New Plastic Cables Developed for AI Infrastructure

As data-transfer speed and energy consumption become critical issues in the development of artificial intelligence and neural networks, Point2 Technology of the United States has raised $136 million to commercialize a new type of cable, as Ixbt.com reports.

Experts say that wired connections inside server racks at modern data centers are becoming a major obstacle to advancing computing systems. As AI systems scale and bandwidth rises to terabit speeds, interprocessor and server-to-server connections are becoming crucial.

Advantages of the New Technology

Point2’s e-Tube platform is based on using radio-frequency signals transmitted through flexible plastic waveguides instead of traditional copper wires or fiber-optic cables. This approach can significantly expand data-center capabilities.

According to its technical specifications, Point2’s solution provides 10 times greater transmission distance at the same cost compared with copper cables. At the same time, it reduces the weight of connections fivefold and cable volume twofold.

Compared with Fiber Optics

Compared with optical communication lines, radio-frequency plastic waveguides consume three times less energy and cost three times less. Most importantly, they reduce data-transfer latency by 1,000 times and eliminate infrastructure issues related to the reliability of laser emitters.

Building data centers for neural networks requires hundreds of billions of dollars, including spending on accelerators, electricity and cooling systems. Amid power shortages, every watt saved on interconnects can be directed straight to computation.

As Paul Williamson, Arm’s vice president of strategic investments, noted, AI infrastructure requires improvements across the entire technology stack, from computing units to memory and interconnects. The funding will accelerate the commercialization of Active RF Cable radio-frequency cables and e-Tube solutions designed for direct integration into chipsets and server racks.