Tencent's Massive Oracle AI Lease
Tencent Reportedly Signs $7 Billion Oracle Lease for 100,000 AI GPUs
Anonymous
Key Points
1
Tencent CFO John Lo said free cash flow would have been 37.6 billion yuan excluding prepayments for compute, attributing the reported negative figure to major AI infrastructure spending and AI-related prepayments.
2
Tencent president Martin Lau said the company sees “clear downside protection” in its AI infrastructure investments: if needed, it could rent the capacity through Tencent Cloud at cost recovery or better.
3
Tencent’s stated plan is to use the leased compute first to train larger Hunyuan models, then for inference workloads, and potentially later offer it as bare-metal capacity or through Model-as-a-Service.
4
Tencent had previously used a similar overseas route, leasing roughly 15,000 Nvidia Blackwell chips in Japan through Datasection under a contract reported to be worth about $1.2 billion.
5
Oracle said most of its new contracts come with prepayments or customers supplying their own hardware; in the same quarter, it reported more than $30 billion in new AI contracts and forecast full-year capital spending of $90 billion to $95 billion.
Tencent has agreed to a roughly $7 billion five-year lease with Oracle for access to about 100,000 advanced AI chips across Southeast Asian data centers, Financial Times reported. The deal marks a major workaround for Chinese tech firms facing U.S. export restrictions that bar powerful chips from entering China. Tencent will pay roughly 30% upfront — about $2.1 billion — to secure computing capacity for training larger Hunyuan AI models.
The lease underscores how tightly U.S. restrictions squeeze Chinese AI development and how aggressively Chinese companies are turning to overseas infrastructure. Oracle declined to confirm the deal, but the company reported more than $30 billion in new AI contracts in its latest quarter and forecast capital spending of $90 billion to $95 billion this year.
How Tencent Plans to Use the Leased Chips
Tencent intends to deploy the 100,000 GPUs in phases, starting with training larger Hunyuan language models, then shifting to inference workloads. Tencent president Martin Lau said the company sees "clear downside protection" in the bet: if needed, Tencent could rent the excess capacity through its cloud service at cost-recovery rates or better. The strategy lets Tencent build its own AI models while monetizing idle compute.
Why China's Tech Giants Turn to Overseas Data Centers
U.S. export controls block most advanced AI chips from reaching China directly. Tencent has already leased roughly 15,000 Nvidia Blackwell chips in Japan through Datasection under a reported $1.2 billion deal. The Oracle lease is the largest such workaround yet, giving Tencent access to cutting-edge compute it cannot easily source at home.
This desperation for overseas chips reflects China's shrinking domestic supply of advanced semiconductors. Tencent CFO John Lo acknowledged the impact: the company posted negative free cash flow of 13.8 billion yuan in the second quarter, primarily due to AI infrastructure spending and prepayments tied to compute leases.
Oracle Prepayments Signal Strong AI Demand
Oracle landed the Tencent deal as part of a broader trend: customers increasingly prepay for compute or supply their own hardware. The company reported more than $30 billion in new AI contracts last quarter and plans to spend $90 billion to $95 billion on capital this year, mostly for data-center buildout. The Tencent lease shows Oracle's infrastructure is competitive even after U.S. trade restrictions.
CFO John Lo noted that excluding the prepayments for compute, Tencent's free cash flow would have been positive at 37.6 billion yuan. The negative result reflects how capital-intensive AI infrastructure has become and Tencent's long-term bet that the investment will unlock strategic value in AI.
Neither Company Confirms the Reported Deal
Financial Times reported the $7 billion figure and terms, but neither Tencent nor Oracle has publicly acknowledged the agreement. The secrecy is typical for deals this large and politically sensitive — U.S. regulators scrutinize tech partnerships involving China, and Tencent may want to avoid drawing attention to how it circumvents chip export rules.
Sentiment
Mostly PositiveNEGATIVENEUTRALPOSITIVE
Publishers
17
Articles
126
Reach
143
A
AnonymousAuthor
