Tencent is reportedly preparing a deal worth roughly $7 billion with Oracle to secure AI compute capacity outside mainland China, according to a TechRepublic report. The arrangement would place Nvidia-powered AI chips in Oracle's Southeast Asian data center footprint rather than on Chinese soil.

The reported structure matters as much as the number. US export controls prohibit shipments of Nvidia's most advanced AI accelerators — the H100, H200 and B200-class products — to customers in mainland China. Chips hosted in third countries, however, sit in a legal gray zone that US regulators have moved to close: Washington has pressured cloud providers and Middle Eastern operators to guarantee that controlled compute is not effectively resold or accessed from China.

Oracle is one of the few Western hyperscalers with a substantial track record of leasing out large GPU clusters to third parties, a model it has already used for OpenAI's multibillion-dollar compute commitments. A Southeast Asia-hosted arrangement with Tencent would extend that playbook: Tencent pays for compute, Oracle owns and operates the silicon, and the chips themselves never enter Chinese jurisdiction.

Tencent needs the capacity. The company is one of China's three largest cloud and AI players, alongside Alibaba and ByteDance, and it builds its Hunyuan family of large language models on top of GPU clusters that have grown increasingly constrained since the first round of US chip restrictions in October 2022. Nvidia's China-specific alternatives — the H20 and its successors — carry sharply reduced performance, and reports of cumulative Chinese orders for downgraded parts running into the millions of units underscore how far the compliant silicon falls short of demand.

For Oracle, a $7 billion contract would rank among its largest infrastructure deals and would deepen a business line that investors already credit for its re-rating over the past two years. Oracle has committed tens of billions of dollars to GPU capacity expansion across the US, Europe, Japan and the Middle East through its Oracle Cloud Infrastructure arm.

The deal, if concluded, would also test the durability of the workaround itself. The Commerce Department has shown willingness to pursue Chinese firms that access controlled compute through overseas intermediaries, and any escalation — stricter rules on US cloud providers serving Chinese customers, or pressure on host governments in Malaysia, Singapore or Indonesia — could unwind the commercial logic of the arrangement. Tencent has not confirmed the talks, and neither company has publicly commented on the reported terms.

Neither the precise chip mix, the host country, nor the contract duration appears in the report, and the $7 billion figure itself remains unconfirmed by either party. What is clear is the direction: China's AI leaders are increasingly willing to spend at hyperscaler scale abroad to keep training runs competitive, while Washington's rules continue to push that spending into offshore structures rather than eliminate it.

If the Oracle deal closes, expect rival Chinese cloud providers to pursue similar arrangements — and export-control enforcement to shift from chips in crates to compute contracts on paper.