Artificial Intelligence (AI) data centers face growing challenges as demand for land and energy surges. A novel solution gaining traction is hollow core fiber (HCF), which promises both logistical and technical advantages over traditional single-mode optical fiber (SMF). HCF, rooted in the photonic-crystal fiber technology of the late 1990s, features a hollow core. This offers a striking contrast to conventional optical fibers with their solid glass cores. While the latter represents a significant improvement from copper, it is still not without signal resistance and attenuation.
In contrast, HCF’s unique properties allow for faster data transmission with considerably less signal loss. Industries, especially those like high-frequency trading which rely on microsecond transactions, find the reduced latency offered by HCF compelling. In such fields, even a few milliseconds saved can be crucial. Bernard Lee, the senior director at SENKO Advanced Components, points out that this technology could lead to faster training times in AI models, providing an edge over competitors.
Beyond speed, the adoption of HCF can address the pressing land and power constraints in the AI sector. In the U.S., there are significant requests for new energy capacity, set to expand to 85 gigawatts by 2030. However, the current supply falls short. Simultaneously, finding available land for new data centers proves challenging.
Given AI’s vast energy needs and the concentrated data center hubs facing land shortages, HCF holds promise. Its ultra-low latency could efficiently connect distributed data centers in less congested areas, offering strategic advantages. This allows companies to choose more economically viable locations and reduces the entry barriers for new data center setups.
Nonetheless, the high costs associated with HCF remain a barrier. Precision manufacturing, specialized handling, and unique physical properties add to its expense. Bernard Lee acknowledged that while the technology is in its early deployment stages, standardization and ecosystem evolution could bring costs down, making it feasible.
Could advancements in AI chip design threaten the utility of HCF? Lee believes both technologies complement each other. He likens HCF to the aerodynamics of a car while AI silicon acts as the engine. Together, they could facilitate a more sustainable and responsible AI deployment.
Realistically, the widespread adoption of HCF hinges on industry collaboration. Cable manufacturers, fiber providers, and test equipment vendors must come together to establish standards.
