Tencent signs $7 billion deal with Oracle for 100,000 AI chips

Five-year lease for advanced processors in Southeast Asia highlights Chinese firms' workaround to export controls

By LineZotpaper
Published
Read Time2 min
Tencent has reportedly struck a five-year deal worth about $7 billion with Oracle to lease approximately 100,000 advanced AI chips that are not available in China, according to the Financial Times. The arrangement places the processors in Oracle data centers across Southeast Asia, allowing the Chinese tech giant to access cutting-edge hardware for AI training while staying within current U.S. export rules.

The deal, which has not been confirmed by either company, is estimated to cost about $1.60 per chip per hour over the lease term, with roughly 30% of the value paid upfront, the Financial Times reported. Tencent's chief strategy officer, James Mitchell, noted on an August earnings call that compute rental prices continue to climb. Tencent president Martin Lau said on the same call that the company could sell existing chip orders at "more than 30% profit" compared to what it "paid just a few months ago."

The specific model of chips has not been disclosed, but the Financial Times reported that such leases could involve Nvidia's top processors. Tom's Hardware analysis suggests the pricing aligns more closely with Nvidia's Hopper architecture (H100 and H200) rather than the newer Blackwell line. H200 imports into China remain limited, meeting the condition that the chips in the deal are unavailable there.

Oracle's data centers are located outside China, and details including the countries, specific sites, and start date remain undisclosed. Oracle has two cloud regions in Singapore and has announced one in Malaysia. The Financial Times noted that the biggest clients of Southeast Asian data centers are ByteDance and Alibaba, with Chinese firms using the capacity especially for AI training, which domestic chips cannot yet do.

Tencent had previously signed a deal worth over $1.2 billion for Datasection Blackwell rentals in Japan and Australia, the FT reported in December 2025. Oracle's fiscal Q4 saw $67 billion in AI contracts added, and the company received $11.4 billion in prepayments from customers in the three months through August, according to its quarterly filing.

§

Analysis

Why This Matters

  • Chinese technology companies continue to secure advanced AI hardware despite U.S. export restrictions, potentially narrowing the gap in AI capabilities between China and the West.
  • The rising cost of compute rental, as noted by Tencent executives, signals increasing demand and supply constraints for AI chips, which may affect the economics of AI development globally.
  • This deal demonstrates a growing trend of Chinese firms leasing overseas data center capacity rather than importing chips directly, a workaround that could prompt further regulatory scrutiny.

Background

U.S. export controls imposed in recent years have restricted the sale of advanced semiconductors and chipmaking equipment to China, aiming to slow the country's military AI advancements. Chinese firms have responded by renting compute power from cloud providers in Southeast Asia and other regions where such restrictions do not apply. Data centers in Singapore, Malaysia, and Japan have become key hubs for this purpose. The demand for AI training compute has surged, driving up prices for GPU rentals and leading to long-term, large-value contracts like this one.

Key Perspectives

Tencent: Secures access to high-end chips for AI model training and inference, maintaining competitiveness in AI while avoiding direct import bans. The company can also profit from reselling existing orders. Oracle: Gains a major customer for its cloud infrastructure, locking in significant revenue and strengthening its position in the AI compute market. U.S. government: Must balance export control enforcement with the reality that Chinese firms can still access advanced chips via cloud services, potentially leading to tightened rules or new restrictions on cloud-based compute. Competing Chinese firms (ByteDance, Alibaba): They are also major clients of Southeast Asian data centers, and such deals could become more common, intensifying competition for limited overseas compute resources.

What to Watch

  • Whether the U.S. government updates export control rules to address cloud-based access to advanced chips.
  • The trajectory of GPU rental prices, which have been increasing; a stabilization or drop could affect the viability of such long-term deals.
  • Other major Chinese AI players, such as ByteDance and Alibaba, may announce similar large-scale lease agreements, further straining global GPU supply.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.