Amidst market fluctuations caused by the US-Iran tensions, a group of 'shovel sellers' have quietly capitalized on the AI computing power surge, reaping nearly hundredfold profits.

On the evening of April 7th, Shannon Semiconductor, a key player in the semiconductor supply chain, unveiled its performance forecast, projecting a staggering net profit of up to 1.48 billion yuan for the first quarter of this year, marking a nearly 90-fold increase year-on-year. Such an extraordinary growth rate is a rarity in the annals of A-share market history.

The driving force behind this remarkable performance surge is the explosive demand for generative AI applications, which has propelled the memory chip industry to new heights.

This phenomenon also underscores that while the global spotlight is on the algorithmic rivalry among AI models, the competition for 'computing power granaries' is intensifying behind the scenes.

Q1 Net Profit, Excluding Non-Recurring Items, Soars Nearly Hundredfold

Shannon Semiconductor's performance forecast has far exceeded market expectations. The company anticipates a net profit attributable to shareholders ranging from 1.14 billion to 1.48 billion yuan for the first quarter, representing a year-on-year surge of 6714.72% to 8747.18%. Even more striking is the performance of net profit excluding non-recurring items, which has skyrocketed by up to 9713.23% year-on-year, indicating a near hundredfold increase in net profit.

This explosive growth is attributed to the company's keen insight into industry trends. Shannon Semiconductor explicitly stated in its announcement, 'The burgeoning demand for generative AI applications has fueled industry prosperity, leading to a rise in enterprise-level storage product prices and an enhancement in the company's profitability.'

Similarly, several A-share memory sector companies have also posted impressive first-quarter results. Demingli expects its revenue to surge by 483.05% to 522.98% year-on-year, with net profit turning from a loss of 69.08 million yuan in the same period last year to a profit of 3.15 billion to 3.65 billion yuan. Biwin Storage's net profit for the first two months has soared by 921.77% to 1086.13% year-on-year.

In the international arena, memory behemoth Samsung Electronics has also reported outstanding performance during the same period. On April 7th, Samsung released its performance guidance, revealing an operating profit of approximately 57.2 trillion won (about 261.1 billion yuan) for the first quarter, soaring by 755% year-on-year and 185% quarter-on-quarter, marking the highest increase in recent years.

Industry Prosperity on the Rise: An AI-Driven Memory Chip Demand Revolution

With the advent of large-scale models like ChatGPT and Sora, the computing power required for AI training and inference has grown exponentially. Supporting this computing power are not only NVIDIA's GPUs but also memory chips, particularly high-bandwidth memory (HBM).

HBM, as the current memory solution capable of meeting the ultra-high bandwidth demands of AI chips, has become a focal point for major AI server manufacturers. It is reported that the global HBM production capacity is nearly monopolized by Samsung, SK Hynix, and Micron. Shannon Semiconductor has forged a deep partnership with SK Hynix and holds the distribution rights for its HBM products.

Industry data indicates that the price of mainstream HBM3e products has surged from about $5 per GB a year ago to over $15 currently, more than tripling. Meanwhile, the price of enterprise-level SSDs for AI training scenarios has nearly doubled in the past six months.

'The demand for memory chips in AI servers differs significantly from that in traditional servers,' noted a memory industry analyst. 'It requires not only large capacity but also extremely high bandwidth and reliability. This structural change in demand has rapidly transformed the high-end memory chip market from a buyer's market to a seller's market.'

Industry Chain Dividends

In fact, the beneficiaries of the AI-driven memory chip boom extend well beyond the original equipment manufacturers. From distributors to solution providers, the entire industry chain is reaping the rewards.

Shannon Semiconductor's business model is highly illustrative. As an important partner of SK Hynix in China, it serves as a 'bridge,' distributing the original manufacturer's memory products to domestic AI server manufacturers and cloud computing companies. Against the backdrop of tight HBM supply and rising prices, this distribution right has become a scarce and valuable resource.

Demingli represents another type of beneficiary. Starting from memory control chip design, the company has gradually evolved into a memory solution provider, capitalizing on the opportunity of AI computing power penetrating various industries.

'The biggest difference between this round of memory chip market and previous ones is the fundamental change in demand structure,' analyzed a senior memory industry insider. 'In the past, demand primarily came from smartphones and PCs, with obvious cyclicality. Now, the incremental demand mainly comes from AI computing power centers, representing a new and sustained growth market.'

The position in the industry chain determines profit distribution. Original equipment manufacturers gain the largest profits due to technological barriers, while distributors and solution providers share in the industry dividends through their customer resources and technical service capabilities.

The Debate Over Cyclicality

The memory chip industry is notorious for its strong cyclicality. This time, can AI demand break the industry's curse of 'sharp rises followed by sharp falls'?

Optimists believe that this round of market conditions is fundamentally different. 'Traditional memory chip cycles are driven by consumer electronics, while AI computing power demand is structural and long-term,' stated Huatai Securities in a research report. 'As AI transitions from training to inference and from the cloud to the edge, memory chip demand will experience multiple growth curves.'

Pessimists, however, caution against potential risks. Demingli admitted in its announcement, 'The prosperity of our industry is influenced by the relationship between upstream raw material supply and downstream market demand. Future changes in industry supply and demand or fluctuations in raw material prices may lead to volatility in our company's performance.'

The reality may lie somewhere in between. In the short term, the HBM supply-demand contradiction is unlikely to be resolved. Samsung, SK Hynix, and Micron have significantly increased their capital expenditures, but HBM production line construction is time-consuming and technically challenging, with capacity release taking time. TrendForce estimates that the HBM supply-demand gap will remain at 10%-15% in 2026.

In the long term, technological advancements may reshape the competitive landscape. HBM3e has begun mass production, and HBM4 research and development are accelerating. Domestic companies like Yangtze Memory Technologies and ChangXin Memory Technologies are making continuous technological breakthroughs. New memory architectures like CXL (Compute Express Link) are gradually maturing. These variables may all influence the future industry landscape.

Behind Shannon Semiconductor's hundredfold growth lies the business logic that supply chain position determines value distribution. Samsung Electronics' performance outburst proves the value of a comprehensive industry chain layout.

For China's memory industry chain, this round of market conditions presents both opportunities and challenges. On the one hand, the huge demand for domestic AI computing power construction provides market space for local memory companies. On the other hand, in high-end memory areas like HBM, domestic companies still have a significant technological gap compared to Samsung and SK Hynix.