South Korea is rapidly cementing its position as a core player in the global artificial intelligence supply chain, following news of an unprecedented capital and capacity cooperation agreement with a major US technology company.
According to reports, South Korean presidential advisor Kim Yong-beom revealed at a press conference last Saturday that Samsung Electronics and SK Hynix are set to finalize total memory chip supply agreements worth up to $950 billion with large US tech firms, including NVIDIA. In this deal, SK Hynix will provide $750 billion worth of long-term memory chips to US companies, while Samsung Electronics will supply chips valued at $200 billion to Broadcom.
SK Group has announced a long-term strategic partnership with NVIDIA worth over $500 billion. This collaboration covers the construction of AI factories, AI cloud platforms, and the supply of next-generation AI memory chips. The two companies plan to build a 2GW AI cloud computing platform and AI factory complex in South Korea, with the first AI factory expected to be operational by 2027.
SK Hynix will also sign a long-term AI memory supply agreement with NVIDIA, ensuring stable delivery of next-generation high-bandwidth memory products, including HBM4. They will also jointly develop memory solutions tailored for future AI computing needs.
Samsung will provide high-performance memory chips for AI accelerators designed by Broadcom and expand its AI chip manufacturing partnership using advanced foundry processes. Additionally, Samsung SDS has established its first strategic partnership with Anthropic, focusing on the Claude large language model to expand into the enterprise AI market. This will involve joint marketing, technical validation, and project development in South Korea, along with training enterprise AI application engineers.
While the $950 billion in orders are impressive, the deals also bring immense pressure to scale up production capacity. Kim admitted that these orders include capacity from "fabs that have not yet been built." This means Samsung and SK Hynix must expand capacity at an unprecedented pace over the next few years. Any delays—whether in equipment delivery, factory construction, or technological breakthroughs—could pose a risk of default.
Global financial markets are closely watching whether this meeting can alleviate recent concerns about a slowdown in AI investment. Due to the slower commercialization of AI relative to infrastructure spending and the potential for decelerating capital expenditure growth from hyperscale cloud providers, shares of major players like Samsung Electronics, SK Hynix, and NVIDIA have fallen by double digits from their peaks. Industry consensus suggests that AI infrastructure investment will maintain steady growth over the next two to three years.
The HBM industry operates differently from previous memory cycles, as it coordinates specifications and quantities with customers before production begins. AI infrastructure investment is planned on a multi-year basis, and the memory supply chain structure is undergoing a transformation.
Last month, South Korea announced an $880 billion investment plan, supported by companies such as SK Hynix, Naver, and Samsung Electronics. The plan aims to solidify the country's AI leadership by aggressively expanding chip production, building data centers, and developing physical AI. Memory chips are seen as the most direct beneficiary of this initiative.
As generative AI drives rapid growth in computing power demand, AI servers are fueling an ever-increasing need for high-bandwidth memory and advanced storage solutions. Samsung and SK Hynix are accelerating capacity expansion to meet the rising procurement needs of global AI companies.
Tianfeng Securities believes that the current memory industry recovery is not solely driven by a single end-market inventory replenishment cycle. Instead, it is the result of a combination of shifting demand structures, a more rational supply expansion, and the recovery of the industry's inventory cycle. As AI computing power construction continues to advance, the importance of high-end memory will only increase, potentially improving the industry's supply-demand balance. Looking ahead, given the evolving demand structure and persistent supply constraints, memory prices and industry profitability are expected to maintain good resilience, with the sector's recovery trend likely to continue. The brokerage recommends focusing on memory modules, memory controller chips, packaging and testing, and upstream equipment and materials.
Related concept stocks include:
Semiconductor Manufacturing International Corporation (00981): SMIC's core products cover several areas, including logic chips, memory chips, and analog chips. First-quarter revenue was $2.5055 billion, up 11.5% year-on-year, with a net profit of approximately $197 million, up 5% year-on-year. For the second quarter, the company guided a sequential revenue growth of 14% to 16% and a gross margin of 20% to 22%.
Hua Hong Semiconductor Limited (01347): Hua Hong's memory business is divided into two main lines: embedded non-volatile memory (eNVM), its core business with the highest revenue share, and standalone NOR Flash, a high-growth second curve. First-quarter revenue reached $660.9 million, up 22.2% year-on-year. Net profit was $20.9 million, a surge of 458.1% year-on-year and up 19.9% sequentially.
GigaDevice Semiconductor (03986): GigaDevice is a leading domestic NOR Flash design company, holding the second-largest global market share. The company is actively building a diversified product portfolio, including NOR Flash, SLC NAND Flash, niche DRAM, MCUs, analog chips, and sensor chips. Central China Securities believes GigaDevice is a leading domestic niche memory and MCU company. With the memory cycle continuing to rise, the company's niche DRAM, SLC NAND, and NOR Flash products are experiencing both volume and price increases. The company is also actively developing customized memory solutions, securing customers in multiple fields, which is expected to contribute to revenue by 2026.
Montage Technology (06809): As a leader in the memory interconnect chip market, Montage Technology is transitioning from a memory interconnect leader to a platform-type interconnect chip company. In the AI era, the company has developed a rich product matrix and holds a leading position in the global market, thanks to its long-term focus on key high-speed interconnect chips between computing power and storage. Its PCIe Retimer products have entered a volume ramp-up phase, and the company is extending along the same technical path into the higher-value PCIe Switch field, which could open up new growth space. As AI shifts from model training to inference, particularly with the development of AI agents, demand for server CPUs is continuously rising, and the company is a major beneficiary of this strong CPU demand.
Comments