Tencent, the Chinese company behind WeChat and some of the world's biggest video games, has agreed to lease around 100,000 advanced AI chips from Oracle, the Financial Times reported on 30 September. The five-year deal is worth about $7 billion, according to the report, with roughly 30% of the value paid upfront and the chips housed in several Oracle data centres in Southeast Asia.
The chips will not go to China. Tencent will reach them remotely over the network and pay for time on the hardware rather than own it, which is the detail that makes the arrangement possible under current US export rules. Reports describe it as the largest overseas compute lease Tencent has signed.
Neither company has confirmed the deal. Oracle and Tencent did not respond to Reuters' requests for comment, Reuters said it could not verify the report independently, and the FT did not name the chip model or the countries where the data centres sit.
What The Reported Deal Covers
The numbers point to a large, long-term reservation rather than a short burst of capacity. A $7 billion contract spread over five years works out at about $1.4 billion a year, and an upfront payment of some $2.1 billion suggests Tencent was willing to pay early to secure a place in the queue for scarce AI hardware.
A cluster of 100,000 advanced accelerators would rank among the larger AI computing sites in Asia, the kind of capacity used to train large language models and to serve them to millions of users. Tencent has been pushing its Hunyuan family of models and building AI features into WeChat, its cloud business and its games, and every one of those efforts needs more computing power than the company can easily get at home.
Oracle runs a cloud region in Singapore and said in 2024 that it would invest more than $6.5 billion in a public cloud region in Malaysia, both natural candidates for the work, although neither company has said where the capacity will come from.
Why Tencent Is Renting Rather Than Buying
Tencent's route to advanced chips at home remains slow and uncertain, which explains the move overseas. The US banned sales of Nvidia's most advanced processors to China in stages from 2022, and although Washington agreed last December to let Nvidia sell its older H200 chip to approved Chinese buyers, with a quarter of the revenue going to the US government, very little has arrived.
The FT reported in August that ByteDance and Tencent had each received roughly 10,000 H200 units, a small share of the licensed volumes. Beijing now controls the pace: China's National Development and Reform Commission reviews each purchase and asks companies to justify why they cannot use domestic alternatives from suppliers such as Huawei, as the government pushes its tech sector towards Chinese silicon.
That leaves Chinese AI developers caught between two governments. Washington limits what they can buy, Beijing limits what they can bring home, and renting capacity abroad sits outside both sets of restrictions for now.
A Spending Surge At Home
The lease also fits a company already spending heavily on AI. Tencent reported second-quarter capital expenditure of 52.8 billion yuan, about $7.4 billion, up 176% on a year earlier, as revenue rose 11% to 204.8 billion yuan. Chief executive Pony Ma said the company was making "substantial progress towards building a new, AI-empowered Tencent," and analysts noted that spending at that pace leaves far less free cash flow than investors are used to from the group.
Renting rather than buying shifts some of that burden. A lease turns a large hardware purchase into a contract paid over time, and it lets Tencent use chips it could not otherwise import.
What Oracle Gets From The Deal
For Oracle, the deal arrives at a useful moment. The company reported first-quarter fiscal 2027 results in September with revenue up 30% to $19.3 billion and cloud infrastructure revenue up 121% to $7.4 billion, and its order backlog reached $664 billion. But roughly half of that backlog comes from OpenAI, and investors have worried about how much Oracle depends on a single customer while it borrows heavily to build data centres.
Oracle spent $28.5 billion on capital projects in the quarter, posted negative free cash flow of $5.4 billion and carries about $125 billion in debt. Its shares had fallen about 30% this year before the report, and they rose more than 2.5% in overnight trading after the news broke, as traders read the contract as proof that Oracle can win large AI customers beyond OpenAI.
The reported upfront payment also matches how Oracle now says it structures new AI deals. Management told analysts in September that most new contracts come with prepayments or with customers supplying their own hardware, so that Oracle does not have to fund all of the build-out itself.
A Gap Washington Has Tried To Close
The deal relies on a gap in US export controls that lawmakers have tried to close for more than a year. Current rules govern where advanced chips are shipped, but they generally do not stop a data centre in a third country from selling remote access to those chips to a customer in China.
The Biden administration's AI diffusion rule would have tightened controls on where advanced chips could go and who could use them, but the Trump administration withdrew it in May 2025 before it took effect. In January, the US House of Representatives passed the Remote Access Security Act by 369 votes to 22, a bill that would extend export controls to cloud access. "This bill brings our laws into the digital age and makes it clear that cloud compute is subject to US export control law, just like physical chips," said John Moolenaar, who chairs the House Select Committee on China.
A Senate version, introduced by senators Dave McCormick and Ron Wyden, has not yet passed. If it becomes law, the Commerce Department would still need to write the rules that decide which kinds of remote access need a licence, and leases like Tencent's could then come under review.
Tencent Is Not The First
Chinese tech companies have used this route for some time. The Information reported in January 2025 that ByteDance planned to spend $7 billion that year on access to Nvidia chips outside China, and last November the FT reported that Alibaba and ByteDance were training their top models in Southeast Asian data centres in Singapore and Malaysia. Chinese rules on moving personal data abroad mean those companies still fine-tune their models at home on Chinese user data.
What sets the Tencent deal apart is its size and its counterparty. A multi-year, $7 billion contract with a major American cloud provider brings the practice into the open in a way that smaller rentals through regional operators did not.
What It Means For Companies Buying AI Compute
The deal carries lessons for businesses well outside China. Southeast Asia has become one of the busiest data-centre regions in the world, with Johor in Malaysia and the area around Singapore drawing large AI builds, and a 100,000-chip reservation by one tenant tightens the supply of GPU capacity, power and floor space for everyone else in the region. Companies that plan to run AI training or inference in Asia may find capacity booked further ahead and prices firmer, a strain that runs from the chips down to the networks, as we saw in how AI's uplink surge reaches the fibre behind 5G towers.
For cloud providers and their enterprise customers, the regulatory risk is now part of the contract. If Washington extends export controls to remote access, providers may have to check who sits behind every large compute order, and multi-year agreements could face licence requirements or forced changes mid-term. Buyers signing long AI capacity deals will want clear terms on what happens if the rules shift.
Compute Has Become The Thing That Crosses Borders
The reported Oracle lease shows how the contest over AI chips has moved from shipping containers to network connections. Export controls were built around physical goods, and the most valuable AI resource now travels as rented time on hardware that never leaves a data centre in Singapore or Malaysia.
For Tencent, the lease buys capacity it cannot get at home on its own timetable. For Oracle, it adds a large customer and a second headline name beyond OpenAI. The next signals will come from the companies themselves, if they confirm the terms, and from the US Senate, whose vote on the Remote Access Security Act will decide whether deals like this remain an ordinary cloud contract or become a licensed export.
