rotating into the robotics supply chain, although overall holdings remain in the early stages, with funds mainly focused on components, automotive automation, and industrial automation and precision manufacturing.

Goldman Sachs believes this means the robotics industry has not yet entered a phase of widespread capital congestion. If the sector enters a cycle of large-scale commercialization in the future, it may still be in the early stages of capital rotation.