"The Next Bottleneck in the AI Era Is 'Connectivity'... Optical Communication Emerges as the Key Battleground"

Samil PwC Releases Report on Optical Communication Technology
Electrical Connectivity Reaches Limits Amid Surging AI Computation
"AI System Architectures Must Prepare for Change"

As artificial intelligence (AI) data centers continue to scale up, the bottleneck that determines AI competitiveness is shifting from securing semiconductors such as graphics processing units (GPUs) and high-bandwidth memory (HBM) to the domain of "connectivity" that links these components together. Analysts noted that the networks that organically integrate thousands to tens of thousands of GPUs, and the improvements in power efficiency, are emerging as new battlegrounds in the AI era.

On August 13, Samil PwC released a report titled “Now the Bottleneck is Connectivity: Opportunities in Optical Communication for the AI Era and Korean Companies.” The report analyzes how the data transmission methods for AI infrastructure are transitioning from traditional copper-based electrical connections to optical communications, and presents key implications for Korean companies.

Optical Communication Technology Emerges as the Next Bottleneck

According to the report, existing copper cables face a limitation: as network bandwidth increases, power efficiency declines. The report pointed out that, in the current high-bandwidth connectivity environments used by major AI data centers, the effective transmission range of copper can be limited to just a few meters. Moreover, as AI data centers expand in size, inter-rack distances become larger, leading to signal attenuation issues for electrically transmitted signals.

The report introduces optical communication technology—which has recently emerged as a promising alternative to overcome the limitations of electrical connections—by dividing it into three distinct phases.

The first phase involves introducing optical communication for long-distance connections between racks, establishing a low-latency, low-power environment. While data is processed in the form of electrical signals, it is converted to light for transmission and travels along optical fibers. Optical fiber cables exhibit minimal signal loss, enabling ultra-high-speed communications over medium to long distances. Optical transceivers are responsible for converting data from electrical signals to optical signals.

The second phase is co-packaged optics (CPO), which integrates an optical engine with a packaged component such as a switch chip. This significantly reduces the length of internal electrical signal pathways within devices, improving power efficiency. The report states that CPO is currently in its early commercialization stage, and that the global market is projected to grow at a compound annual rate of 38 percent, reaching USD 580 million by 2030.

The third phase refers to optical conversion of GPU-to-GPU chip interfaces, which is expected to overcome the constraints of physical distance between computing chips. This is regarded as a next-generation technology with the potential to transform AI system architecture.

Nvidia and TSMC are Investing as Well... Korean Firms Seek Competitiveness

The report shows that leading AI infrastructure companies are leveraging optical communication technology to maximize the speed and efficiency of data connectivity and processing. In March, Nvidia formalized its optical communication strategy by investing USD 2 billion each in optical technology companies Lumentum and Coherent. TSMC, the world’s largest foundry, is also building a manufacturing base for co-packaged optics by developing silicon photonics chips and related technology.

Korean companies are mainly active in the optical cable and optical transceiver segments. Taihan Optical Communication has gained cost competitiveness through vertical integration and has signed supply contracts for optical cables for U.S. AI infrastructure. OE Solutions and BITQuantum are developing optical transceiver products to respond to expanding market demands.

Meanwhile, the report suggests that the advancement of optical communication could shift AI systems from “integration-centric” to “flexible, distributed architectures.” In a high-speed, optical communication-based connectivity environment, the constraint of placing GPUs and HBM in close physical proximity is reduced, which could change memory placement and system design methodologies. The report advises that Korean memory companies should consider optimized memory usage structures and architectures for optical environments, and that proactive R&D and strategic planning are required.

Further details of the report can be found on the Samil PwC website. Seunghwan Lee, Samil PwC AX Node Leader (Partner), commented, “Just as Korea established a technological edge in memory semiconductors, the country could also develop specialized strengths in the optical communication sector. We need to leverage our accumulated semiconductor capabilities and advanced manufacturing competitiveness to expand into adjacent areas such as silicon photonics and high-speed interconnect chips.”

This content was produced with the assistance of AI translation services.