China has launched the Future Network Test Facility (FNTF), a distributed network of computing centers spanning more than 2,000 km and operating almost as efficiently as a single giant data center. The system is aimed at training large-scale AI models, telemedicine, and the industrial internet.
China has launched the Future Network Test Facility (FNTF), a distributed network of computing centers spanning more than 2,000 km and operating almost as efficiently as a single giant data center. The system is aimed at training large-scale AI models, telemedicine, and the industrial internet.
According to Interesting Engineering, FNTF connects data centers in 40 cities using a high-speed optical network with a line length of over 55,000 km. This allows you to synchronize distributed resources as if they were operating within a single cluster, and achieve up to 98% efficiency from a single data center.
The key feature is what’s called a «deterministic» network approach. Engineers say that for training large models with hundreds of billions of parameters, a single iteration on this network takes about 16 seconds. Without such infrastructure, each iteration could take more than 20 seconds, stretching training out to months.
The project presents itself as an answer to the headache of modern AI — the lack of capacity and the difficulty of scaling. Instead of building a single megadata center, FNTF is trying to «glue» existing sites into a single system with predictable delays and guaranteed bandwidth.
The network is capable of simultaneously supporting hundreds of heterogeneous networks and thousands of parallel service tests. This is important not only for AI, but also for scenarios like remote surgery or real-time industrial facility management, where extra milliseconds can be critical.
The FNTF also fits into the government’s «East Data, West Computing» strategy, in which China is locating data centers in regions with cheaper energy and trying to assemble them into a single national computing platform. The new project is intended to become one of the technology «hubs» of this infrastructure.
A special emphasis is placed on integration with promising technologies such as photonic and quantum chips. In the long term, this should reduce power consumption and increase computing density, which is critical for large AI clusters.
At the same time, analysts draw attention to the risks: maintaining stable operation of such a stretched system is difficult, and the energy consumption of dozens of connected data centers will be enormous. In addition, scaling such an infrastructure inevitably raises issues of cybersecurity and control of access to computing resources.
For the global AI market, FNTF is a signal that the race is already moving beyond individual companies and to the level of state network projects. For Ukraine and the EU, this is another reminder: without their own strategies for developing computing infrastructure, competition with such players will become increasingly difficult.
Previously, dev.ua wrote about how US President Donald Trump announced that the United States would allow Nvidia to sell its H200 processors, the second-most powerful AI chips, in China. At the same time, the government would collect a 25% commission on these sales.
