GF Securities released a research report stating that with the surge in AI demand, data centers within individual facilities are experiencing power and capacity bottlenecks, making inter-data center connectivity an industry trend. NVIDIA's newly launched Spectrum-XGS Ethernet technology can combine multiple distributed data centers into gigawatt-scale AI superfactories, significantly enhancing multi-node communication performance. Leading optical module companies are actively positioning themselves in the Scale-Up market, warranting continued attention to investment opportunities among leading optical module manufacturers.

Event: On August 21st Beijing time, NVIDIA announced the launch of NVIDIA Spectrum-XGS Ethernet. This scale-across technology can combine multiple distributed data centers into a single gigawatt-scale AI superfactory.

GF Securities' main viewpoints are as follows:

**Data Centers Face Power and Capacity Bottlenecks, Inter-Data Center Connectivity Becomes Industry Trend**

With the surge in AI demand, data centers within individual facilities are experiencing power and capacity bottlenecks. To achieve data center expansion, it is necessary to break through building limitations and overcome existing Ethernet network infrastructure bottlenecks including high latency, high jitter, and performance issues.

NVIDIA Founder and CEO Jensen Huang stated: "The AI industrial revolution has arrived, and massive AI factories are the core infrastructure of this transformation. Following scale-up and scale-out technologies, we will provide scale-across technology through the launch of NVIDIA Spectrum-XGS Ethernet, combining data centers across different cities, countries, and even continents into massive gigawatt-scale AI superfactories."

**Multi-Node Communication Performance Enhancement, CSPs Accelerate Deployment**

Spectrum-XGS Ethernet is fully integrated into the Spectrum-X platform, achieving dynamic adaptation between network and data center facility distances through algorithms. Spectrum-XGS Ethernet has nearly doubled the performance of NVIDIA's collective communications library, accelerating multi-GPU and multi-node communication performance, and achieving predictable performance for remote AI clusters.

North American cloud providers have begun accelerating the deployment of Spectrum-XGS platforms, with CoreWeave becoming one of the first companies to use Spectrum-XGS Ethernet. The Spectrum-X Ethernet network platform is designed for hyperscale AI factories, providing 1.6 times the bandwidth density of traditional Ethernet and achieving ultra-low latency. NVIDIA Spectrum-XGS Ethernet is now an extension of the NVIDIA Spectrum-X Ethernet platform and is currently available for delivery.

**Optical Module Companies Advance into Scale-Up Market, Continue Monitoring Investment Opportunities in Leading Optical Module Manufacturers**

To ensure that subsequent network architectures adapt to rapidly growing computing power demands, NVIDIA plans to introduce optical interconnects in the Scale-Up field, with leading optical module companies actively positioning themselves. From AOC to light sources to optical engines, the firm believes that leading optical module companies will leverage their strong R&D capabilities, flexible solution design, and agile global delivery to help CSP customers reduce equipment costs, drive technological iteration in the Scale-Up market, and play important roles.

The firm continues to recommend leading optical module companies.

**Risk Warnings**: Risks of AI application development falling short of expectations; risks of localized shortages in AI hardware; risks of tightening overseas export controls; risks of cloud provider capital expenditure falling short of expectations.