The Value Potential of AI Ether Network

Network Connection

NVIDIA is actively embracing Ethernet technology. In March, it launched the SpectrumX-800 AI-optimized network, and the May financial report guidance indicates that the Spectrum Ethernet platform will contribute billions of dollars in revenue next year, showing that it is actively promoting network connectivity towards Ethernet development.

Ethernet and IB networks each have their advantages:

• Ethernet is widely used in local area networks, utilizing standard devices, offering high compatibility, low cost, and good flexibility.
• IB network is suitable for high-performance computing and AI scenarios, with excellent performance, high bandwidth, and low latency.

Ethernet has a wide user base, but its performance in the AI field is lacking. The IB network has obvious advantages backed by Nvidia, but it is highly priced, resulting in a lower quantity of hardware purchases under the same budget.

Looking ahead, Ethernet is expected to become the mainstream network interconnection solution:

• The performance advantages are gradually becoming apparent.
• Stronger capabilities for large-scale clusters
• Openness is the core advantage.
• Adapt to the computing power demand for AIGC development

Currently forming two major camps:

1. The Ultra Ethernet Alliance UEC: Composed of device manufacturers such as AMD and Arista, as well as cloud providers like Meta and Microsoft.
2. NVIDIA Spectrum-X Platform: A Comprehensive System Architecture Combining Software and Hardware

Regardless of which solution dominates, it will promote the expansion of the switch market and industry development. It is recommended to focus on the entire industry chain from chips, manufacturing to brand owners.

Optical Module

LPO( linear drive pluggable optical module ) solution or the best choice for Nvidia AI Ether:

• Low power consumption, low latency, low cost, hot-swappable
• Needs to cooperate with the switch, with high requirements for collaboration in the vendor supply chain.
• Leading companies like Zhongji Xuchuang and Xinyi Sheng have greater advantages.

The AI-driven optical module market is expected to grow by 45% year-on-year in 2024. The surge in demand for 400G and 800G is driving significant revenue growth. The market size is projected to reach $22.4 billion by 2029.

Integrated photonic technology, especially silicon photonics, provides low-cost scalable solutions for high-speed short-distance optical links. LPO, as a transitional solution, can reduce power consumption and latency, making it suitable for various connections in the AI field.

Storage Technology

Micron Technology's HBM memory sales are expected to reach hundreds of millions of dollars this fiscal year and rise to several billion dollars next year. By 2025, HBM supply negotiations are largely expected to be completed.

Storage prices are expected to rise in the second half of 2024 to 2025, driven by demand from AI devices and data centers. HBM demand has significantly increased, and capacity constraints are affecting the overall supply of DRAM.

Micron's 12-layer HBM3E is currently undergoing validation and will enter production based on customer progress. The industry is expected to gradually transition from 8 layers to 12 layers by 2025.

Telecommunications Market Trends

T-Mobile is acquiring the majority of US Cellular's wireless business for $4.4 billion, including customers, stores, and 30% of spectrum assets. The transaction is expected to be completed by mid-2025, enhancing T-Mobile's market competitiveness.

Development of AI Mobile Phones

Canalys predicts:

• In 2024, 16% of global smartphone shipments will be AI smartphones.
• The share will reach 54% by 2028.
• The average annual compound growth rate of the AI mobile phone market from 2023 to 2028 is 63%.

AI capabilities will gradually become more widespread from high-end models to mid-range ones. Consumers are increasingly interested in AI functionalities, but they still focus on privacy and data security.

Public Cloud Market

Gartner predicts that global public cloud spending will reach $675.4 billion in 2024, a year-on-year increase of 20.4%. Main drivers:

• Generative AI
• Modernization transformation

IaaS and PaaS are growing the fastest, at 25.6% and 20.6%, respectively. SaaS remains the largest expenditure area, expected to reach $247.2 billion.

National Integrated Circuit Industry Investment

The country's "Big Fund" Phase III officially established:

• Registered capital of 344 billion yuan
• The six major state-owned banks account for 33.14%
• China Mobile Capital and others participated in the investment.

Key support for integrated circuit manufacturing, while also considering design, packaging testing, equipment, materials and other aspects.