China has launched a large-scale centralized network for artificial intelligence that connects data centers across the country over a distance of more than 1,900 kilometers. The new system is named the 'Future Network Test Facility' (FNTF) and was presented on December 3 after 10 years of research.
According to Interesting Engineering, the Chinese development represents 'the world's largest distributed supercomputer for AI.' Journalists call the project 'the optical highway' — a unifying infrastructure for AI data centers.
Impressive scales
The system covers 40 cities and includes over 55,000 kilometers of optical transmission lines — enough to wrap around the Earth at the equator one and a half times. The performance of the new infrastructure is impressive: data transmission that would take about two years on the regular internet is completed in just an hour and a half through this 'AI highway.'
FNTF operates round the clock, supporting 128 heterogeneous networks and 4,096 parallel service trials. This makes it one of the most extensive testbeds ever deployed in the world. The project team has created 206 international and national standards, received 221 patents for inventions, and built what they call the world's first distributed large-scale network operating system.
Technological foundation
At the core of FNTF are three data transmission protocols: DWDM, ROADM, and OTN. Each of these technologies plays its role in ensuring high system performance.
Dense Wavelength Division Multiplexing (DWDM) is a fiber-optic technology that uses the principle of the color wheel to increase data transmission capacity. It transmits individual data streams over a single fiber in multiple channels, placing each channel at a different wavelength of light corresponding to a specific color.
Reconfigurable Optical Add-Drop Multiplexer (ROADM) acts as a switch that can remotely add new channels or 'colors' to the fiber, remove a signal of a specific wavelength, or pass it through a point unchanged.
Optical Transport Network (OTN) unites multiple services, including Ethernet and 5G, into standardized frames with powerful error correction, state checks, and other monitoring.
American response
The USA is not staying out of the technology race but is choosing a different approach to infrastructure development for AI. Instead of a single monolithic project, Americans are developing a portfolio of testbeds, spectral policy, and general infrastructure for artificial intelligence.
Within this strategy, three key initiatives are implemented. NSF FABRIC is a research network of the National Science Foundation designed to prototype new architectures at scale. DOE ESnet is a network of energy sciences from the U.S. Department of Energy, connecting national laboratories and supercomputing centers. NSF PAWR includes urban platforms — real outdoor wireless testbeds for experiments in Harlem (Manhattan), Salt Lake City (Utah), and the Research Triangle in North Carolina.
A look into the future
The Chinese FNTF project is an open testbed for experiments with 'next-generation internet' technologies on a national scale. It is a research and validation platform that will stimulate a particular type of progress in a rapidly changing world.
The system can be compared to CERN or the Large Hadron Collider — unique research objects that set development trends. China, in a certain sense, has created a significant technological advantage for itself.
The development of such infrastructure projects makes one think about how the next generations of AI will be delivered, how we will personally interact with technology, and what our future interfaces will look like. All of this becomes an extremely instructive experience for understanding the direction of technological development.
