Tencent Leases 100,000 GPUs From Oracle Outside China
3 min readTencent has found a way to train on chips it cannot buy at home. The Chinese tech giant has signed a five-year lease for roughly 100,000 GPUs inside Oracle data centers in Southeast Asia, according to a Financial Times report cited by Data Center Dynamics. The Tencent Oracle GPU deal is valued at about $7 billion, with 30 percent paid upfront.
Why Tencent Needs Oracle
US export controls bar the most advanced AI accelerators from being sold into China. That leaves Chinese labs with a choice: wait for domestic silicon to catch up, or rent restricted hardware wherever it is legally hosted. Renting capacity in a third country has become a quiet but growing workaround, and Southeast Asia, with its new wave of hyperscale campuses, sits closest to home.
Tencent builds the Hunyuan family of models and a growing set of agentic tools. Its executives have said repeatedly that compute, not talent or data, is the constraint on how fast those models improve.
The Deal
Two people familiar with the agreement told the FT that the lease spans multiple Oracle facilities in the region and runs for five years. The hardware is not available inside China under current rules. Oracle and Tencent had not commented at the time of the report.
The scale of the commitment showed up in Tencent’s books before the deal was public. The company reported negative free cash flow of RMB 13.8 billion, about $2.06 billion, in its second quarter, its first negative figure in more than a decade. Chief financial officer John Lo attributed it to large AI infrastructure spending and AI-related prepayments, and said cash flow would have been RMB 37.6 billion without the compute prepayments. That gap of roughly $7 billion to $8 billion lines up with the reported Oracle contract.
President Martin Lau has defended the spending as a bet with limited downside. If Tencent ever has more capacity than it can use, he said, it can rent the infrastructure out through Tencent Cloud at cost recovery or better. For now the priority is training bigger Hunyuan models, then inference, then renting out bare metal or model-as-a-service capacity.
Why It Matters
A $7 billion lease is one of the largest offshore compute arrangements by a Chinese company on record, and it tests how far export controls really reach. Washington has focused on where chips are shipped. Deals like this shift the question to who is allowed to rent them, and from where.
For Southeast Asia, it confirms the region’s role as neutral ground in the AI compute race. For Oracle, it is another anchor tenant for a cloud business that has been growing on exactly these kinds of large, long-dated contracts. Watch for whether US regulators respond with rules on remote access, and whether Alibaba and ByteDance follow Tencent offshore.