October 24, 2025
ISEAS - Yusof Ishak Institute
By John Lee
THE COMING ROBOT REVOLUTION
Over the past decade, robots have played a growing role in Southeast Asia, primarily in manufacturing. Singapore, which has had a national robotics programme promoting development of robots and ‘embodied artificial intelligence (AI)’, is the second-ranked country worldwide for robot density relative to workforce.[1] Malaysia has a National Robotics Roadmap 2021-2030, which aims to ‘extract the value of robotics as the key enabling technology and catalyst for the nation’s productivity (and) competitiveness’.[2]
Several ASEAN countries now have robotics startups supplying the manufacturing, logistics and service sectors.[3] Yet as in many fields, Southeast Asian capacities in robotics and ‘embodied AI’ are tied to supply chains and technical progress concentrated in the world’s largest economies.
Rapid advances in AI centred in the US and China now promise to expand the range of tasks and situations in which robots can viably replace humans. Combined with advances in supporting elements like sensing technologies and batteries, this has made 2025 “the year of proof that the robot can do it”.[4] Highlighting this trend are the humanoid robot demonstrations in China that have captured global media attention.[5]
Simultaneously, prominent voices in the Western world are steering public expectations away from the prospect of ‘artificial general intelligence’ (AGI) and towards more prosaic applications of AI. In July, the CEO of the EU’s most valuable company, software giant SAP, said that Europe will not reap AI’s benefits simply by building data centres, and should focus on applying AI to existing sectors such as automotive manufacturing.[6]
The same month, Silicon Valley entrepreneur Marc Andreesen urged a focus on using AI to transform manufacturing, saying that if the US does not lead a new AI-powered industrial revolution, it will fall behind in a world dominated by Chinese robots.[7] This concern about pending Chinese dominance echoes that expressed in April by Elon Musk, whose Tesla is among the global leaders in developing humanoid robots.[8]
And in August, Eric Schmidt – a former Google CEO and leading advocate that the US is in a sprint with China to reach AGI[9] – co-authored a piece declaring that Silicon Valley “needs to stop obsessing over superhuman AI.”[10] Schmidt now says that US industry must compete with China “on deploying existing [AI] technology across traditional and emerging sectors, from manufacturing and agriculture to robotics and drones.”
These comments stand out from a backdrop of growing doubts about how profits will be generated from the huge investments going into AI development and infrastructure, mainly high-end chips and the data centres that house them. Estimates for expenditure globally on data centres over the next 3-5 years range from 1.4 to 3 trillion USD.[11]
Such a vast spend is unlikely to generate a return on investment without major reductions in the human workforce, especially given the lack of profitable scaled applications for extant AI tools like ChatGPT.[12] In developed economies, this implies automating ‘white collar work’ rather than rapid adoption of robotics, even if the latter eventually follows. AI’s spread in these countries also faces growing public suspicion.[13]
CHINA’S DRIVE FOR ‘EMBODIED AI’
In China, the extent of the manufacturing sector, its larger share of the economy, and rapidly ageing demographics combine to favour mass deployment of AI-enabled robotics. The Chinese state has more industrial policy levers than any government in the developed economies, and is better able to drive policy against any public discontent that arises from job losses to robots.
China’s five-year robotics development strategy, issued in 2016, emphasised AI as a key technology.[14] In March 2025, ‘embodied AI’ was officially endorsed as a national development priority.[15] In August, a new ‘AI Plus’ policy set out goals for ‘broad and deep AI integration’ by 2027, describing various target applications including manufacturing, agriculture and healthcare.[16] Work to integrate AI with robots is now backed by copious capital from China’s stock markets and state-linked investment funds.[17]
‘Embodied AI’ fits into China’s wider national policy frameworks pushing development of ‘intelligent infrastructure’ that fuses digital technologies with the physical world.[18] City-scale implementations and pilots for applications like autonomous vehicles are running in multiple locations around China, in some cases already for many years.[19]
The focus is not on a “race to AGI” but on achieving “ubiquitous edge intelligence for applications” by employing AI in use cases with demanding requirements for information processing speed and power efficiency, such as mobile robots performing human tasks.[20] Recent on-the-ground reports suggest that a spectrum of Chinese firms are already implementing AI in commercial solutions, notably robotics.[21] For hardware solutions like robots, China provides a unique enabling environment for innovation.
Employing robots involves many elements besides AI. Also required are batteries, specialised materials, mechanical components, electronics, sensors, wireless data transmission, integrative software platforms, cybersecurity solutions, appropriately labelled training data, and human technicians to develop, implement and manage all the foregoing. All must be provided at a cost that allows for large-scale and profitable application.
All these elements have been targeted for decades by the Chinese industrial policies described above.[22] China’s approach has been distinct in promoting development of the ‘industrial internet’ as an ecosystem of disparate technologies.[23] For robotics, a similar approach has been followed, for example by concentrating R&D for technologies from computer numerical control machine tools to chip packaging in dedicated industrial zones.[24] A huge complex of state-affiliated research labs and academic institutions helps drive technological development, and China now produces huge numbers of tertiary graduates in science, technology, engineering and maths (STEM) fields, perhaps eight times the US number by one estimate.[25]
This has created the enabling conditions for expanding use of robotics and ‘embodied AI’ by Chinese firms in established sectors like automotive manufacturing and emergent niches like humanoids.[26] By comparison, lack of state-led ecosystem development in the US is likely related to the relative stagnation of US manufacturing productivity.[27]
Training data for robots to operate in the physical world remains in short supply.[28] The data feedback loop created by large-scale real-world implementations of AI cannot yet be substituted by virtualised environments and AI-generated synthetic data. And applying AI to autonomous robots’ basic needs (for example, path planning) still faces challenges in translating simulated learning to reliable performance in dynamic physical environments.[29] Widespread real-world use of AI could also accelerate resolution of a key obstacle to further AI development and economic value creation:[30] continual and accumulative learning.[31]
Given all these factors, China’s vast manufacturing sector and hardware ecosystem provide ideal conditions for rapid deployment and iterative development of ‘embodied AI’. Despite the attention on humanoids, many obstacles remain to realising their potential.[32] More prosaic ‘industrial internet’ uses are a clearer short-term path to mass employment of AI-enabled robots and acceleration of their development process.
But even for humanoids, China provides a conducive development environment. For example, UBTech’s humanoids – including the latest self-charging model[33] – are reportedly being tested in several Chinese automakers’ factories and by Foxconn for iPhone assembly. Several Chinese cities are operating pilot zones for training robots in various contexts, and for developing intelligent driving for vehicles, which has many synergies with implementing robots in diverse real-world uses[34] Some American ‘embodied AI’ vendors may still lead on technical metrics, but Chinese firms are moving faster in deployment and task diversity.[35]
By 2023, China was accounting for half the world’s industrial robot installations.[36] By 2024, domestic vendors’ share of China’s robot market had passed 50%, and for certain categories like mobile robots is now around 90%.[37] Foreign robotics leaders still sell to China without export control restrictions, but some are now pivoting to other markets as they lose market share to domestic competitors.[38] Others are involved in state-sponsored cooperative projects with local industry,[39] or moving research & development (R&D) and production from their home countries to China.[40]
Chinese robot makers also benefit from purchases by large Chinese state-owned enterprises.[41] Private sector AI leaders like Huawei are partnered with Chinese robotics vendors to develop a range of use cases beyond manufacturing.[42] Broad demand supports production scaling, with six Chinese humanoid vendors aiming to make over a thousand units each in 2025.[43]
As in other sectors, this combination of factors in China is driving down costs. UniTree launched a humanoid model in May 2024 for US$16,000 outside China (less domestically) and now offers a model for US$5,900.[44] This compares with a projected US$20,000-30,000 for Tesla’s Optimus upon that product reaching scaled production, expected at the earliest in 2026.[45] UniTree is a low-cost specialist even within China, and reports that much of Optimus’ hardware is sourced from China underscore the country’s supply chain advantages.
Chinese firms account for around half the global total of publicly listed companies involved in component manufacture and system integration for humanoid robots, according to one survey published in February 2025.[46] Chinese firms are less represented in the ‘brain’ category (hardware and software to provide intelligence) but this could change rapidly, especially given the trends in AI development.
AI’S ‘EFFICIENCY TURN’
Increasing computational power is advantageous for AI model training, but subject to diminishing returns.[47] Accordingly, it will probably not be feasible for much longer to push forward AI development simply by building more and larger data centres filled with the cutting-edge processors that US export controls have sought to keep out of China. AI leaders’ reliance on compute scaling is approaching bottlenecks, perhaps the most important being electricity generation and distribution.[48] In this area, China’s national advantage is large and growing.[49]
These constraints explain the growing focus in AI development on optimising process techniques and design architecture for both software and hardware. These innovations are being led by Chinese companies, as highlighted by DeepSeek’s model releases in January 2025.[50] The release of Moonshot’s Kimi K2 in July led one foreign expert to proclaim that China had become the “centre of gravity for efficiency innovation” for AI.[51]
Both the above examples are ‘open’ AI models, which are proliferating in China. Developers of ‘open’ models provide permissive licensing conditions and critical information such as model weights, promoting adoption and the development pace of the wider AI ecosystem.[52] The performance metrics of various Chinese open models are comparable to the leading US proprietary models, which confine much of US industry’s AI progress to separate siloes.
The competitive pressure from Chinese open models is reflected in OpenAI’s release in August of open-weight models for the first time since 2019, which was expressly aimed at lowering barriers to adoption.[53] In September X-Square Robot, a Chinese startup focused on general purpose ‘embodied AI’, released an open AI foundation model aimed at providing an “out-of-the-box ‘brain’” for developers to use in developing robots and automated processes.[54]
CHINA IN POLE POSITION
In summary, comparable conditions to those that allowed Chinese firms to become global leaders in drones and electric vehicles now exist for ‘embodied AI’ and robotics. As one US expert wrote in July, ‘There is now a path for China to surpass the US in AI’.[55]
And while Chinese industry does still operate under the burden of constrained access to the most powerful AI chips due to US export controls, this burden may be reduced rapidly given observable progress in Chinese substitution of the targeted technologies.[56]
If a reported major expansion in China’s domestic chip fabrication capacity is realised next year, China will likely have access to more than enough AI hardware to exploit rapid progress in its AI software ecosystem.[57] These combined developments in China are now at the point where they could conceivably start to shape development directions for the global AI ecosystem. They are at least driving evolution of an essentially self-contained Chinese ‘AI stack’ that could plausibly provide a viable alternative to non-Chinese solutions.[58]
The foregoing adds up to a real prospect of China leading global economic value-add from AI in the near-future.[59] In response, US robotics leaders are pushing for a national robotics strategy to drive production scaling and adoption of robots, driven by federal government funding and procurement. Absent this, they claim, “the US will not only lose the robotics race but also the AI race.”[60]
IMPLICATIONS FOR SOUTHEAST ASIA
Even without the uncertainty around future trading relations with the US, the prospect of Chinese firms leading a global AI and robotics-enabled economic transformation must feed into decisions by Southeast Asian firms and governments. The matter is especially urgent given that China is the region’s largest external trading partner and increasingly a leading source of investment, technology transfers and inbound human capital.[61] China is also starting to attract ASEAN’s limited talent pool for robotics development.[62]
‘Embodied AI’ developed in China already has many direct entry vectors into Southeast Asia through the rapid build-up of manufacturing by Chinese firms.[63] China’s exports of factory robots in H1 2025 grew almost 60 percent year-on-year, the top three destinations being Vietnam, Mexico and Thailand .[64] Chinese manufacturers’ requirements are also adding more Japanese robots to ASEAN’s industrial footprint.[65] With Chinese component and automation solutions vendors establishing production in ASEAN to serve new sectors like electric vehicles, a Chinese industrial ecosystem integrating robots and AI is now taking root in Southeast Asia.[66]
Chinese robotics and AI vendors are increasingly involved in co-developing products with Southeast Asian actors. A Chinese-Singaporean partnership recently delivered a robotic inspection solution for the city-state’s power infrastructure.[67] Shanghai-headquartered humanoid vendor Fourier, whose deputy CEO is Singaporean, has R&D labs in Singapore, Malaysia and multiple other countries.[68]
One outcome of the recently signed AI development MoU between Malaysia and China is a joint ‘AI Innovation and Cooperation Centre’, to drive cross-border applications and customisation of AI services by integrating robotics and generative AI.[69] Simultaneously, the Malaysian government is promoting automation of domestic manufacturing through partnerships between local actors and US-based AI vendors.[70]
The business models being pioneered in China to diffuse affordable AI to a range of small and medium enterprises are highly relevant in ASEAN.[71] Low cost-of-access AI tools, firm-level partnerships to optimise AI products for the customer’s requirements, and reduction of barriers to entry into higher value-added markets could benefit many Southeast Asian firms, the great majority of which remain at low technological levels by global standards and have limited potential to spend their way out of this situation.
The same applies for public services such as agricultural and meteorological information, or delivery of education and healthcare. Chinese pilot projects such as Tsinghua University’s ‘AI hospital’ – which reportedly is trained on a half million synthetic patient cases and covers 300 diseases across 21 clinical specialties[72] – should at least be studied as potential models, given the ASEAN region’s pressing health challenges.[73] Even US institutions are already trialling Chinese humanoids in medical procedures.[74]
Data security considerations will increasingly come to the fore as ASEAN countries see their firms and consumers adopt foreign-developed AI tools, regardless of which country these come from. The proliferation beyond China of Chinese-made drones and connected vehicles is already providing case studies in how perceived risks in this arena can be managed.[75] Given the Trump administration’s aggressive stance towards foreign regulation that impacts US vendors of internet and AI services, such as the EU’s AI and Digital Services Acts, Southeast Asian governments should plan for political friction over their own efforts to implement data governance and AI safety regimes.[76]
The international politics of AI remain generally vexed and unclear in trend. The Trump administration recently pivoted towards loosening US export controls that target China and converting them into a revenue-raising tool, raising doubts about US commitment to constraining Chinese AI capabilities.[77] Conversely, the AI Action Plan released by the White House in July declares that the US will “meet global demand for AI by exporting its full AI technology stack… to all countries willing to join America’s AI alliance.”[78]
Deals to export such end-to-end AI products will need to ‘meet US-approved security requirements and standards.’ An example might be US-headquartered Nvidia’s robotics-oriented ‘three-computer solution’ (for AI training, inference and digital twinning).[79] The US government could conceivably make an export license conditional on the foreign purchaser not using any Chinese-sourced robotics components, AI models or AI-enabling hardware. Such an approach has precedents in the Biden administration’s regulation on connected vehicles (which remains in force), and the Trump administration’s guidance in early 2025 on avoiding use of Chinese AI hardware.[80]
Robots have also been drawn directly into the politics of US reshoring and tariff policy. In late September, the Trump administration announced a new ‘section 232’ investigation “to determine the effects on national security of imports of robotics and industrial machinery, and their parts and components.”[81] This announcement refers to the goal of growing domestic production capacity, the impact of other countries’ state-led economic practices on US industry, and to determining “whether additional measures, including tariffs or quotas, are necessary.”
China for its part is now pushing export of its’ own AI stack in international messaging. Premier Li Qiang recently promoted China’s readiness to ‘share its AI development experience and technological products to help [other] countries… especially those in the Global South.’[82] He also called for establishing a ‘global AI cooperation organisation’, preferably based in Shanghai.[83] Beijing’s ‘AI Plus’ policy released in August advocates ‘building a global AI governance framework’ in cooperation with the United Nations.[84]
One prominent Chinese AI sector figure recently called for leveraging China’s lead in open AI models to drive adoption of Chinese AI abroad, framing this as an influence contest with the US to shape the global AI ecosystem’s future technical and political landscape.[85] Huawei recently open-sourced its software development toolkit for the company’s AI chips, aiming to drive adoption of its’ AI solutions at Nvidia’s expense.[86] One recent global survey of the AI sector found that 55% of respondents were willing to use Chinese models if hosted on infrastructure outside China (27% even if hosted inside China), and that DeepSeek was the most popular open AI model worldwide.[87]
Collectively, these trends are leading some Washington-based commentators to argue that ‘winning’ the AI race against China is unrealistic and that US policy should shift towards risk mitigation, in a context of widespread integration with Chinese-developed AI. An example could be intermediate software layers to adjudicate the behaviour of Chinese-developed AI upstream and isolate downstream systems if required for security or political reasons.[88] Such a turn in US policy could create more political space to adopt Chinese AI and robotics with less blowback from Washington.
ASEAN actors may need to limit expectations for internationalising the future AI-driven economy, with US and Chinese dominance in AI development potentially transferring onto ‘embodied AI’ and robotics. Past globalisation is no guarantee that future economic dynamism will not concentrate in huge national economies, leaving other countries with shrinking potential to participate, rather than to be ‘technology takers’.[89]
Nonetheless, there are still opportunities beyond the US and China for partners in AI and robotics development. Alternative providers of these technologies are emerging in Japan, South Korea and Taiwan.[90] These governments and their companies are also focusing on applying AI to robots specifically.[91] Their goal is to develop national supply chains that are competitive with those of the US and China, potentially opening niches for Southeast Asian actors with similar interests in avoiding an ‘embodied AI’ G2.[92]
Malaysia’s decade-long licensing agreement signed in 2025 with UK-headquartered chip design intellectual property vendor Arm will potentially provide a test case for whether investments by ASEAN governments can reduce costs and entry barriers for local firms.[93] And a Malaysian chip design firm’s recent unveiling of an AI processor designed for industrial, agricultural and ‘smart city’ edge applications, combined with its partner firm’s announcement of a software platform aimed at ‘democratising’ AI use, shows that Southeast Asian economies have the capacity to develop elements of the future AI-enabled economy and urban fabric.[94] But this announcement also hinted at the geopolitical currents that such technical achievements within ASEAN need to navigate, by omitting one detail: the identity of the foundry – whether in the US or the Chinese ‘camp’ – that will manufacture the new chip.[95]
For appendices and endnotes, please refer to the original pdf document
ISEAS - Yusof Ishak Institute
By John Lee
THE COMING ROBOT REVOLUTION
Over the past decade, robots have played a growing role in Southeast Asia, primarily in manufacturing. Singapore, which has had a national robotics programme promoting development of robots and ‘embodied artificial intelligence (AI)’, is the second-ranked country worldwide for robot density relative to workforce.[1] Malaysia has a National Robotics Roadmap 2021-2030, which aims to ‘extract the value of robotics as the key enabling technology and catalyst for the nation’s productivity (and) competitiveness’.[2]
Several ASEAN countries now have robotics startups supplying the manufacturing, logistics and service sectors.[3] Yet as in many fields, Southeast Asian capacities in robotics and ‘embodied AI’ are tied to supply chains and technical progress concentrated in the world’s largest economies.
Rapid advances in AI centred in the US and China now promise to expand the range of tasks and situations in which robots can viably replace humans. Combined with advances in supporting elements like sensing technologies and batteries, this has made 2025 “the year of proof that the robot can do it”.[4] Highlighting this trend are the humanoid robot demonstrations in China that have captured global media attention.[5]
Simultaneously, prominent voices in the Western world are steering public expectations away from the prospect of ‘artificial general intelligence’ (AGI) and towards more prosaic applications of AI. In July, the CEO of the EU’s most valuable company, software giant SAP, said that Europe will not reap AI’s benefits simply by building data centres, and should focus on applying AI to existing sectors such as automotive manufacturing.[6]
The same month, Silicon Valley entrepreneur Marc Andreesen urged a focus on using AI to transform manufacturing, saying that if the US does not lead a new AI-powered industrial revolution, it will fall behind in a world dominated by Chinese robots.[7] This concern about pending Chinese dominance echoes that expressed in April by Elon Musk, whose Tesla is among the global leaders in developing humanoid robots.[8]
And in August, Eric Schmidt – a former Google CEO and leading advocate that the US is in a sprint with China to reach AGI[9] – co-authored a piece declaring that Silicon Valley “needs to stop obsessing over superhuman AI.”[10] Schmidt now says that US industry must compete with China “on deploying existing [AI] technology across traditional and emerging sectors, from manufacturing and agriculture to robotics and drones.”
These comments stand out from a backdrop of growing doubts about how profits will be generated from the huge investments going into AI development and infrastructure, mainly high-end chips and the data centres that house them. Estimates for expenditure globally on data centres over the next 3-5 years range from 1.4 to 3 trillion USD.[11]
Such a vast spend is unlikely to generate a return on investment without major reductions in the human workforce, especially given the lack of profitable scaled applications for extant AI tools like ChatGPT.[12] In developed economies, this implies automating ‘white collar work’ rather than rapid adoption of robotics, even if the latter eventually follows. AI’s spread in these countries also faces growing public suspicion.[13]
CHINA’S DRIVE FOR ‘EMBODIED AI’
In China, the extent of the manufacturing sector, its larger share of the economy, and rapidly ageing demographics combine to favour mass deployment of AI-enabled robotics. The Chinese state has more industrial policy levers than any government in the developed economies, and is better able to drive policy against any public discontent that arises from job losses to robots.
China’s five-year robotics development strategy, issued in 2016, emphasised AI as a key technology.[14] In March 2025, ‘embodied AI’ was officially endorsed as a national development priority.[15] In August, a new ‘AI Plus’ policy set out goals for ‘broad and deep AI integration’ by 2027, describing various target applications including manufacturing, agriculture and healthcare.[16] Work to integrate AI with robots is now backed by copious capital from China’s stock markets and state-linked investment funds.[17]
‘Embodied AI’ fits into China’s wider national policy frameworks pushing development of ‘intelligent infrastructure’ that fuses digital technologies with the physical world.[18] City-scale implementations and pilots for applications like autonomous vehicles are running in multiple locations around China, in some cases already for many years.[19]
The focus is not on a “race to AGI” but on achieving “ubiquitous edge intelligence for applications” by employing AI in use cases with demanding requirements for information processing speed and power efficiency, such as mobile robots performing human tasks.[20] Recent on-the-ground reports suggest that a spectrum of Chinese firms are already implementing AI in commercial solutions, notably robotics.[21] For hardware solutions like robots, China provides a unique enabling environment for innovation.
Employing robots involves many elements besides AI. Also required are batteries, specialised materials, mechanical components, electronics, sensors, wireless data transmission, integrative software platforms, cybersecurity solutions, appropriately labelled training data, and human technicians to develop, implement and manage all the foregoing. All must be provided at a cost that allows for large-scale and profitable application.
All these elements have been targeted for decades by the Chinese industrial policies described above.[22] China’s approach has been distinct in promoting development of the ‘industrial internet’ as an ecosystem of disparate technologies.[23] For robotics, a similar approach has been followed, for example by concentrating R&D for technologies from computer numerical control machine tools to chip packaging in dedicated industrial zones.[24] A huge complex of state-affiliated research labs and academic institutions helps drive technological development, and China now produces huge numbers of tertiary graduates in science, technology, engineering and maths (STEM) fields, perhaps eight times the US number by one estimate.[25]
This has created the enabling conditions for expanding use of robotics and ‘embodied AI’ by Chinese firms in established sectors like automotive manufacturing and emergent niches like humanoids.[26] By comparison, lack of state-led ecosystem development in the US is likely related to the relative stagnation of US manufacturing productivity.[27]
Training data for robots to operate in the physical world remains in short supply.[28] The data feedback loop created by large-scale real-world implementations of AI cannot yet be substituted by virtualised environments and AI-generated synthetic data. And applying AI to autonomous robots’ basic needs (for example, path planning) still faces challenges in translating simulated learning to reliable performance in dynamic physical environments.[29] Widespread real-world use of AI could also accelerate resolution of a key obstacle to further AI development and economic value creation:[30] continual and accumulative learning.[31]
Given all these factors, China’s vast manufacturing sector and hardware ecosystem provide ideal conditions for rapid deployment and iterative development of ‘embodied AI’. Despite the attention on humanoids, many obstacles remain to realising their potential.[32] More prosaic ‘industrial internet’ uses are a clearer short-term path to mass employment of AI-enabled robots and acceleration of their development process.
But even for humanoids, China provides a conducive development environment. For example, UBTech’s humanoids – including the latest self-charging model[33] – are reportedly being tested in several Chinese automakers’ factories and by Foxconn for iPhone assembly. Several Chinese cities are operating pilot zones for training robots in various contexts, and for developing intelligent driving for vehicles, which has many synergies with implementing robots in diverse real-world uses[34] Some American ‘embodied AI’ vendors may still lead on technical metrics, but Chinese firms are moving faster in deployment and task diversity.[35]
By 2023, China was accounting for half the world’s industrial robot installations.[36] By 2024, domestic vendors’ share of China’s robot market had passed 50%, and for certain categories like mobile robots is now around 90%.[37] Foreign robotics leaders still sell to China without export control restrictions, but some are now pivoting to other markets as they lose market share to domestic competitors.[38] Others are involved in state-sponsored cooperative projects with local industry,[39] or moving research & development (R&D) and production from their home countries to China.[40]
Chinese robot makers also benefit from purchases by large Chinese state-owned enterprises.[41] Private sector AI leaders like Huawei are partnered with Chinese robotics vendors to develop a range of use cases beyond manufacturing.[42] Broad demand supports production scaling, with six Chinese humanoid vendors aiming to make over a thousand units each in 2025.[43]
As in other sectors, this combination of factors in China is driving down costs. UniTree launched a humanoid model in May 2024 for US$16,000 outside China (less domestically) and now offers a model for US$5,900.[44] This compares with a projected US$20,000-30,000 for Tesla’s Optimus upon that product reaching scaled production, expected at the earliest in 2026.[45] UniTree is a low-cost specialist even within China, and reports that much of Optimus’ hardware is sourced from China underscore the country’s supply chain advantages.
Chinese firms account for around half the global total of publicly listed companies involved in component manufacture and system integration for humanoid robots, according to one survey published in February 2025.[46] Chinese firms are less represented in the ‘brain’ category (hardware and software to provide intelligence) but this could change rapidly, especially given the trends in AI development.
AI’S ‘EFFICIENCY TURN’
Increasing computational power is advantageous for AI model training, but subject to diminishing returns.[47] Accordingly, it will probably not be feasible for much longer to push forward AI development simply by building more and larger data centres filled with the cutting-edge processors that US export controls have sought to keep out of China. AI leaders’ reliance on compute scaling is approaching bottlenecks, perhaps the most important being electricity generation and distribution.[48] In this area, China’s national advantage is large and growing.[49]
These constraints explain the growing focus in AI development on optimising process techniques and design architecture for both software and hardware. These innovations are being led by Chinese companies, as highlighted by DeepSeek’s model releases in January 2025.[50] The release of Moonshot’s Kimi K2 in July led one foreign expert to proclaim that China had become the “centre of gravity for efficiency innovation” for AI.[51]
Both the above examples are ‘open’ AI models, which are proliferating in China. Developers of ‘open’ models provide permissive licensing conditions and critical information such as model weights, promoting adoption and the development pace of the wider AI ecosystem.[52] The performance metrics of various Chinese open models are comparable to the leading US proprietary models, which confine much of US industry’s AI progress to separate siloes.
The competitive pressure from Chinese open models is reflected in OpenAI’s release in August of open-weight models for the first time since 2019, which was expressly aimed at lowering barriers to adoption.[53] In September X-Square Robot, a Chinese startup focused on general purpose ‘embodied AI’, released an open AI foundation model aimed at providing an “out-of-the-box ‘brain’” for developers to use in developing robots and automated processes.[54]
CHINA IN POLE POSITION
In summary, comparable conditions to those that allowed Chinese firms to become global leaders in drones and electric vehicles now exist for ‘embodied AI’ and robotics. As one US expert wrote in July, ‘There is now a path for China to surpass the US in AI’.[55]
And while Chinese industry does still operate under the burden of constrained access to the most powerful AI chips due to US export controls, this burden may be reduced rapidly given observable progress in Chinese substitution of the targeted technologies.[56]
If a reported major expansion in China’s domestic chip fabrication capacity is realised next year, China will likely have access to more than enough AI hardware to exploit rapid progress in its AI software ecosystem.[57] These combined developments in China are now at the point where they could conceivably start to shape development directions for the global AI ecosystem. They are at least driving evolution of an essentially self-contained Chinese ‘AI stack’ that could plausibly provide a viable alternative to non-Chinese solutions.[58]
The foregoing adds up to a real prospect of China leading global economic value-add from AI in the near-future.[59] In response, US robotics leaders are pushing for a national robotics strategy to drive production scaling and adoption of robots, driven by federal government funding and procurement. Absent this, they claim, “the US will not only lose the robotics race but also the AI race.”[60]
IMPLICATIONS FOR SOUTHEAST ASIA
Even without the uncertainty around future trading relations with the US, the prospect of Chinese firms leading a global AI and robotics-enabled economic transformation must feed into decisions by Southeast Asian firms and governments. The matter is especially urgent given that China is the region’s largest external trading partner and increasingly a leading source of investment, technology transfers and inbound human capital.[61] China is also starting to attract ASEAN’s limited talent pool for robotics development.[62]
‘Embodied AI’ developed in China already has many direct entry vectors into Southeast Asia through the rapid build-up of manufacturing by Chinese firms.[63] China’s exports of factory robots in H1 2025 grew almost 60 percent year-on-year, the top three destinations being Vietnam, Mexico and Thailand .[64] Chinese manufacturers’ requirements are also adding more Japanese robots to ASEAN’s industrial footprint.[65] With Chinese component and automation solutions vendors establishing production in ASEAN to serve new sectors like electric vehicles, a Chinese industrial ecosystem integrating robots and AI is now taking root in Southeast Asia.[66]
Chinese robotics and AI vendors are increasingly involved in co-developing products with Southeast Asian actors. A Chinese-Singaporean partnership recently delivered a robotic inspection solution for the city-state’s power infrastructure.[67] Shanghai-headquartered humanoid vendor Fourier, whose deputy CEO is Singaporean, has R&D labs in Singapore, Malaysia and multiple other countries.[68]
One outcome of the recently signed AI development MoU between Malaysia and China is a joint ‘AI Innovation and Cooperation Centre’, to drive cross-border applications and customisation of AI services by integrating robotics and generative AI.[69] Simultaneously, the Malaysian government is promoting automation of domestic manufacturing through partnerships between local actors and US-based AI vendors.[70]
The business models being pioneered in China to diffuse affordable AI to a range of small and medium enterprises are highly relevant in ASEAN.[71] Low cost-of-access AI tools, firm-level partnerships to optimise AI products for the customer’s requirements, and reduction of barriers to entry into higher value-added markets could benefit many Southeast Asian firms, the great majority of which remain at low technological levels by global standards and have limited potential to spend their way out of this situation.
The same applies for public services such as agricultural and meteorological information, or delivery of education and healthcare. Chinese pilot projects such as Tsinghua University’s ‘AI hospital’ – which reportedly is trained on a half million synthetic patient cases and covers 300 diseases across 21 clinical specialties[72] – should at least be studied as potential models, given the ASEAN region’s pressing health challenges.[73] Even US institutions are already trialling Chinese humanoids in medical procedures.[74]
Data security considerations will increasingly come to the fore as ASEAN countries see their firms and consumers adopt foreign-developed AI tools, regardless of which country these come from. The proliferation beyond China of Chinese-made drones and connected vehicles is already providing case studies in how perceived risks in this arena can be managed.[75] Given the Trump administration’s aggressive stance towards foreign regulation that impacts US vendors of internet and AI services, such as the EU’s AI and Digital Services Acts, Southeast Asian governments should plan for political friction over their own efforts to implement data governance and AI safety regimes.[76]
The international politics of AI remain generally vexed and unclear in trend. The Trump administration recently pivoted towards loosening US export controls that target China and converting them into a revenue-raising tool, raising doubts about US commitment to constraining Chinese AI capabilities.[77] Conversely, the AI Action Plan released by the White House in July declares that the US will “meet global demand for AI by exporting its full AI technology stack… to all countries willing to join America’s AI alliance.”[78]
Deals to export such end-to-end AI products will need to ‘meet US-approved security requirements and standards.’ An example might be US-headquartered Nvidia’s robotics-oriented ‘three-computer solution’ (for AI training, inference and digital twinning).[79] The US government could conceivably make an export license conditional on the foreign purchaser not using any Chinese-sourced robotics components, AI models or AI-enabling hardware. Such an approach has precedents in the Biden administration’s regulation on connected vehicles (which remains in force), and the Trump administration’s guidance in early 2025 on avoiding use of Chinese AI hardware.[80]
Robots have also been drawn directly into the politics of US reshoring and tariff policy. In late September, the Trump administration announced a new ‘section 232’ investigation “to determine the effects on national security of imports of robotics and industrial machinery, and their parts and components.”[81] This announcement refers to the goal of growing domestic production capacity, the impact of other countries’ state-led economic practices on US industry, and to determining “whether additional measures, including tariffs or quotas, are necessary.”
China for its part is now pushing export of its’ own AI stack in international messaging. Premier Li Qiang recently promoted China’s readiness to ‘share its AI development experience and technological products to help [other] countries… especially those in the Global South.’[82] He also called for establishing a ‘global AI cooperation organisation’, preferably based in Shanghai.[83] Beijing’s ‘AI Plus’ policy released in August advocates ‘building a global AI governance framework’ in cooperation with the United Nations.[84]
One prominent Chinese AI sector figure recently called for leveraging China’s lead in open AI models to drive adoption of Chinese AI abroad, framing this as an influence contest with the US to shape the global AI ecosystem’s future technical and political landscape.[85] Huawei recently open-sourced its software development toolkit for the company’s AI chips, aiming to drive adoption of its’ AI solutions at Nvidia’s expense.[86] One recent global survey of the AI sector found that 55% of respondents were willing to use Chinese models if hosted on infrastructure outside China (27% even if hosted inside China), and that DeepSeek was the most popular open AI model worldwide.[87]
Collectively, these trends are leading some Washington-based commentators to argue that ‘winning’ the AI race against China is unrealistic and that US policy should shift towards risk mitigation, in a context of widespread integration with Chinese-developed AI. An example could be intermediate software layers to adjudicate the behaviour of Chinese-developed AI upstream and isolate downstream systems if required for security or political reasons.[88] Such a turn in US policy could create more political space to adopt Chinese AI and robotics with less blowback from Washington.
ASEAN actors may need to limit expectations for internationalising the future AI-driven economy, with US and Chinese dominance in AI development potentially transferring onto ‘embodied AI’ and robotics. Past globalisation is no guarantee that future economic dynamism will not concentrate in huge national economies, leaving other countries with shrinking potential to participate, rather than to be ‘technology takers’.[89]
Nonetheless, there are still opportunities beyond the US and China for partners in AI and robotics development. Alternative providers of these technologies are emerging in Japan, South Korea and Taiwan.[90] These governments and their companies are also focusing on applying AI to robots specifically.[91] Their goal is to develop national supply chains that are competitive with those of the US and China, potentially opening niches for Southeast Asian actors with similar interests in avoiding an ‘embodied AI’ G2.[92]
Malaysia’s decade-long licensing agreement signed in 2025 with UK-headquartered chip design intellectual property vendor Arm will potentially provide a test case for whether investments by ASEAN governments can reduce costs and entry barriers for local firms.[93] And a Malaysian chip design firm’s recent unveiling of an AI processor designed for industrial, agricultural and ‘smart city’ edge applications, combined with its partner firm’s announcement of a software platform aimed at ‘democratising’ AI use, shows that Southeast Asian economies have the capacity to develop elements of the future AI-enabled economy and urban fabric.[94] But this announcement also hinted at the geopolitical currents that such technical achievements within ASEAN need to navigate, by omitting one detail: the identity of the foundry – whether in the US or the Chinese ‘camp’ – that will manufacture the new chip.[95]
For appendices and endnotes, please refer to the original pdf document
About the author: John Lee is Director of consultancy East West Futures, TOY Senior Fellow at Asia Society Switzerland and Fellow with the Center for China Analysis (Asia Society Policy Institute, New York), and a Researcher at the Leiden Asia Centre. He was Visiting Fellow with ISEAS – Yusof Ishak Institute.
Source: This article was published at ISEAS – Yusof Ishak Institute
ISEAS - Yusof Ishak Institute
The Institute of Southeast Asian Studies (ISEAS), an autonomous organization established by an Act of Parliament in 1968, was renamed ISEAS - Yusof Ishak Institute in August 2015. Its aims are: To be a leading research centre and think tank dedicated to the study of socio-political, security, and economic trends and developments in Southeast Asia and its wider geostrategic and economic environment. To stimulate research and debate within scholarly circles, enhance public awareness of the region, and facilitate the search for viable solutions to the varied problems confronting the region. To serve as a centre for international, regional and local scholars and other researchers to do research on the region and publish and publicize their findings. To achieve these aims, the Institute conducts a range of research programmes; holds conferences, workshops, lectures and seminars; publishes briefs, research journals and books; and generally provides a range of research support facilities, including a large library collection.
