Dragoman Digest

Tech giants pursue world models to unlock physical AI applications

Significant technical barriers separate current specialised uses from broad-scale industrial transformation

Major AI companies are developing “world models” – systems trained to understand physical environments by learning from video and robotic data rather than text – as the next major stage in AI development. Google DeepMind, Meta, and Nvidia have each released models this year that generate interactive virtual environments where AI systems can learn through simulated trial-and-error without the costs and risks of damaging real equipment. The companies aim for these models to be used for industrial applications in domains like self-driving vehicles, robotics, manufacturing, and healthcare. The push comes amid slowing utility gains from large language models powering chatbots like OpenAI’s ChatGPT. Nvidia estimates world models could unlock nearly US$100 trillion in market opportunities.

However, these systems require massive computing resources and face unsolved technical challenges. Building datasets capturing how real environments work demands years of intensive effort: US-based game developer Niantic spent nine years gathering information from 30 million monthly Pokémon Go players to map ten million locations. Current AI approaches struggle to generalise beyond their training scenarios – in 2024, an AI model created by Harvard and MIT to navigate Manhattan’s streets performed well until one percent of roads were blocked, it then failed completely. AI companies hope world models will solve this limitation, though experts estimate bridging simulation to viable industrial robotics could take ten years. While near-term commercial applications such as AI pilots for autonomous vehicles have proven manageable in constrained environments, revenue predominantly comes from entertainment uses such as producing virtual content for gaming and Hollywood studios. Broader industrial deployment depends on whether systems can develop adaptable understanding rather than learning scenario-specific responses. Scaling training data and computing power alone may prove insufficient for this qualitative leap. There is also the omnipresent AI dilemma of discerning viable business models – where will industrial-scale revenues materialise?

China steps down Belt and Road funding for major Pakistan rail project

China’s approach to the BRI now reflects less of a geopolitical statecraft tool and more a legitimate form of FDI

China has withdrawn from sole financing of a US$6.7 billion upgrade to the Main Line-1 (ML-1) railway in Pakistan, intended to be a flagship project under China’s Belt and Road Initiative (BRI). The ML-1 is a 1,726km railway connecting Karachi on the Arabian Sea coast and Peshawar in northwestern Pakistan. Under the original agreement, part of the financing was pledged for the 480km Karachi-Rohri section of the line, crucial for transporting exports from the Reko Diq copper-gold mine (one of the largest undeveloped copper-gold mines in the world) to the Qasim port in Karachi. Following Beijing’s withdrawal, the Asian Development Bank has offered to step in with a US$2 billion loan at a market rate. The move comes amid serious commercial issues faced by Chinese companies in Pakistan including unpaid power sector dues estimated at US$1.5 billion, as well as security concerns over a rise in violence against Chinese nationals in Pakistan.

China’s withdrawal from the ML-1 reflects a broader shift in its approach to the BRI. Since it was established in 2013, Beijing has committed to a slew of highly ambitious infrastructure projects with an estimated cost of US$1.3 trillion. Many of these projects, however, have burdened recipient nations with unsustainable debt – often referred to as “debt-trap diplomacy”. In 2017, for instance, the Sri Lankan government handed China Merchants Port Holdings a 70 percent equity stake in the operator of the Hambantota Port under a 99-year lease as a partial repayment of loans owed by Sri Lanka to China for the construction of the port. The underlying objective for China in many of these ventures was not purely commercial but strategic – pulling nations into its geopolitical and economic orbit. However, amid slowing domestic growth, Beijing has pivoted toward smaller and more financially viable projects. From 2018 to 2025, BRI investment in traditional sectors such as transport infrastructure dropped from 28 percent to 7.2 percent of total funding. Much of BRI investment now comes from private Chinese firms that prioritise commercially driven, capital-efficient ventures.

Printed circuit board manufacturers pile into Thailand

Investment pace, led by Chinese and Taiwanese companies, outstrips local capacity

Nearly 60 Chinese and Taiwanese manufacturers of printed circuit boards (PCBs) – the base layers connecting electronic components in all devices – have established operations in Thailand in the past three years. China-based Victory Giant Technology, Nvidia’s leading PCB supplier, opened its second Thai facility less than a year after acquiring its first, while Taiwan’s Gold Circuit Electronics runs its inaugural plant at full capacity. According to Taiwan’s Industry, Science and Technology International Strategy Center, Thailand’s PCB production value is set to grow from US$3.5 billion in 2024 to US$5.6 billion by 2030 as Bangkok aims to make the country the world’s second-largest PCB hub after China. The shift stems from concerns over manufacturing concentration in China and Taiwan (seen as vulnerable to Chinese influence) – which together account for over half of global PCB capacity – amid rising geopolitical tensions. Growing pressure on US tech giants including Microsoft, Amazon, and Apple to diversify critical component sourcing has accelerated relocation efforts by their suppliers.

However, the pace of the expansion has strained Thailand’s nascent PCB ecosystem. Acute talent shortages have resulted in salaries for experienced factory managers doubling over the past two to three years to around 100,000 baht (US$3,100) per month. Combined with a supply chain still reliant on imported equipment and materials, personnel costs now run approximately triple those in China when factoring in lower worker productivity and relocation incentives. While established PCB makers like Victory Giant that secured customer orders before investing can justify higher operating costs with guaranteed revenue streams, newer entrants lacking a solid order pipeline are struggling to stand up to intense price competition. Thailand ultimately faces an uphill battle in achieving the operational efficiencies – particularly labour upskilling – that Chinese and Taiwanese manufacturers have developed within China’s ecosystem over decades. Additionally, uncertain US tariff policies and soft non-AI demand for PCBs risk triggering exits before the ecosystem has a chance to mature.

Chinese wind turbine manufacturers increase presence in India

The trend reflects China’s saturated domestic sector pushing firms overseas

Chinese turbine manufacturers are stepping up their presence in India’s expanding wind energy sector. In September, China’s Envision Group began work on its third manufacturing facility in India, where it already operates a blade plant and a factory for nacelles (part of the turbine that houses components such as the gearbox and generator). The company accounted for 41 percent of turbine orders in India as of 2024. Another major Chinese firm, Sany Group, made a full-scale entry into India in 2024 and intends to expand local production of core components by 2026. India aims to generate 50 percent of its electricity from non-fossil fuel sources by 2030, under which wind power capacity is planned to be boosted to 100GW from 50GW as of November 2025.

The expansion into India by Chinese turbine manufacturers has been driven by intense competition in their home market. Turbine procurement costs in China have fallen by 60 percent over five years through the first half of 2025 due to significant cost cutting, ultimately leading to a drop in the average operating profit margin among Chinese wind manufacturers to around zero. To offset the lack of profitability at home, a number of players have sought to sell their cheap turbines into international markets, where procurement costs have risen 40 percent over the same timeframe.

However, the influx of lower-cost Chinese turbines has caused a major realignment in India’s wind industry where Chinese manufacturers have been able to offer turbines at a roughly 10-15 percent lower price than local players such as Suzlon Energy. In March 2025, Spain’s Siemens Gamesa divested 90 percent of its Indian operations amid mounting pricing pressures. Similar trends are emerging in other markets such as the EU and the US – both of which have launched probes into unfair trade practices. For its part, India has introduced a directive that requires turbine makers to source key components – including blades and generators – domestically. Manufacturers will also have to comply with strict data localisation rules. Despite such measures, India will face an uphill battle in overcoming the deep cost advantage of Chinese manufacturers.