Tencent has leased 100,000 chips from Oracle to accelerate its artificial intelligence buildout, the Financial Times reported, citing people familiar with the arrangement. The deal gives one of China's largest cloud and gaming companies access to substantial AI compute capacity without directly purchasing advanced processors that face US export restrictions.

The Financial Times did not specify in the Reuters summary the exact processor models involved, the financial terms of the lease, or the duration of the arrangement. Oracle declined to comment and Tencent did not respond to requests for comment, according to the report.

The structure of the deal matters as much as its size. Under US export controls tightened since 2022 and expanded in 2023 and 2024, Nvidia's most advanced AI accelerators — including the A100, H100 and subsequent H200 and B200 families — cannot be sold to Chinese customers. However, chips already deployed inside data centers outside China can still serve Chinese clients through cloud rental arrangements. By leasing compute from Oracle, Tencent taps into exactly this mechanism: the hardware sits in Oracle's facilities, and Tencent rents its capacity rather than importing the silicon.

Oracle has spent heavily to build out AI infrastructure. The company has signed multi-billion-dollar GPU contracts with Nvidia and hosts large clusters of Nvidia accelerators in its US data centers, which it rents to customers including OpenAI, xAI and other AI developers through its Oracle Cloud Infrastructure unit. Adding Tencent as a lessee extends that customer base to a Chinese hyperscaler that cannot buy comparable hardware outright.

For Tencent, the 100,000-chip lease addresses a concrete bottleneck. The company, which operates the Hunyuan family of large language models and sells AI compute through its Tencent Cloud unit, has faced the same constraint as domestic rivals Baidu, Alibaba and ByteDance: access to top-tier training hardware is capped by export policy, and domestic alternatives such as Huawei's Ascend accelerators do not yet match peak Nvidia performance at equivalent scale. Leasing overseas capacity lets Tencent keep training and inference workloads running while it waits for the domestic supply chain to mature.

The reported scale is significant in absolute terms. One hundred thousand data-center GPUs would rank among the larger single-customer deployments disclosed anywhere, comparable in order of magnitude to the clusters that frontier AI labs have assembled in the United States over the past two years. The Financial Times report did not break down whether the figure covers a single Oracle data-center region or spans multiple sites, nor did it state the chip generation.

The deal also illustrates a growing commercial pattern in the AI supply chain: compute as a tradable, leasable commodity that flows across jurisdictional lines even when the underlying hardware cannot. US regulators have watched this channel closely. Proposals have circulated in Washington to close the cloud-rental loophole that lets Chinese firms access restricted US compute, though no rule reported in connection with the Tencent-Oracle arrangement has been finalized.

That regulatory overhang is the main risk hanging over the deal. If the Commerce Department moves to restrict Chinese companies' access to US-hosted AI compute, leases of the kind Tencent has reportedly signed could face curtailment or non-renewal. For now, no such restriction has been confirmed, and the arrangement as reported gives Tencent near-term headroom for its AI roadmap.

Neither company has publicly confirmed the lease, and the figure of 100,000 chips rests on the Financial Times' sourcing. If the report holds, it marks one of the largest known instances of a Chinese tech giant renting restricted-class AI hardware from a US cloud provider — and a signal that, until export rules change or domestic chips catch up, Chinese AI labs will keep buying their compute by the hour rather than by the wafer.