Tencent has launched the Full-Stack Intelligent AI Suite, a set of software tools for building and running AI agents inside a company’s own private cloud. The suite runs on Tencent Cloud Enterprise (TCE), Tencent’s private cloud platform, and was unveiled on 6 November at the company’s Global Digital Ecosystem Summit in Shenzhen, China. Tencent also launched Lite Deployment, a smaller version of TCE that can start on three servers.
A private cloud gives a company cloud-style computing on servers it controls, rather than renting capacity from a public provider. TCE is sold as software for customers’ own hardware, and Tencent says it scales from three servers to tens of thousands of machines. It can run connected to the internet or fully air-gapped, meaning cut off from all outside networks.
Tencent Cloud vice-president Kaibo Sha said that once AI agents handle core business data and critical systems, the main question is no longer whether the AI model is strong enough. Companies need to know where their data is processed and who runs the system, he said, as well as what the agents run on.
What The Suite Includes
Customers can adopt the suite one layer at a time, starting at any level. The base layer combines computing, storage and networking. TokenHub sits on top of it. This private platform turns a company’s computing power into access to AI models, measured in tokens. Tokens are the small chunks of text that models process, and providers usually charge by them. TokenHub keeps different teams separate and can track usage down to individual users.
The next layer runs the agents in a sandbox, a sealed-off space with controls on outbound traffic. Each agent also gets its own credentials, so it can be identified. The top layer holds the agent applications themselves.
Tencent says every agent action can be audited. It adds that data stays within the customer’s own boundary, along with the model weights that make an AI model work and the requests and answers that pass through the model in use.
Tencent says TokenHub handles more than 250,000 tokens per minute (TPM) on a single AI chip. It also says scheduling model training and model use together can raise the use of mixed computing hardware by up to 60%. According to the company, the platform runs on AI chips from several vendors and on x86 and ARM, the two main types of server processor. Tencent says it has been tested on 10,000 graphics processors.
Starting On Three Servers
Lite Deployment is aimed at organisations that want to start small. Tencent says a full private cloud needs eight servers, while an AI-only setup with private model services can start on three. From there, customers can grow from one availability zone to two running side by side in the same city, and later add backup across regions. An availability zone is a separate block of data centre capacity.
The option runs on hardware customers already own and has no minimum spend, according to Tencent. One customer started its AI work on three servers. Tencent describes the firm as one of the largest futures brokerages by trading volume and client assets, and says it runs its core trading systems on TCE across chips of different designs.
Other named customers include the Xuhui district government in Shanghai, China, which runs a cloud on chips from different makers. Shenzhen Second People’s Hospital uses TCE as the base for AI agents in clinical, service and management work. A large organisation in Hong Kong is moving off VMware’s server software. In Thailand, a digital infrastructure provider dropped VMware and opened two data centres on TCE in eight weeks.
The Wider Push For Private AI
Analyst firm Forrester looked at TCE in a September 2025 blog post on Tencent’s sovereign cloud plans, meaning cloud computing kept under a customer’s own legal and operational control. The post said TCE shares its source code with Tencent’s public cloud. It also said China Construction Bank was building a cloud platform supporting more than 80,000 physical servers and moving its core banking systems onto TCE.
Tencent also released a guide to sovereign AI clouds with research firm Gartner. The guide cites a Gartner forecast that by 2028, strict data sovereignty and privacy rules will push more than a quarter of companies in regulated industries to move their main generative AI work back onto their own premises. Gartner also expects dependence on proprietary AI models to leave more than half of generative AI users locked into one public cloud provider through 2029.
Other companies are pitching similar control over AI. Context Labs, which grew out of research at the Massachusetts Institute of Technology, recently launched software for running AI on company-owned data. Tencent itself was reported in September to be leasing about 100,000 AI chips from US software group Oracle. Neither company has confirmed the deal.
Tencent is pitching TCE to financial firms, hospitals and governments as a way to run AI agents while keeping data, models and operations under their own control. The company says the three-server option gives them a smaller way in, on hardware they already own, with room to grow into a full private cloud.
