China's AI firms drive Southeast Asia data centre boom amid US chip export curbs

Business Times

09-05-2025

[HO CHI MINH CITY] Chinese firms are ramping up artificial intelligence (AI) infrastructure leases across South-east Asia in a long-term push to globalise compute workloads as tighter US controls on Nvidia's advanced chips drive demand to digital hubs such as Singapore, Malaysia, Indonesia – and increasingly Vietnam and Thailand.
Industry sources note that Chinese hyperscalers such as Alibaba Cloud, Tencent and Huawei have also been steadily deepening their presence in South-east Asia through joint ventures (JVs) as they seek to sidestep geopolitical friction and secure scale.
Mickael Driol, chief executive at Mekong Partners, a Shanghai-based advisory firm specialising in corporate services, said: 'We're already seeing targeted capacity reservations by Chinese clients that stretch five to seven years into the future, particularly for AI-focused data centres in Malaysia and Indonesia.'
Vietnam and Thailand are also gaining traction alongside established data hubs amid this rising influx of Chinese AI demand, Driol added.
According to Dedi Iskandar, Asia-Pacific regional director at datacenterHawk, a global data centre real estate analytics platform, demand from China began surging in the third quarter of last year, with investor interest in Johor, Malaysia, and Batam, Indonesia, fuelled by favourable regulations, lower costs and generous subsidies.
Throughout 2024 and into 2025, both have experienced data centre investment booms, driven by spillover from Singapore's temporary moratorium on new developments. This had happened particularly within their respective Special Economic Zones that offer attractive incentives and strengthened infrastructure in terms of submarine cable connectivity and electricity supply.
A NEWSLETTER FOR YOU
Friday, 8.30 am Asean Business
Business insights centering on South-east Asia's fast-growing economies.
Sign Up
Sign Up
The shift is being felt on the ground by regional operators, who are seeing an uptick in enquiries and bookings from Chinese clients shut out of direct server access.
'We actually see a lot of interest in China, because people there would no longer get their hands on servers, and that has led to a big demand for data centres and AI infrastructure (in the region),' said Zhao Jiajun, ST Telemedia Global Data Centres (STT GDC) senior business development director of South-east Asia, at an event last month in Ho Chi Minh City, Vietnam.
Headquartered in Singapore, STT GDC operates data centres globally, including facilities in Indonesia, the Philippines, Malaysia and Vietnam.
Strategic nodes of China's AI clout
After the US imposed its first export controls on AI chips to China in October 2022, Chinese buyers allegedly turned to third-party traders in Taiwan and South-east Asia to acquire Nvidia's high-end graphics processing units (GPUs). They include the Blackwell and H100 chips, which are globally sought after for their ability to accelerate AI workloads.
However, growing regulatory scrutiny has prompted Chinese firms to adopt indirect access models. Instead of importing restricted chips, they are now leasing AI compute from foreign providers that legally operate high-end hardware.
Regional infrastructure providers, say market watchers, are increasingly acting as neutral intermediaries, often structured through holding companies in Singapore or Jakarta to deliver GPU-as-a-service and AI-as-a-service solutions. Their geographical proximity helps ensure low latency and operational efficiency for Chinese clients.
According to Nvidia's latest filings, China accounted for roughly 13 per cent of the US chip giant's 2024 sales, down from 21 per cent in 2022.
In contrast, revenue billed through Singapore – used as a centralised invoicing hub – rose to 18 per cent, up from 8.5 per cent two years earlier. With actual uptake in Singapore accounting for less than 2 per cent, Nvidia noted that products were 'almost always shipped elsewhere'.
'Some (Chinese) hyperscalers have also expanded quietly through joint ventures with local telcos or real estate investors to avoid geopolitical friction while gaining scale quickly,' Driol added, referring to the likes of Alibaba Cloud, Tencent Cloud and Huawei Cloud.
He also observed the increasing demand from China-based AI labs and machine learning startups in fields such as online education, fintech, gaming and industrial automation, which are looking to train foundation models abroad by leveraging US-sourced GPU clusters located within more geopolitically stable ecosystems in Asean.
'This is reshaping the digital map of South-east Asia, turning previously secondary locations into strategic nodes of China's AI and cloud expansion plans,' he added.
Long-term shift
While China is a major consumer of AI chips, its domestic production of advanced AI chips is limited. Experts believe that as long as this imbalance persists, China's appetite for overseas compute power will continue to grow, especially in the light of its push for generative AI, autonomous driving and robotics.
'With China's domestic supply bottlenecked and hyperscaler capacity in Beijing and Shenzhen already oversubscribed, the logical overflow will continue to be routed through data centres in Asean,' Driol stated.
Unlike AI inference workloads, which require processing servers located closer to the point of consumption, AI model training is location agnostic.
Mayank Srivastava, chief executive at BDx Data Centers, said: 'It does not depend on the location. What matters most is the service cost and renewable power supply.'
Headquartered in Singapore, BDx extends its AI infrastructure footprint across Indonesia and Greater China. In April, BDx announced that its 500-megawatt CGK4 facility in Indonesia was among the archipelago's first data centre parks to achieve Nvidia certification to deliver scalable and high-performance infrastructure for AI and advanced computing workloads.
Srivastava declined to disclose BDx's customer portfolio of the AI-specialised build, but he confirmed that it is serving clients both within and outside of Indonesia.
It is also worth noting that trends in terms of data localisation and digital transformation are creating growing domestic needs for compute resources, either from local AI-focused startups or global hyperscalers seeking to enhance end-users' digital experience by utilising edge infrastructure inside the country.
While shifting geopolitical dynamics have raised concerns about short-term disruptions in some countries' access to critical US-supplied chips, industry insiders remain optimistic that these challenges will be resolved within a few years.
'AI is going to be like a utility,' said Srivastava. 'People will find a way. Over a period of two to three years, you will find a lot of democratisation happening for all of the infrastructure, especially chip hardware that is required for AI.'
This is attributed to rapid technological advancements, exemplified by China's development of the AI chatbot DeepSeek with lower cost and less advanced chips compared with its US equivalent ChatGPT.
'Governments have taken note of AI; they realise that it's better to be prepared for AI than lose out in the race,' he said.
Additional reporting by Tan Ai Leng in Kuala Lumpur