Understanding the Bandwidth and Server Scale of Douyin (TikTok) Data Centers
This article explains how Douyin (TikTok) and its parent companies operate massive data centers with tens of thousands of servers and multi‑terabit outbound bandwidth, using dual‑link designs, CDN acceleration, and load‑balancing to support hundreds of millions of concurrent users.
Recently a curious question arose: how large is Douyin's (TikTok's) server bandwidth, and why can it serve so many users simultaneously? This article provides a technical overview.
Douyin, Baidu, and Alibaba Cloud all run self‑built data centers with T‑level (1 TB = 1024 GB/s) total outbound bandwidth. Their server fleets exceed 200,000 machines, with Alibaba alone surpassing one million.
ByteDance’s data center total bandwidth is estimated around 10 TB and expected to approach 15 TB soon.
In practice, a total outbound capacity of 1 TB may translate to roughly 100 GB of actual room‑level outbound bandwidth, achieved through dual‑ or multi‑link designs that aggregate to T‑level capacity.
To support hundreds of millions of concurrent users, TB‑level bandwidth, CDN acceleration, multi‑node deployment, and load‑balancing are all essential.
How many servers does ByteDance operate?
In early 2017, ByteDance rented about 20‑30 k servers. In 2018 it built its own data center in Hebei Huailai, with Phase 1 housing 50 k servers and Phase 2 90 k. By the end of 2018 the total reached 170 k, and a 2020 recruitment announcement indicated 420 k servers.
These servers primarily serve Chinese‑region products such as Douyin, Xigua Video, Toutiao, and Feishu. In the United States, TikTok operates independently, renting nearly 100 k servers in 2020.
Business Insider reported that in early 2020 ByteDance rented a data center in northern Virginia with a 53 MW power capacity, capable of housing hundreds of thousands of servers across tens of thousands of square feet.
TikTok is also investing in data centers in India and Singapore.
What is the outbound bandwidth of ByteDance’s large data centers?
Outbound bandwidth is essentially the download bandwidth experienced by end users—the total data delivery speed from servers to devices.
Small IDC companies typically have total outbound bandwidth of about 5 G, while enterprises with >30 G are considered sizable.
Many enterprises now prefer cloud hosts such as Alibaba ECS, Tencent Cloud, Baidu Cloud, or AWS instead of building their own facilities.
A typical corporate website might use 20 M bandwidth, 4 G RAM, and 100 G storage, costing roughly 4,000–5,000 CNY per year.
Bandwidth is the most expensive resource; increasing it incurs higher costs than adding memory or storage.
China Mobile’s data center in Shijiazhuang (174 mu, 130 k m²) plans 10 buildings, offering about 30 k racks and 15 T of bandwidth. Each rack can host up to 16 U‑density servers, yielding an estimated capacity of 30 k servers.
Roughly estimating ByteDance’s self‑built data center at 170 k servers, the total outbound bandwidth likely falls between 7 TB and 10 TB.
Using dual‑outbound and multi‑link designs, an actual outbound of 800 G–1 TB can achieve an aggregated total of around 10 TB.
For context, Shanghai’s total outbound bandwidth was only 1.14 TB in 2009; a decade later, major enterprises like ByteDance exceed 1 TB.
CDN acceleration ensures smooth video playback by distributing content to edge nodes close to users, reducing latency and network congestion.
CDN works by publishing server content to edge nodes, allowing users to retrieve data locally, and by compressing static pages for rapid delivery (often within 2 seconds).
Dynamic video streams use intelligent routing, protocol optimization, long‑connection handling, and compression to minimize redundancy.
In 2015, Tencent’s CDN handled 5 hundred million daily active users across music, messaging, and other services, delivering up to 10 TB of bandwidth and processing trillions of requests daily.
Given Douyin’s 600 million daily active users and additional services totaling roughly 800 million, ByteDance’s overall bandwidth requirement is plausibly around 10 TB.
This massive bandwidth and technical infrastructure enable the seamless video experience users enjoy today.
Source: Internet
Architecture Digest
Focusing on Java backend development, covering application architecture from top-tier internet companies (high availability, high performance, high stability), big data, machine learning, Java architecture, and other popular fields.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.