China Internet Speed in 2026: What Actually Limits It
China internet speed is fast domestically and slow to global services. Why that gap exists, and what actually helps a remote professional in China.
A lot of people in mainland China run a speed test, get a strong number, and assume the connection is fine. Then the workday starts. Zoom stutters. Teams lags. Google Drive uploads crawl. GitHub pulls time out. The line looks fast on paper, and international work still feels unstable.
That gap is the actual story behind China internet speed. Domestic performance and international performance are not the same product. China has modern fibre, dense urban infrastructure, and very fast local routes. The moment traffic has to leave the mainland and reach global services, the experience changes. For anyone working with overseas clients, distributed teams, or cloud tools hosted outside China, the useful question is not "how fast is the plan?" It is "how fast is the path to the services that matter?"
Table of contents
- The paradox of China's gigabit internet
- Domestic speed versus international speed
- Why international internet from China is slow
- How carrier choice changes everything
- How to actually test your China connection
- What actually helps
The paradox of China's gigabit internet
Someone in Shanghai can pay for a premium home line, connect over fibre, and still watch a client call freeze the moment screen sharing starts. That is not unusual. It is the normal contradiction of the Chinese internet.
China's network is not slow in the simple sense. Urban infrastructure is strong. Many cities offer gigabit service. Median fixed-broadband download speeds rank competitively against developed markets, and large parts of the country sit comfortably above 200 Mbps on local tests.
But advertised speed is a vanity metric for international professionals. A local speed test can look excellent while everyday work remains frustrating. The bottleneck is rarely the apartment router, the office Wi-Fi, or the fibre package. It is the route out of the country.
Practical rule: if a connection works well for local Chinese platforms but degrades on Google, Zoom, Slack, GitHub, or overseas cloud storage, the issue is almost never local access speed.
For domestic apps and Chinese-hosted services, the network feels sharp and responsive. For foreign platforms, the same connection feels brittle. Browser tabs hang. File sync pauses. Calls reconnect several times in one meeting. Most remote workers do not need a higher headline number. They need consistency.
Domestic speed versus international speed
On domestic metrics, China performs well and the picture deserves to be taken seriously before evaluating cross-border problems. Beijing and Shanghai both run solidly above 200 Mbps on typical fixed-broadband measurements. Gigabit service is available in major cities. Domestic apps load quickly because they are physically close, well-peered inside the country, and not crossing any filtered boundary.
Once a workflow depends on foreign endpoints, the same line behaves differently. The polished story about fast national internet meets the working reality of multinational teams.
International access from mainland China commonly degrades by half or more compared to the same line's domestic performance, and on bad evenings on the wrong carrier the gap is much larger. A line that runs at 300 Mbps to a domestic Speedtest server can deliver under 50 Mbps to a foreign endpoint, before any VPN is in the picture.
A network engineer looks at any Chinese speed-test number and asks the second question immediately: where is the server? If the benchmark is inside mainland China, the result describes local reach. It does not describe the outbound path.
Why international internet from China is slow
Three forces compound to produce the gap.
Limited international gateways. Cross-border traffic from mainland China is funnelled through a small number of Internet Exchange Points run by the major state-owned carriers. There is no equivalent to the dense, openly peered fabric that connects, say, Frankfurt or Singapore to the rest of the global internet. The pipe is narrower than the country's domestic capacity, by design.
Inspection overhead. Every cross-border connection passes through deep packet inspection. The system identifies traffic that matches restricted patterns and slows or drops it. Even traffic that is not restricted pays a small inspection tax. At peak times, that overhead becomes visible, and it has grown heavier through the 2026 China VPN crackdown.
Peering quality varies sharply by carrier. China Telecom and China Mobile maintain comparatively strong peering with East Asian and global routes. China Unicom historically peers poorly with Japan, especially in the evening, which is exactly when many professionals need the link most. The same line, the same VPN, the same time of day can feel completely different on a different carrier in the same building.
This is not a hardware problem. The infrastructure exists. The architecture is the constraint. Adding more bandwidth on the local side does very little when the choke point is fifteen hops away in the cross-border layer.
How carrier choice changes everything
For a single user this is the variable with the largest practical impact, and it is invisible to most VPN reviews.
- China Telecom is generally the strongest choice for international connectivity. East Asian routes peer well, and most VPNs that work in China work better on Telecom than on the alternatives.
- China Mobile is a reasonable second. Its mobile network in particular is often the best place to test whether a problem is the home line or the carrier.
- China Unicom is the carrier where most of the avoidable pain happens. Strong domestically. Weak peering with Japan, especially during peak evening hours. A user on a Unicom home line who consistently sees slow international speeds is often not looking at a VPN problem; they are looking at a carrier problem.
| Carrier | Domestic performance | International peering | Evening peak behaviour | Best for |
|---|---|---|---|---|
| China Telecom | Strong | Strong East Asian and global routes, peers well with CN2 GIA | Holds up better than the alternatives | Cross-border work, the default recommendation |
| China Mobile | Strong | Reasonable, solid mobile peering | Workable, good as a diagnostic line | A solid second, and the easiest way to test home line versus carrier |
| China Unicom | Strong | Weak peering with Japan exits | Degrades noticeably in the evening | Domestic use, the weakest pick for international work |
If you have any flexibility on the carrier for an apartment line or an office, this is a more important choice than it sounds. It also explains why two people running the same app from the same building can report opposite results, a pattern explored further in our ranked guide to the best VPN for China.
How to actually test your China connection
Most speed tests are misleading because they default to a domestic server. Two tests, run separately, give a useful picture.
Step one. Test against a local Chinese server. This number describes the line into the building. If it is low, the bottleneck is local: a poor router, a shared building line, a 2.4 GHz Wi-Fi setup in a dense apartment block, or a cheap Ethernet cable. Fix this first. There is no point evaluating the international route when the local one is broken.
Step two. Test against a foreign endpoint. Pick a server outside China, ideally in Japan, Singapore, Korea, or the United States. The gap between this number and the local one is the cross-border degradation on this line at this time on this carrier. That is the number that matters for international work.
A few records to keep alongside the numbers: city, carrier, time of day, what application was being used. A pattern emerges within a week of measurement.
One useful note: tracert and traceroute do not go through most VPNs. They use ICMP packets that take the raw internet path, so the trace shows the route as if no VPN were connected. To verify a tunnel is actually carrying your traffic, visit a "what is my IP" page in a browser and confirm the IP is not in mainland China.
What actually helps
The honest list of things that move the needle for international internet from mainland China is short.
- The right carrier. China Telecom or China Mobile beats China Unicom for cross-border work in most of the country, most of the time.
- A wired connection. Sustained uploads and video calls reveal Wi-Fi problems that look like VPN problems but are not.
- A real route out of China. A private international link with reserved capacity does not collapse at 8 p.m. the way a shared commodity link does. This is the part that distinguishes a work-grade connectivity service from a consumer VPN, and it is the single biggest reason most consumer VPNs cannot hold a workday open in China.
- Operations on the provider side. A network that is actively maintained will hold up. A network that was set up once and left alone will degrade over weeks.
- A second path for important work. A mobile hotspot on a different carrier, or a colleague's connection, gives you somewhere to fail over to during a bad evening. This is normal infrastructure hygiene, not an indictment of any provider.
What does not help is upgrading the home line in the hope that more local Mbps will fix the international experience. If the workload is domestic, a faster plan is great. If the workload is overseas, a faster plan is mostly noise.
The Chinese internet is fast inside the country. It is the boundary that is slow. Designing around that fact, instead of fighting it with a bigger router, is the difference between a workday that flows and one that does not.
Throughwire was built for exactly the boundary problem described here: a private route out of mainland China on CN2 GIA premium bandwidth, actively operated, so the cross-border layer does not collapse at 8 p.m. For professionals and teams whose work depends on overseas tools, Throughwire is worth evaluating on the only number that matters, peak-hour performance to a foreign endpoint, rather than the headline speed on a domestic test.