MMarket PulseScopeSolicitud / Contacto
All market news
AI HARDWARE

Tencent Leases 100,000 AI Chips From Oracle in $7 Billion Overseas Compute Deal

A five-year lease for about 100,000 advanced accelerators across Southeast Asian data centres shows how Chinese buyers route around export controls, converting scarce GPU capacity into contracted cloud capacity instead of hardware purchases.

Tencent has signed what is reported to be its largest overseas leasing agreement, gaining access to about 100,000 advanced AI chips from Oracle under a five-year deal spanning multiple Southeast Asian data centres and estimated to be worth around $7 billion, with roughly 30% paid upfront, TrendForce reported on October 1, citing Reuters and the Financial Times. The economic logic is straightforward: while Chinese technology companies cannot buy advanced AI chips directly, US rules do not currently prohibit them from leasing compute capacity hosted outside China, so the deal converts an otherwise prohibited hardware purchase into a service contract and gives Tencent access to advanced NVIDIA accelerators that are generally unavailable to its domestic operations.

The spending behind that capacity is accelerating fast. According to Tencent's own results filing, second-quarter capital expenditure jumped 176% year on year to RMB 52.8 billion, and free cash flow turned negative at RMB 13.8 billion as large AI-related prepayments hit. TrendForce noted that Tencent is simultaneously expanding AI integration across products including WeChat, and that its latest Hunyuan models have improved substantially in recent months, narrowing the gap with leading domestic models from developers such as DeepSeek and Alibaba. For hardware suppliers, the important detail is that this demand arrives as contracted multi-year commitments rather than spot purchases, which lengthens order books even though no chips change hands inside China.

Tencent is not alone in the workaround. Tom's Hardware, citing the Financial Times, reported that ByteDance's Singapore-based subsidiary Spring gained access to 2,304 NVIDIA B200 GPUs through UK cloud provider Nscale's data centre in Norway, another example of Chinese groups reaching restricted hardware through offshore infrastructure. Both arrangements pull scarce accelerator capacity into tightly held, long-dated contracts that sit outside normal spot-market pricing, which reduces the volume available to merchant buyers and makes quoted accelerator and AI-server pricing less representative of what incremental capacity actually costs.

Sources