Oracle has canceled its agreement to lease AI chips from Tencent worth $7 billion.

Oracle canceled an AI chip leasing agreement with Tencent, worth approximately $7 billion.

This agreement was signed by Tencent this year and has a term of five years. According to the original plan, Tencent will lease approximately 100,000 advanced AI chips through Oracle’s data centers in Southeast Asia, with the initial payment accounting for around 30% of the total cost.

In other words, this is a detour: the chips are not sent to China but are placed in data centers overseas, allowing Tencent to remotely rent computing power. Such arrangements have always been situated in the gray area of export controls.

Later, American officials expressed their concerns regarding this transaction to Oracle, and Oracle subsequently faced pressure from the Trump administration, ultimately deciding to cancel it.

At the very same time, another piece of news presented a completely opposite scenario: According to reports, the US Department of Commerce has approved approximately 10 Chinese companies to purchase Nvidia’s H200 graphics processing units. Each company is allowed to acquire up to 75,000 units, but so far, no deliveries have been completed yet.

On one hand, these policies allow the sale of chips; on the other hand, they prevent the use of overseas rental computing power. The logic behind such policies seems to be to ensure that all transactions remain under control: sales are permitted, but only through channels that are clearly monitored and approved by the authorities.

For large domestic companies, pursuing the option of renting computing power overseas can basically be considered a dead end. The remaining choices are to wait for permits to be issued, stock up on relevant resources, or accelerate the transition to using domestically produced graphics cards.

The situation regarding oracle bone inscriptions is no less challenging. Those data centers in Southeast Asia were originally designed to serve this large-scale order. Now that the client has withdrawn its request, the available computing power needs to be quickly resold to new buyers.