Under the arrangement, Oracle would have placed up to 100,000 AI chips in its Southeast Asian data centers while Tencent rented the computing power remotely. That structure is the point. US export controls currently stop Chinese companies from buying such chips directly, so renting capacity abroad would have delivered much of the same processing power by another route.
US officials reportedly raised concerns with Oracle directly, though informally. There was no enforcement order or legal ruling. What Washington appears to be signaling is that overseas cloud infrastructure serving Chinese customers may fall under export-control rules too.
Tencent Was Pulled in Two Directions
Tencent reportedly faced pressure from both capitals at once. Officials in Guangdong urged the company to use domestically produced semiconductors, in line with Beijing’s push to cut reliance on foreign chips. Washington, meanwhile, wants to limit Chinese access to advanced AI computing. Those competing pressures reportedly contributed to the deal falling apart.
A separate layer of constraint may be coming from the chip supplier itself. Nvidia has reportedly told cloud customers to limit advanced computing capacity serving Chinese users. As reported:
- Under the new agreements, at least 51 percent of services must go to non-Chinese customers
- The requirement applies to future contracts, not existing ones
- Nvidia has not publicly confirmed it, and how compliance would be measured is unclear
ByteDance Faces the Same Question
The Oracle and Tencent situation is unlikely to stay isolated. ByteDance reportedly maintains active chip-leasing arrangements with US cloud providers, and Oracle is reportedly a significant supplier to the company across Asian data centers. Those arrangements now sit under the same scrutiny that appears to have ended the Tencent deal.
Hashlytics Take
The most telling detail is how the pressure was applied. A conversation, not a rule, appears to have been enough to kill a $7 billion contract. That leaves cloud providers pricing regulatory risk with no written standard to point to, which is a harder position than complying with a clear restriction. It also shifts the question from where the chips sit to who is renting them, a test that most hosting contracts were never built to answer. If the reported Nvidia cap is real, the compliance burden moves further down the chain, from the company selling the chips to the clouds hosting them. All of this still hangs on a single report with no confirmation from Oracle, Tencent, Nvidia or the White House, so the direction is clearer than the details. The same unease about who controls cloud infrastructure shows up elsewhere too, as when a Swiss data chief flagged US cloud risk to authorities.
Follow Hashlytics on Bluesky, Facebook, LinkedIn , Telegram and X to Get Instant Updates
