In the first half of this year, many A-share listed companies achieved remarkable performance, driven by factors such as AI, which led to continued growth in storage industry demand. In the second half of the year, demand in the PC (personal computer) and smartphone markets is expected to rebound, thereby driving an increase in shipments of storage products such as NAND Flash.

According to market voices, Guanghe Technology recently stated that demand for general-purpose servers remains stable and growing. Compared to demand for general-purpose servers, demand for AI servers has a smaller base but is growing rapidly, and the proportion of AI server business will gradually increase.

A new wave of AI server demand is poised to explode.

01

Ultraman and Musk Join Forces

OpenAI and Musk are simultaneously attacking AI infrastructure, igniting a new wave of AI server demand. Contract manufacturers such as Quanta Computer, Wistron, Wistron NeWeb, and Inventec in Taiwan will benefit from this opportunity.

OpenAI's ChatGPT has become incredibly popular in the global generative AI market, with its weekly active user count surpassing 200 million, double that of last November. It has attracted 92% of the Fortune 500 companies to use its ChatGPT services, and its commercial model subscriptions have surpassed 1 million. It is rumored that OpenAI executives are discussing higher-priced subscription plans for its next large language model.

Industry insiders point out that the surge in ChatGPT users not only reflects a significant increase in market acceptance of AI tools but also demonstrates the corporate world's emphasis on generative AI technology. OpenAI currently dominates the AI landscape and holds a leading position.

As OpenAI's commercial model user base grows steadily, news about the next-generation GPT 5 has also emerged. Jung-Bae Lee, President of Samsung's Memory Business, revealed at SEMICON Taiwan 2024 that the new GPT 5 will have up to 3 to 5 trillion parameters and will be trained using 7,000 NVIDIA B100 chips, with a potential launch this year.

Separately, Ultraman's efforts to secure global investor funding for AI infrastructure are gradually coming to light. It is rumored that the initiative will start in various US states, involving the construction of data centers and the expansion of semiconductor fabs, with investments expected to reach tens of billions of dollars, continuously fueling the AI server wave.

Similarly, Musk's AI layout is also in full swing. He announced that his startup AI company, xAI, has adopted 100,000 NVIDIA H100 chips to build Colossus, currently the world's most powerful AI training system and supercomputer. It was recently officially launched.

Musk also revealed that the system's scale will double in the coming months, with the number of H100 chips increasing from 100,000 to 200,000, including 50,000 H200 chips.

With Ultraman and Musk joining forces, a powerful growth engine for AI chip demand is ignited, fueling synchronized growth in AI server demand. Contract manufacturers such as Quanta Computer, Wistron, Wistron NeWeb, and Inventec will soar on this AI growth train. Yang Qi, Executive Vice President of Quanta Computer, also noted that AI development is gradually improving in the second half of the year and will only get better. Although some products are entering a transition phase, demand for existing products such as NVIDIA's Hopper architecture remains robust in the market.

02

Can AI Servers Save the Slumping Mobile Phone Market?

The smartphone industry is facing a new turning point.

Foxconn, Apple's primary manufacturing partner, reported record-high sales of NT$548.3 billion (US$17.1 billion) in August, a significant increase from July's 22% growth. Foxconn's revenue has started to recover from the long-term slump in the smartphone market, thanks to its growing server business, which includes NVIDIA AI accelerators for data center operators. In July, Foxconn projected revenue growth for the rest of the year, reversing several quarters of decline. Its share price has risen nearly 70% in 2024.

Why? Foxconn has set a goal of securing a 40% share of the global AI server market, relying on its relationships with many of the world's largest technology companies and its manufacturing expertise.

Bill Gates once declared, "ChatGPT is as important as the invention of the internet and will change the world."

This change essentially lies in transforming human-machine interaction, which is gradually becoming a reality through AI-powered smartphones. On January 18, Samsung officially unveiled its new flagship S24 series, emphasizing AI capabilities and proclaiming the dawn of a "new era of mobile AI." The new devices introduce various AI features, including video AI processing, AI chatbots, image processing, and real-time call translation.

As September arrived, Huawei and Apple held press conferences on the same day, with AI undoubtedly taking center stage at Apple's event.

Apple Intelligence, a comprehensive AI suite introduced by Apple at its Worldwide Developers Conference in June, is described as a generative AI system based on personal scenarios, providing intelligent functions such as assistance and image creation. Essentially, it serves as an AI large model for Android phone manufacturers, empowering all Apple devices and enabling intelligent networking within the iOS closed-loop system.

Previously, it was widely anticipated that Apple's iPhone 16 would be its first AI-powered phone, potentially sparking a replacement wave for Apple devices. Stimulated by this anticipation, Apple-related stocks in the A-share market performed impressively in the lead-up to Apple's fall product launch event.

An AI phone is defined as one that possesses general AI functions supported by current AI phones, such as AI image removal, AI document organization and summarization, meeting minutes generation, and automatic sound differentiation. Some AI phones also offer proactive services, learning user habits and offering relevant services in different contexts, including features like "one-click taxi booking."

AI phones must also be equipped with large end-side AI models and have the capability to access cloud-based large models. Due to the massive data processing requirements of large AI models, AI phones have higher requirements for memory, chips, storage, heat dissipation, and battery life compared to ordinary phones. In comparison, Apple's iPhone 16 series has indeed been upgraded to an AI phone in terms of both hardware and software, marking Apple's first AI phone.

With hardware upgrades such as chips, memory, and camera buttons, Apple has clearly prepared its devices for future needs, particularly those requiring large amounts of memory to process AI workloads for large language models. These upgrades pave the way for personalized and enhanced features driven by Apple's intelligence. Specifically, the iPhone 16 abandons the practice of using last year's Pro chip in its lower-end models, upgrading the entire lineup to the A18 series chip and increasing memory from 6GB to 8GB across the board. Its camera button also incorporates Apple's new AI-driven visual intelligence features, enabling users to search for content and products using images, display business information and rating percentages for photographed stores, and boasts comprehensive upgrades in heat dissipation and battery life.

Huawei's Mate XT, the world's first tri-foldable phone, has also coincidentally positioned AI capabilities as a major selling point, attracting numerous fans. Like other Huawei flagship products, the Mate XT boasts exceptional communication capabilities, supporting Tiantong satellite communication for connectivity in areas without terrestrial network coverage. In daily use, it leverages AI Lingxi algorithms to optimize network selection, ensuring stable and seamless connectivity even in weak network environments such as elevators and subways.

Furthermore, the Mate XT integrates deeply with the Pangu large model, enabling Xiaoyi to comprehend user intent and function as a text editor, information consultant, translation expert, and more, enhancing users' daily lives, work, and studies. Its AI erasure, AI cloud enhancement, and AI image expansion capabilities are also outstanding, easily rescuing low-quality phone images.

Today, global technology companies are racing to incorporate AI into their products, with smartphones expected to become one of the most critical battlegrounds.

03

The AI Computing Power Black Hole: Only Server Upgrades Can Counteract It

The computing power demand generated by AI is virtually limitless. In the cloud computing era, the focus was more on optimizing idle computing power and innovating the middleware and software layers, essentially a form of sharing economy rather than a technological revolution. However, in the face of new-era demands, enhancing server and cluster computing power has become the most crucial bottleneck, necessitating hardware technology upgrades that we can no longer avoid.

The primary difference between AI servers and traditional servers lies in their use of high-performance GPUs and HBMs, ultimately delivering outstanding heterogeneous computing capabilities. This makes AI servers the only viable hardware solution to fill the computing power black hole. For example, in AI training servers, GPUs account for over 70% of costs, while in basic servers, this proportion is less than 20%. The evolution of server platforms has shifted from following the generational changes of Intel CPUs to those of NVIDIA GPUs. The value of individual servers has soared from around US$10,000 to approximately US$200,000. Undoubtedly, NVIDIA has emerged as the biggest winner in the AI server revolution, with its market value surging from US$300 billion to US$3 trillion.

According to the latest research report from CITIC Securities, the computing power industry chain is undergoing unprecedented transformation, and the profit elasticity of chips and AI servers is expected to quadruple by 2025 compared to 2023.

In terms of user experience, the design and functionality of AI servers are continually being optimized. For instance, new servers are equipped with intelligent cooling systems that maintain stable operation under high loads, reducing failure rates. This is particularly crucial for industries requiring 24/7 uninterrupted operation, such as financial services, healthcare, and big data analysis. Additionally, users can remotely monitor and manage servers through cloud services, enhancing operational flexibility and boosting overall efficiency.

In the market competition, this AI server undoubtedly occupies a favorable position. Compared to similar products on the market, it boasts significant advantages in both performance and cost-effectiveness. On the one hand, proper cost control allows end-users to enjoy more competitive prices; on the other hand, its superior performance attracts the attention of large enterprises and research institutions, further boosting its market share.