"Behind-the-Scenes Heroes": Characteristics of Wiring
The popularity of ChatGPT has ushered us into the era of artificial intelligence, bringing significant impacts and changes across many fields. Some changes are visible to the naked eye, while others, like wiring, are silent and often overlooked 'behind-the-scenes heroes'.
In the development of artificial intelligence, computing centers or data centers play a crucial role as primary infrastructure, with wiring acting as the 'blood vessels' that connect these infrastructures. Wu Jian, Technical Director of CommScope Enterprise Network in Greater China, states, "Data centers can be simply understood as our brain, and the intelligence of the brain comes from the connections between numerous neurons. Connectivity is a fundamental issue."
As artificial intelligence technology continues to evolve, the volume of data it needs to process is also growing exponentially. Recent estimates suggest that general artificial intelligence may require as many as 10 trillion parameters to achieve the desired complexity and performance level. Thus, data is the cornerstone of artificial intelligence development. Handling such vast volumes of data requires sufficient computing power and fast connections, as there is a synergistic interaction between computing power and network connectivity capabilities, akin to the brain (computing power) and legs (network connection) of society, both of which need to be improved synchronously for efficiency. According to a report by Dell’Oro in June 2022, 400G, 800G, and 1.6T will be the future trends of Ethernet development. This indicates that connection technologies before 400G, such as 100G and below Ethernet connections or data center internal connections, are rapidly evolving to 400G, 800G, and 1.6T rates.
Wu Jian comments, "The pace of iteration indeed exceeds our imagination. Especially after 2023, the proportion of network speeds above 400G has increased rapidly. In the next 2-3 years, the proportion above 800G will exceed half, which is also beyond our imagination. We believe that this rapid change is driven by AI technology."
The development of artificial intelligence technology also promotes the rapid deployment and development of data centers. Due to their high density, high bandwidth, low latency, and lossless characteristics, the overall wiring deployment strategy of data centers is significantly different from that of ordinary computer rooms or buildings. Wu Jian points out several key characteristics: firstly, there is a significant increase in the use of optical fibers. Data center connections are high-density connections at a centralized location, with increasing speed demands leading to more fiber usage.
Thus, as data center operations continue to grow, the proportion of optical fibers keeps increasing; secondly, data center connections are generally high-speed and short-distance, often utilizing multimode connections; thirdly, as server rates and densities increase, more optical fiber ducts need to be added within the computer room. Compared to traditional mesh bridges or metal ducts, optical fiber ducts are better suited for fiber wiring needs; fourthly, data centers have higher demands for reliability and importance, and their requirements for maintenance and efficiency are also higher compared to traditional industries. Because a data center outage can have a severe impact; finally, data centers have shorter delivery times. Ms. Chen Lan, General Manager and Vice President of CommScope Enterprise Network Greater China, states, "Compared to the construction structure or decoration timing of buildings and campuses, data center delivery times are more urgent, which is an important consideration for customers when choosing data center solutions."
In response, Wu Jian concludes that AI further strengthens the position of data centers, especially as China's data center market still has a lot of development space and potential; AI also makes data center connections more than ordinary connections, at least improving connection quality significantly; moreover, AI makes high-speed interconnectivity a necessity, thus driving the development of some emerging technologies in the fiber field, including LPO, CPO, etc.
To meet these demands, CommScope has introduced the Propel product. Wu Jian states, "The launch of Propel is mainly to meet the demands of some new technologies: firstly, MPO16, which is based on multi-core parallel technology, offering both 8-core and 16-core options. With the acceleration of deployments beyond 400G, the demand for MPO16 will gradually increase; secondly, OM5, which is based on multimode width technology, allowing a single fiber to accommodate multiple wavelengths for parallel multi-wavelength; thirdly, VSFF, which allows the size of connection devices to be reduced to meet future high-density needs. The last one is the fiber connector APC end face, grinding surface, mainly used to reduce return loss."
In summary, Propel provides an integrated platform that can adapt to different needs, from 100G, 400G to even 800G, with strong integration capabilities.
Wu Jian says, "Propel was developed after listening to the voices of many data center customers, including demands for maintenance convenience and support diversity. Therefore, the establishment of the Propel platform is forward-looking, while also responding to current customer needs."
As the future development of artificial intelligence technology continues and data centers grow larger, the demand for computing power and transmission rates will further increase. CommScope's Propel can be seen as a forward-looking technology, grounded in the present but also focused on the future.