Oracle reportedly lands $7B GPU deal for Tencent
Network World ·
Oracle has reportedly signed a five-year, $7 billion agreement to provide Chinese technology conglomerate Tencent with access to roughly 100,000 advanced AI accelerators deployed in Oracle data centers across Southeast Asia, helping Tencent to sidestep restrictions on advanced AI equipment. The deal, first reported by the Financial Times , would represent Tencent’s largest overseas cloud commitment with a U.S. provider. Neither Oracle nor Tencent has publicly confirmed the agreement or disclosed the specific AI processors involved. Under the reported terms, Tencent would pay about 30% of the contract value upfront—roughly $2.1 billion—with the remainder paid over the five-year term. That would give Oracle a substantial source of upfront funding for the expensive data center infrastructure required to build out the infrastructure. Tencent executives have said the company intends to use newly acquired computing capacity initially for training larger and more capable models, followed by inference workloads. Excess capacity could eventually be offered through Tencent Cloud. An important unanswered question is the identity of the processors. Oracle cloud infrastructure offers Nvidia and AMD AI accelerators to choose from. It is not a given that Oracle will go with Nvidia. But Tencent isn’t going to wait for Oracle, it is making some big moves in AI now. FT says its newly released Hunyuan family of AI models and generative AI and agentic-AI capabilities are showing vast improvement over prior versions, Also, Tencent is rolling out Xiaowei, an embedded agent in WeChat with stronger models. WeChat is a “ superapp ” with 1.4 billion use and has millions of mini programs, such as ordering food or booking services, based on natural language commands. The Tencent agreement also highlights an emerging problem for U.S. semiconductor policy : controlling access to AI accelerators is becoming more difficult when customers can rent computing capacity outside the country. Current U.S. export restrictions primarily target the shipment and transfer of advanced semiconductor technology, especially as relates to China. Remote access to processors physically located in third countries has created a potential gap between hardware export controls and cloud computing access. The Commerce Department has initiatives and proposed legislation to close that loophole. If such rules are adopted, the Oracle/Tencent arrangement could become an important test case for how the U.S. regulates AI compute rather than simply AI chips.
Oracle has reportedly signed a five-year, $7 billion agreement to provide Chinese technology conglomerate Tencent with access to roughly 100,000 advanced AI accelerators deployed in Oracle data centers across Southeast Asia, helping Tencent to sidestep restrictions on advanced AI equipment. The deal, first reported by the Financial Times , would represent Tencent’s largest overseas cloud commitment with a U.S. provider. Neither Oracle nor Tencent has publicly confirmed the agreement or disclosed the specific AI processors involved. Under the reported terms, Tencent would pay about 30% of the contract value upfront—roughly $2.1 billion—with the remainder paid over the five-year term. That would give Oracle a substantial source of upfront funding for the expensive data center infrastructure required to build out the infrastructure. Tencent executives have said the company intends to use newly acquired computing capacity initially for training larger and more capable models, followed by inference workloads. Excess capacity could eventually be offered through Tencent Cloud. An important unanswered question is the identity of the processors. Oracle cloud infrastructure offers Nvidia and AMD AI accelerators to choose from. It is not a given that Oracle will go with Nvidia. But Tencent isn’t going to wait for Oracle, it is making some big moves in AI now. FT says its newly released Hunyuan family of AI models and generative AI and agentic-AI capabilities are showing vast improvement over prior versions, Also, Tencent is rolling out Xiaowei, an embedded agent in WeChat with stronger models. WeChat is a “ superapp ” with 1.4 billion use and has millions of mini programs, such as ordering food or booking services, based on natural language commands. The Tencent agreement also highlights an emerging problem for U.S. semiconductor policy : controlling access to AI accelerators is becoming more difficult when customers can rent computing capacity outside the country. Current U.S. export restrictions primarily target the shipment and transfer of advanced semiconductor technology, especially as relates to China. Remote access to processors physically located in third countries has created a potential gap between hardware export controls and cloud computing access. The Commerce Department has initiatives and proposed legislation to close that loophole. If such rules are adopted, the Oracle/Tencent arrangement could become an important test case for how the U.S. regulates AI compute rather than simply AI chips.