Guangxi provincial officials report 44 AI cooperation agreements with ASEAN countries and joint large‑model research with eight.
Guangxi and the China‑ASEAN AI Application Cooperation Centre
China’s effort to embed itself in Southeast Asia’s AI infrastructure is consolidating around place as much as product. The first ministerial roundtable on AI between China and ASEAN, held in September 2025, led to the launch of the China‑ASEAN AI Application Cooperation Centre in Nanning, Guangxi; a follow‑up meeting in September 2026 brought representatives from all 11 ASEAN members. At the fourth China‑ASEAN AI Cooperation Forum on 16 September organisers unveiled five outputs, including a Lao‑language model for Laos’s state electricity corporation — described by Chinese officials as their first overseas AI project in the power sector.
That Lao project shows the delivery model plainly: compute and core infrastructure remain in Guangxi while an adapted application runs in Laos. The service is local, the infrastructure is not.
Huawei, Telkomsel, ZTE, Tencent and ByteDance: AI arriving through networks and cloud
Chinese companies are embedding AI into the telecommunications and cloud layers that already connect Southeast Asia. Huawei has long worked with operators from Thailand to Brunei, and AI is being introduced as an extension of those relationships. At Mobile World Congress 2026, Huawei and Indonesian operator Telkomsel announced an agreement that combines Indonesian 5G and broadband upgrades with embedded AI; Telkomsel also expanded cooperation with ZTE on AI‑native networks.
Elsewhere, Chinese firms are building out data‑centre capacity and approvals: Tencent has outlined further data‑centre investment in Indonesia, and ByteDance’s TikTok secured approval to expand data infrastructure in Thailand. None of these moves requires a standalone decision to buy a Chinese large language model — AI can arrive as part of a telecom contract, a cloud agreement or a routine software upgrade, deepening practical reliance on a single ecosystem.
This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildCommercial scale, division of labour and the durability of dependence
The financial picture is modest but organized. Organisers of the 23rd China‑ASEAN Expo reported nearly 100 preliminary agreements worth around US$150 million (A$210 million). Officials describe a division of labour: research centred in Beijing, Shanghai and Guangzhou; integration in Guangxi; and application across Southeast Asia.
Cost and delivery advantages make Chinese suppliers commercially attractive. The more immediate concern is not espionage theatre but the practical problem of dependency: an operator that relies on proprietary AI network management faces switching costs; a utility using externally hosted models has limited visibility into how those models are updated or secured. The central question is operational: who can audit, maintain, modify and replace the technology once it is embedded?
Open‑weight models, Sea‑lion v4.5, and the limits of model sovereignty
Software trends are reinforcing the hardware and network shifts. Open‑weight models have shown capable performance at low cost: DeepSeek demonstrated that such models could be offered affordably, and Alibaba’s Qwen has become an important foundation for development beyond China. In May, Singapore released Sea‑lion v4.5, a collection of large language models optimised for use in Southeast Asia that included models adopted from China’s Qwen and Gemma. That release shows how Chinese model architectures can enter a wider ecosystem without implying a political takeover: sovereignty at the model layer does not automatically equal independence across chips, cloud infrastructure or technical support.
Governments can swap a model without changing where the compute sits or who provides the chipsets and maintenance — and those deeper dependencies are the ones that are hardest to reset.
What this means for technologists, policymakers, and utilities
- Technologists and security teams: expect AI functionality to be introduced through network and cloud upgrades. The practical risk is loss of auditability and control when core compute is hosted abroad; teams will need to identify who maintains code, who pushes updates and how to assess supply‑chain and operational risk.
- Policymakers and regulators: ASEAN’s Digital Economy Framework Agreement, concluded in May, will shape cross‑border data flows and digital trade; Malaysia’s sovereign computing decisions will test how governments balance Chinese technology against US restrictions. Those policy choices will interact with procurement defaults made at the telecom and cloud levels.
- Utilities and procurement leaders: the Lao electricity example and Malaysia’s May 2025 episode — when Deputy Communications Minister Teo Nie Ching described plans involving Huawei’s Ascend AI hardware and DeepSeek, only for the trade ministry to distance the government within two days and for Huawei to say it had not sold Ascend chips in Malaysia — underline the confusion that can follow blurred public‑private deals. Organisations should map where compute physically resides, who controls updates, and what switching would cost.
China’s footprint is growing through commercially attractive investments, infrastructure upgrades and technical partnerships. Governments may think they are avoiding a binary choice between Chinese and Western AI, but many of the most consequential choices are being made at the level of networks, chips and data centres. Once those dependencies are in place, the analysis and the options look very different — and changing course can become much harder.
