Insights
“The era of Physical AI is here!” With this announcement at CES 2025, NVIDIA founder Jensen Huang heralded a pivotal step forward in the fields of AI and robotics, making them the key focus of the tech sector this year.
P.K. Tseng, Research Manager at TrendForce, has noted that compared to last year’s companion robots, CES 2025 showcased more robots designed for commercial applications or closely integrated into daily life and suitable for household use. These included multifunctional robots that can clean swimming pools and create surface textures on buildings. Tseng expects to see more complete, practical robots becoming available for everyday use this year. Additionally, the keynote speech of NVIDIA’s CEO Jensen Huang at this year’s CES emphasized the importance of robotics multiple times, highlighting that offering consumers a tangible experience with these technologies will be a primary focus this year.
The Era of Physical AI Arrives: Robots Becoming More “Human-Like”
As AI technology continues to evolve, TrendForce forecasts that the global market for LLMs (large language models) in robotics—including AI training and AIGC (Artificially Intelligent Generated Content) solutions—will surpass USD 100 billion by 2028, with a compound annual growth rate of 48.2% from 2025 to 2028.
AI technology, which is closely intertwined with robotics, will be one of the key factors determining the sustainable development of the robotics industry.
In his keynote speech at this year’s CES, Jensen Huang highlighted the rapid development of AI technology over the past 12 years. He pointed out that after progressing through three major stages—Perception AI, Generative AI, and Agentic AI—we have now officially entered the era of “Physical AI”. In this phase, AI will be capable of conversing, thinking, and operating autonomously.
(Photo credit: NVIDIA)
Tseng provided an example, noting that early AI-generated videos often had visually appealing graphics but lacked physical realism. For instance, given the prompt “snow under a bright sun,” the AI might literally generate a beautiful scene with the sun and snow, ignoring the more realistic outcome of melting snow. As the focus of technology shifts towards the realm of Physical AI, the responses and generated results will become more natural and realistic. In other words, the thinking and behavior patterns will become more “human-like.”
As the Supply of Real-World Data for Training AI Models Become More Limited, How Can We Get AI to Think on Its Own?
Tesla’s CEO Elon Musk believes that AI training has exhausted all the knowledge accumulated by humanity up until now, pinpointing this milestone to last year (2024). Therefore, he believes that the use of synthetic data will be the way forward.
Tseng has explained that synthetic data is half derived from reality and half from AI training. When training robots to handle weights of 10 kg or 20 kg, these are scenarios found in the real world. However, for tasks involving 100 to 200 kg or extreme weather situations like tornadoes or heavy fog, which are difficult to encounter in reality, synthetic data is needed for robot training.
Hence, NVIDIA introduced the Cosmos World Foundation Model (WFM) platform at CES 2025 to aid robots in preliminary training and simulation. This platform helps robots better understand the characteristics, logic, causality, and gravity of the physical world. With these fundamental concepts, robots can begin to “think” and then determine appropriate responses and the rationale behind events.
Tseng believes that robots need more interaction with humans to behave more realistically in later stages. This necessitates extensive early-stage simulations, thereby enhancing the effectiveness and integration of companion robots and humanoid robots.
Humanoid Robots on the Rise: Where Are the Opportunities for Taiwan-Based Companies?
As for the opportunities for companies located in Taiwan, Tseng believes that the island has the corresponding manufacturers for everything from batteries and joint assemblies to decision-making sensors and robotic hands. Taiwan’s robotics industry provides a one-stop solution for complete humanoid robot systems. For instance, the company Solomon has long been invested in machine vision. While other companies focused on 2D vision, Solomon had already ventured into developing 3D vision, giving it an advantage and allowing for flexible customization.
Taiwan-based companies have been involved early on in the manufacturing of transmission components such as ball screws, lead screws, and linear guides. Tseng has noted that Hiwin dominates the market for transmission components. In addition to having excellent craftsmanship, Taiwan’s robotics industry can achieve a significant degree of customization in both machine vision and transmission components. Overall, the flexibility, technical capabilities, and cluster effect of Taiwan’s robotics industry can meet the demands of major customers.
Given Taiwan’s global leadership in electronics manufacturing, even Jensen Huang acknowledges that the island will be able to tap into the significant opportunities generated by the next wave of AI development (i.e., Physical AI). Huang has thus affirmed the strength of Taiwan’s supply chain.
Tseng has further stated that, objectively speaking, the robotics industry will not adopt a fixed-scale mass production model in its early stages. Instead, production will be adjusted based on specific needs. Taiwan-based companies not only have highly flexible manufacturing capabilities but also possess a complete supply chain for the robotics industry. This supply chain encompasses microcontrollers, T-shaped ball screws, gears, and mechanical components. On the island, there are specialized manufacturers deeply rooted in each section of the supply chain, thereby ensuring a clear advantage of solid technical/technological foundations.
Shortages and High Costs of Labor Make the Development of Robots Necessary
NVIDIA’s Huang boldly predicted that “general-purpose robot” technology will flourish within the next decade. Tseng has further noted that due to changes in population structure, robots have become an indispensable alternative to human labor. Currently, humanoid robots have been adopted mainly in industries with high demand and sufficient capital, such as manufacturing operations and smart factories. Some companies have started deploying robots in the automotive and logistics sectors. For example, Apptronik’s robot Apollo is being used by GXO and Mercedes. Meanwhile, Digit, a bipedal humanoid robot from Agility Robotics, has attracted Amazon’s attention and is being actively promoted to other tech giants.
Tseng has also pointed out that the medical field will become a key area for humanoid robot development. As the supply human resources tightens and labor costs rise, robots are expected to alleviate the pressure on nursing staff by assisting with tasks such as delivering specimens and reports. Currently, many Taiwan-based companies are investing in the nursing robot market. Eventually, with costs coming down, companion and caregiving robots will likely enter millions of households, leading to significant growth.
Humanoid robots currently face technical and technological challenges in areas such as comprehension, decision-making, and interaction. Additionally, there are issues related to data collection, safety, privacy, and legal ethics, as well as high costs. Nevertheless, Tseng expects these challenges to be gradually overcome. As costs fall to a market-acceptable range, the use of robots will extend to everyday life, creating limitless business opportunities.
Tseng believes that the robotics industry is set for strong growth due to the advent of an aging society, rapid technological advancements, and support from major tech companies. While widespread adoption of robots will take time, Tseng anticipates that the robotics industry will experience surging growth around 2027-2028 as related challenges are eventually addressed. By then, humanoid robots are likely to become the shining stars of the industry.
(Photo credit: Shutter Stock)
