How Tencent Scaled to Over 1 Million Servers and Cut Costs by 30%

Infrastructure Milestones

Tencent announced that in 2019 its total server count surpassed one million, making it the first Chinese company and one of only five worldwide to reach this scale. Its network bandwidth peak also broke 100 Tbps, the first in China to do so.

Evolution Toward the Cloud

At the 2019 Tencent Tech Developer Conference, the Vice President of Tencent Cloud and General Manager of the Cloud Architecture Platform, Xie Ming, disclosed the company’s infrastructure evolution, describing a transition to a "born‑as‑cloud" era driven by continuous upgrades to servers, networks, IDC, compute, storage, and databases.

Server Innovation and Cost Reduction

Tencent’s server journey moved from generic hardware to custom designs and now to self‑developed servers named "Starry Sea." These servers feature a trusted‑chain hardware tamper‑proof mechanism, advanced heat‑pipe cooling, and refined airflow control, achieving a 30% reduction in total cost of ownership (TCO), a 50% improvement in load efficiency, and a 35% boost in overall performance.

Modular Data Centers

The fourth‑generation T‑Block data center adopts a modular, building‑block approach, cutting construction time in half compared to traditional large‑scale data centers. Its natural cooling technology reduces PUE to below 1.2, and a 300 k‑server campus can save 250 million kWh of electricity annually.

Network Advancements

With connections to over 800 global carriers, Tencent has replaced traditional commercial routers with self‑developed, SDN/NFV‑based distributed routing clusters and SONiC‑based switches, lowering network equipment TCO by more than 20%.

Rapid Cloud Resource Provisioning

The VStation cloud computing operating system can provision thousands of virtual machines per minute, meeting the compute demands of medium‑sized enterprises. By leveraging DPDK, smart NICs, and other hardware‑software solutions, Tencent offloads network and storage I/O to hardware, achieving zero‑loss VM operation and zero interference between VMs.

Storage and Database Evolution

Since 2006, Tencent has operated the distributed storage system TFS for Q‑zone photo uploads, handling billions of daily images. Its Cloud Object Storage (COS) now engages all cluster nodes for throughput, achieving over 90% disk utilization and fully autonomous operation, dramatically reducing manual maintenance.

Database offerings have grown from MySQL to a portfolio of nearly 20 products, including CynosDB, which decouples compute and storage, supports instances up to hundreds of terabytes, and enables up to 15 compute nodes to share storage.

Extreme Elasticity and Serverless Computing

Tencent Cloud’s third‑generation general‑purpose serverless platform delivers near‑100% resource utilization, cutting user costs by more than 50%. Function cold‑start latency is reduced to less than 0.01%, and a lightweight virtualization technology can launch a VM in 35 ms—the fastest record in the industry—now used in "Mini‑Program Cloud Development".

Intelligent Services and AI‑Driven Optimization

Leveraging reinforcement learning, the CDBTune system provides online database performance tuning that can outperform dedicated DBAs.

Future Vision

Xie Ming emphasized that cloud computing will increasingly integrate intelligence to improve service quality and reduce costs, and that Tencent aims to open its battle‑tested infrastructure to developers worldwide, fostering the growth of the industrial internet.