Oracle–Tencent Deal Canceled Under U.S. Pressure, Deepening Uncertainty Over Southeast Asian Data Center Investment

U.S. expands pressure on overseas transactions to block China's AI chip procurement
Collapse of major leasing deal disrupts plans to attract Chinese demand
Delays in securing new customers could prolong investment recovery for Southeast Asian data centers

The United States is tightening restrictions on China's access to advanced artificial intelligence (AI) chips. Oracle canceled a major chip-leasing agreement with Tencent under U.S. government pressure, and Nvidia has reportedly asked its cloud customers to limit the share of new business they conduct with Chinese companies. Legal proceedings are also underway against those who illegally supplied AI servers to China through third countries. Southeast Asian operators that expanded their facilities in anticipation of leasing demand from Chinese tech giants now face the twin challenges of finding replacement customers and recouping their investments.

USD 7 Billion AI Chip-Leasing Deal Falls Through

According to the Financial Times (FT) and Reuters on the 5th (all dates hereafter are local time), U.S. cloud company Oracle recently canceled a USD 7 billion AI chip-leasing agreement with Chinese tech giant Tencent. Signed earlier this year, the agreement was Tencent's largest overseas leasing deal and centered on securing large-scale AI computing resources through Oracle's data centers in Southeast Asia. The size and key terms of the deal were first disclosed in late September, when Tencent was increasing investment in computing resources needed to develop AI models and expand services. News that the deal had fallen through emerged roughly a month after its signing became public.

The original agreement was structured to give Tencent access to as many as 100,000 advanced AI chips installed at multiple data centers in Southeast Asia for five years. Tencent planned to access chips at overseas facilities operated by Oracle and use their computing resources to develop AI models and agents, among other purposes. The cancellation of the long-term lease for a large volume of chips has disrupted those plans to secure computing resources. Given the scale and duration of the chip commitments, the payment terms also included a substantial advance payment. At the time of signing, the advance was reported to be about 30% of the total, equivalent to USD 2.1 billion on a USD 7 billion contract.

U.S. Pressure Extends to AI Computing Services for China

U.S. government pressure reportedly played a role in the cancellation. According to two people familiar with the agreement, U.S. officials contacted Oracle directly to express concern about its transaction with Tencent, after which the contract was canceled. As U.S. export controls on advanced AI chips made domestic procurement more difficult, Chinese companies have turned to renting computing resources at overseas data centers. This arrangement allows them to access the performance needed for AI training by connecting remotely to overseas servers, without bringing the chips into China. Although overseas leasing transactions have been permitted under existing rules, the U.S. government appears to be pressuring cloud providers that offer advanced computing resources to Chinese companies to scale back those dealings.

Efforts to restrict Chinese customers' use of overseas computing resources have also appeared in Nvidia's business terms. According to AFP on the 3rd, Nvidia asked cloud customers using its chips to sell at least 51% of their total services to companies outside China. The condition limits the share of services allocated to Chinese customers to less than half and applies to new contracts signed going forward. The requirement was reportedly communicated verbally rather than written into contracts. In response to questions, Nvidia said only that data centers and cloud services built with U.S. chips provide a foundation for American AI companies to compete in global markets.

Table 1. Legal Proceedings and Concealment Methods in the Illicit Diversion of Advanced AI Servers to China

Illegal AI Server Exports to China Via Southeast Asia Uncovered

Alongside pressure on Chinese companies' use of overseas cloud services, the U.S. government is pursuing legal proceedings against those who secretly supplied AI servers to China through third countries. According to Reuters, former Super Micro Computer contractor Willy Sun pleaded guilty in federal court in Manhattan, New York, on the 8th of last month to four charges, including conspiracy to violate export controls, conspiracy to smuggle, and obstruction of justice. U.S. prosecutors had indicted three people in March this year—Sun, a Supermicro co-founder, and a sales representative at its Taiwan office—on allegations of involvement in illegally exporting advanced AI servers to China. A Southeast Asian company used in the transactions under investigation purchased USD 2.5 billion worth of servers in 2024–2025. Shipments to China identified by prosecutors totaled at least USD 510 million in just roughly three weeks from late April to mid-May last year.

The diversion scheme involved documents concealing the end users and even dummy servers intended to deceive on-site inspectors. According to an indictment released by the U.S. Department of Justice, the defendants allegedly falsified purchase documents to make it appear that a Southeast Asian company would use the servers itself, then delivered the U.S.-assembled servers to Chinese customers through Taiwan and Southeast Asia. During shipping, they used a logistics company to repackage the servers in unmarked boxes to conceal their contents. Investigators found that, after sending the actual servers to China, they placed thousands of nonfunctional dummy servers in warehouses in preparation for manufacturer inventory checks. Surveillance cameras also captured Sun and others using hair dryers to remove labels and serial-number stickers and transfer them to dummy servers and boxes. Some of the dummy servers were also used during on-site inspections by the U.S. Commerce Department.

Table 2. Evidence of Advanced Server Supplies to Southeast Asia and Diversion to China

Equipment Supplied to Southeast Asia Through U.S. Subsidiary

Evidence of advanced server procurement through Southeast Asia has also emerged in Chinese company Inspur's overseas trading network. According to The New York Times (NYT) on September 6, after being placed under U.S. export restrictions, Inspur supplied advanced equipment through its U.S. subsidiary Aivres, which was not on the restricted list. Shipping records show that Aivres exported at least USD 5.6 billion worth of equipment from the United States to Southeast Asia between April 2024 and February this year. More than USD 3 billion of that consisted of computers equipped with Nvidia Blackwell chips and related equipment. Recipients included Southeast Asian data centers providing computing services to Chinese companies such as Alibaba and ByteDance. Under this arrangement, the overseas subsidiary supplied equipment that enabled Chinese companies to access local facilities and use advanced chips.

Evidence has also emerged that servers supplied to Southeast Asia were re-exported to China. According to an analysis by the NYT and trade data company ImportGenius, Chinese company Maginfra imported more than USD 700 million worth of servers from Malaysia over six months through companies including those doing business with Inspur affiliates. More than 1,500 of those servers had unit prices in the range of high-performance AI servers, raising suspicions that advanced equipment was being diverted. Server transactions routed through Indonesia have also led to indictments. In August, Taiwanese prosecutors indicted nine people, including Nvidia and Supermicro employees, on allegations of sending 74 servers equipped with B300 chips to China through Japan and Indonesia. Prosecutors sought five-year prison terms for four key defendants, including an Nvidia manager.

Limits on Chinese Demand Complicate Investment Recovery for Southeast Asian Data Centers

Tighter U.S. controls on AI chips for China are expected to make it harder for Southeast Asian data centers to recoup their investments, as restrictions constrain the leasing demand from Chinese tech giants that has underpinned local expansion. According to industry sources cited by Caixin in August, 60–70% of the 1.5 gigawatts (GW) of data center projects approved in Thailand were linked to Chinese investment or had Chinese technology companies as their main customers. In Malaysia, ByteDance was also the anchor tenant at a 110-megawatt (MW) campus developed by Bridge Data Centres in Johor. Because Chinese companies have served as the customers filling these large facilities, restrictions on their new contracts would inevitably increase uncertainty over rental income from planned capacity expansions.

Finding customers to replace Chinese companies is also proving difficult. A Singapore-based cloud company interviewed by the FT on the 3rd said it was difficult to find enough non-Chinese customers outside the United States and warned that strict enforcement of the transaction restrictions would hurt its business. The burden of securing additional leasing demand will grow as data centers currently under construction begin operating. According to global real estate services firm Cushman & Wakefield, roughly half of the 4.8 GW of data center capacity under construction in the Asia-Pacific region in the first half of this year was concentrated in Southeast Asia. Capacity under construction in Johor, Malaysia, and Bangkok, Thailand, alone totaled 602 MW and 859 MW, respectively. If operators fail to secure enough customers even after the facilities are completed, rental income is expected to fall short of initial projections, extending the time needed to recover construction costs already incurred.