Tencent reportedly rents 100,000 Oracle GPUs abroad that US export controls keep out of China

On Tencent's second-quarter earnings call, chief financial officer John Lo explained a negative cash flow of RMB 13.8bn ^[10]. According to the FT, it was the company's first negative figure in more than a decade ^[10]. Lo said the quarter was "reflecting large AI infrastructure capex and AI-related prepayments" ^[11]. He added that "excluding the prepayments for compute procurement, free cash flow would have been RMB 37.6bn" ^[12]. Using DCD's dollar conversions, the gap between those two figures is about $7.7bn ^[19].

The usual case for renting compute is that it keeps cash free and lets you walk away when your own chips arrive. The Oracle lease, as the FT described it, commits Tencent for five years across multiple data centers ^[1]. Spread evenly, the roughly $7bn price averages $1.4bn a year ^[16].

Divided across 100,000 chips, the reported price is about $70,000 per GPU for the term, or $14,000 a year ^[18]. Run around the clock, each GPU would cost roughly $1.60 an hour ^[18]. Reuters described the hardware only as advanced AI chips ^[6].

Reuters placed the deal among Chinese firms racing to secure AI chips as US export curbs tighten and Beijing promotes home-grown alternatives ^[9]. For Tencent, the route is location. The chips stay in Oracle's facilities in Southeast Asia, and Tencent uses them there ^[1]. Neither report says whether US officials have examined arrangements of this kind.

Tencent president Martin Lau said the company has "clear downside protection" on its AI spending ^[8]. "The AI investments we're making are mostly in AI infrastructure, and in the worst case, which we do not believe would happen, we can choose to rent that infrastructure out at cost recovery or even better prices via Tencent Cloud if needed," he said ^[8]. Applied to the Oracle GPUs, that fallback would mean re-renting capacity Tencent itself leases, inside another company's data centers, for whatever remains of a five-year term ^[1].

The strategy side has its own timeline. In May 2026, chief strategy officer James Mitchell said Tencent had been "consciously late to monetize the AI opportunity through Tencent Cloud," and that "as the supply of China-designed GPUs progressively ramps up, we'll be remedying that situation" ^[14]. The Oracle lease was agreed in 2026 for five years, according to the FT as carried by Reuters ^[7].

An operator weighing any compute contract can place it with two numbers. The first is the share of the price paid before the compute is used. The second is the length of the term measured against the date your preferred supply arrives. A small prepayment on a term shorter than that wait is renting in the sense teams usually mean. A large prepayment on a term that outlasts the wait is close to buying, with the hardware sitting in someone else's facility. Tencent's reported deal sits at about 30 percent upfront over five years ^[1][2].