A reported $7 billion, five-year cloud infrastructure agreement between Oracle and Chinese technology giant Tencent is highlighting a growing complication in the US-China AI chip race: restricting where advanced AI chips are physically shipped does not necessarily prevent Chinese companies from accessing their computing power remotely.
According to a Financial Times report, Oracle has agreed to provide Tencent access to approximately 100,000 advanced AI chips through data centres in Southeast Asia. The arrangement would represent Tencent’s largest overseas infrastructure lease and could give the Chinese company access to high-performance AI computing hardware that is otherwise difficult to obtain inside China.
The reported deal comes at a time when Washington is tightening controls around the export, reexport and transfer of advanced computing chips and AI systems to China and certain Chinese-headquartered entities.
Oracle-Tencent Deal Highlights a New AI Compute Route
The reported arrangement is significant not simply because of its size, but because of where the computing infrastructure is located.
Rather than shipping restricted AI accelerators directly to China, the chips can be deployed in data centres in countries where their export is permitted. Chinese companies can then potentially rent computing capacity from those facilities and send AI workloads to the hardware remotely.
This creates a different model for accessing advanced AI infrastructure.
Under such an arrangement, the physical GPU may remain outside China while the computing workload originates from a Chinese company. The distinction between owning a chip and accessing its computing capacity therefore becomes increasingly important in the global AI infrastructure race.
Oracle’s extensive cloud and data-centre footprint across Southeast Asia provides a potential foundation for this model.
Why Southeast Asia Is Becoming Critical to AI Compute
The reported Tencent deal also reflects a broader shift in the geography of AI infrastructure.
As access to the most advanced AI accelerators becomes more restricted within China, Chinese technology companies have increasingly looked outside the country for additional computing capacity.
Markets including Malaysia, Singapore, Indonesia and Thailand have emerged as important destinations for data-centre investment and cloud capacity. Chinese companies are reportedly reserving substantial amounts of capacity across the region to support growing AI workloads.
For these companies, overseas infrastructure can provide access to computing resources without requiring advanced accelerators to be physically installed inside China.
This is particularly relevant as AI development becomes increasingly dependent on enormous amounts of GPU capacity for model training, inference, reasoning and agentic applications.
US Chip Controls Face a Remote-Access Challenge
The development also exposes a complicated question for US technology controls.
Washington has imposed restrictions designed to limit China’s access to advanced computing chips and systems, particularly high-performance AI accelerators developed by companies such as NVIDIA.
However, controlling the physical destination of a chip is different from controlling who can ultimately use its computing power.
If an advanced accelerator is legally installed in a permitted country, a cloud provider or data-centre operator can potentially make that hardware available as an infrastructure-as-a-service resource. A customer does not necessarily need to physically possess the GPU to use it.
A recent Reuters legal analysis described a similar scenario involving a Chinese company potentially renting infrastructure at a Malaysian data centre and remotely running workloads on advanced computing hardware located there.
The issue is important because current US rules do not necessarily amount to a blanket prohibition on ordinary remote access to every advanced computing resource located outside China.
That leaves regulators facing a more complicated problem: how do you restrict access to computing power when the hardware itself can remain thousands of kilometres away?
Tencent’s AI Ambitions Are Driving Demand
For Tencent, access to large-scale AI computing capacity comes as the company continues to expand its artificial intelligence portfolio.
The company is developing the Hunyuan family of AI models and has been pushing deeper into reasoning, coding and agentic AI.
Tencent’s AI products include WorkBuddy, CodeBuddy and Yuanbao, reflecting a broader industry transition from conventional generative AI tools toward systems capable of completing multi-step tasks.
Its newer Hy3 and Hy4 models have also been developed and optimised around areas such as reasoning, coding and agentic workloads.
These applications require substantial computing resources, particularly as AI systems become more capable of handling longer, more complex tasks.
For companies operating at Tencent’s scale, access to advanced GPUs is therefore becoming not merely a technology question but an infrastructure and strategic capacity issue.
Tencent Is Not the Only Chinese Tech Giant Looking Overseas
The reported Oracle agreement fits into a wider trend involving China’s largest technology companies.
ByteDance and Alibaba are also among the major Chinese users of data-centre capacity in Southeast Asia. Their overseas infrastructure activity reflects the increasing importance of regional cloud markets as Chinese companies seek additional computing resources.
Southeast Asia is consequently becoming an increasingly important bridge between the world’s AI infrastructure supply and China’s growing demand for computing power.
The region already has several advantages: expanding data-centre markets, growing cloud adoption, proximity to China and increasing investment in digital infrastructure.
As AI workloads continue to expand, these factors could make Southeast Asia an increasingly important part of the global AI computing ecosystem.
A New Chapter in the Global AI Chip Race
The reported Oracle-Tencent agreement illustrates how quickly the AI infrastructure landscape is evolving.
US export controls can determine where advanced chips can be shipped, but cloud computing introduces another layer: where the customer can access the computing power generated by those chips.
That distinction could become increasingly important as AI companies move toward massive infrastructure requirements.
For cloud providers, the opportunity is substantial. For regulators, it creates a more complicated enforcement challenge. And for Chinese technology companies, overseas data centres could provide an important source of computing capacity at a time when access to cutting-edge AI hardware inside China remains constrained.
The reported 100,000-chip Oracle-Tencent arrangement, if completed as described, therefore represents more than a major cloud lease. It underscores a fundamental shift in the AI economy—from the question of who owns the chips to the increasingly important question of who can access their computing power.
Read more: Google DeepMind Develops ‘Invisible’ Watermark to Track AI-Generated Proteins
