This story was originally published in the EastAsiaForum.
Technologies are never just tools, they are accompanied by patterns of labour, know-how and social relations. In Southeast Asia, unequal ownership structures and geopolitical factors have always shaped technological sovereignty . Deepening this techno-societal divide is growing investment by US technology companies in artificial intelligence (AI) in the region.
The deals struck by US firms such as OpenAI, Meta and Google with telecommunications and digital service providers from India and Southeast Asia demonstrate this trend. Companies like Singapore’s Singtel , Indonesia’s Telkomsel and India’s Bharti Airtel and Reliance Industries provide US technology companies with two fundamental resources — nearly a billion consumers whose usage and behavioural data train and refine their models, as well as local infrastructure, payment interfaces, identity systems and regulatory cover. This integration drives customer acquisition costs towards zero, embeds their models inside existing data flows and shapes governance frameworks.
While 83 per cent of Southeast Asian firms are still in the early stages of AI adoption, telecommunications companies have been early adopters. These companies sell partnerships as enabling their evolution from connectivity providers into AI distribution platforms. States are promised productivity gains and a futureproofed economy while the public is promised access, empowerment and the democratisation of AI . US hyperscalers promote their data centres and large language models as ways to end global poverty , advance equality of access and facilitate innovation.
The public reception has, on the surface, been optimistic. The 2025 AI Index and 2026 Anthropic survey across affluent urban Southeast Asian populations found that AI was more associated with growth and economic opportunity than job displacement and surveillance.
But applications of the AI stack must be analysed beyond an interface running on some ethereal cloud. It must be understood in the context of how they interact with divisions of labour, production methods, trade regimes and geopolitical realities. These deals, dictated by foreign ownership, could carry considerable consequences.
At the model development layer, data harvested through these deals is repackaged and monetised under US technology companies’ control, including via commercialisation through third parties . Despite Southeast Asia’s progress towards data localisation and governance — and the region’s limited development of ‘sovereign large language models’ like Malaysia’s ILMU , Cambodia’s Khmer model and Singapore’s SEA-LION — these deals rely on US companies’ proprietary algorithms and fall under foreign laws like the US Cloud Act. Jurisdiction is dictated by corporate control rather than physical storage and app development, enabling US technology companies to use native data to train models like Gemini, Llama and ChatGPT.
Beneath this sits the infrastructural layer, which involves rare earth extraction , graphics processing unit manufacturing, data centres , energy grids, computational power and satellite connectivity networks . The governance layer sets the assumptions by which infrastructure, models and applications are built, trained and integrated into governance institutions.
Central to each stage is the ‘ hidden abode of AI ‘, labour. The exploitation of miners, image taggers and gig workers demonstrates the power of monopoly finance and platform capital, both of which extend their control of telecommunications companies via joint infrastructure investments and direct ownership . AI is projected to impact 57 per cent of Southeast Asia’s workforce.
Ownership of model development and infrastructure allows US technology companies to export high-value goods to the United States, while Southeast Asia receives minimal returns, only gaining 6–18 months of free app access.
Local telecommunications conglomerates, with their own billionaire-class and monopolistic ambitions , operate as intermediaries between core and periphery, with limited control over the models, as demonstrated by Washington’s suspension of foreign access to Anthropic’s Mythos .
Asian multinationals also play a role in building the material parts of the AI stack. Samsung, SK Hynix and Naver lead South Korea’s US$1 trillion drive to produce semiconductors, physical AI and data centres. Other projects include Japan’s US$180 million GENIAC Program , Japanese company Softbank and Chinese transportation service provider Didi’s US$2 billion investment in Grab and Indian conglomerate Adani’s US$100 billion investment into AI infrastructure.
The rise of this semi-periphery ‘pole’ will not necessarily lead to technological sovereignty . Regional capital remains tethered to foreign technology, investment and production networks , with local economies unable to retain surplus value. Regional companies can only maintain profit levels by exploiting local natural resources and labour power, which applies downward pressure on wages.
As this uneven reality becomes apparent, optimism is subsiding, even among the region’s urban and affluent populations. The 2026 AI Index Report recorded a 14 percentage point rise in ‘nervousness’ about AI in India, with majorities in Malaysia, Singapore and Thailand also expressing ‘nervousness’.
Neither US technology companies, local capital, their partnerships nor the regional ‘ third way ‘ offer a lasting solution. The latter advocates for a public–private consortium pooling compute, model co-development and talent to provide an alternative to AI monopoly. Yet it risks only challenging who extracts value , rather than whether extraction, exploitation and control continue. The ‘third way’ remains silent on the question of private ownership and the role of local corporations and regional powers whose aspirations mirror those of US technology companies.
Production and the AI stack must be reorganised so that local populations, not US technology companies and their regional collaborators and competitors, own and govern AI. This requires critically rethinking the private appropriation and centralised control of AI’s core layers and bringing their productive capacities under collective ownership, democratic planning and social use.
Govand Khalid Azeez is Lecturer in Politics and International Relations at the School of International Studies
