Oracle's Reported $7 Billion Tencent Chip Deal Collapses

By Nathan Brooks

A reported $7 billion Oracle agreement to rent as many as 100,000 advanced AI chips to Tencent for five years has collapsed after U.S. officials raised security concerns. The Financial Times, citing multiple sources, said Oracle withdrew the contract shortly after government officials contacted the company; Oracle, Tencent and the White House have not issued official positions.

What would the Tencent contract have covered?

The deal would have supplied up to 100,000 advanced AI chips through Oracle data centers in several Southeast Asian locations over five years. Its reported $7 billion value, about 9.4 trillion won, would have required Tencent to pay 30% upfront and would have represented the largest overseas semiconductor rental agreement the Chinese company had signed.

This structure would have given Tencent indirect access to advanced AI chips that remained difficult to obtain at home under U.S. export controls and Chinese regulations. The proposed contract value exceeded one-third of Oracle's $19.35 billion fiscal first-quarter revenue.

What triggered the withdrawal?

U.S. officials raised the security implications of giving Chinese firms indirect access to advanced chips through third-country facilities. The report framed the withdrawal as direct intervention to block that route. Separately, senior Guangdong provincial officials visited Tencent and demanded more use of Chinese-made semiconductors, reflecting Beijing's push to reduce major technology companies' reliance on foreign chips.

How does "backdoor leasing" work?

Chinese technology companies have expanded so-called "backdoor leasing," remotely renting computing resources from overseas data centers. The structure can sidestep U.S. export controls and domestic supply shortages, often by placing Nvidia high-performance AI chips in third countries for large-scale model training. Current rules permit overseas leasing, but U.S. companies face pressure to reduce their dependence on Chinese customers.

Since June, the U.S. Department of Commerce has applied the latest AI-chip export controls in full to overseas subsidiaries of Chinese companies established in third countries, including Malaysia. Nvidia has also reportedly required its data-center and cloud customers to sell at least 51% of total service capacity to non-Chinese customers, a measure aimed at blocking overseas access to American-made advanced chips.

What does the setback change for Oracle?

Oracle serves ByteDance in Asia through several data centers in Malaysia. It previously stored U.S. user data for TikTok and participated in the consortium that acquired TikTok's U.S. business late last year. The canceled Tencent contract had offered a way to ease reliance on Oracle's heavily cited OpenAI cloud contract and diversify its customer base.

Oracle shares fell more than 2% in after-hours trading after the withdrawal report, and the stock was down more than 34% from the start of the year. The source also linked investor sentiment to the U.S. midterms, where Democrats outperformed expectations and a candidate critical of data-center expansion won the Texas gubernatorial primary.

Oracle has been winning a string of AI computing contracts while financing an infrastructure buildout. It is issuing $20 billion, about 26.8 trillion won, in new shares and corporate bonds for data centers and infrastructure.

Fiscal first-quarter revenue, announced in September, rose 30% year over year to $19.35 billion, about 25.9 trillion won, ahead of Wall Street's $19.14 billion estimate, about 25.6 trillion won. In that quarter, it secured more than $30 billion, about 40.2 trillion won, in new AI-cloud contracts, lifting remaining performance obligations to $664 billion, about 889.3 trillion won.

Oracle raised its fiscal 2027 adjusted net-income and revenue forecasts. It maintained annual capital expenditure at $70 billion, about 93.7 trillion won.

What is the constraint on Tencent?

Tencent has narrowed the gap with Alibaba's Qwen by improving the performance of its Hunyuan AI model. Analysts say the expansion of training and inference infrastructure could slow as the path to advanced chips narrows. Better domestic-model performance does not remove that capacity constraint.

Washington seeks to curb China's AI development pace while preserving U.S. leadership. Beijing is pressing Chinese technology companies to use domestic chips, creating a dual structure in which corporate cooperation itself becomes harder.

ByteDance and Alibaba maintain large-scale chip-rental contracts with U.S. operators. Those agreements make the reach of regulatory enforcement the key commercial question, because similar limits could constrain several Chinese model builders at once. Authorities have not confirmed whether they issued a specific administrative prohibition order, and it remains unclear whether Tencent will receive a refund of any advance already paid.

The next test is whether U.S. regulators extend similar capacity restrictions to ByteDance and Alibaba, turning one canceled contract into a broader ceiling on Chinese access to overseas AI capacity.