Samsung Electronics' semiconductor industry version, especially Samsung Foundry, has been facing major challenges. Although it has worked hard to solve the problem of low yields in the first-generation process, it will inevitably affect the next...
Samsung Electronics' semiconductor industry version, especially Samsung Foundry, has been facing major challenges. Although it has worked hard to solve the problem of low yields in the first-generation process, it will inevitably affect the next generation's 1.4 nanometer progress. In the popular high-frequency wide memory (HBM) market, Samsung is also unable to enter the supply chain of GPU manufacturer NVIDIA, allowing the memory dragon head to compete with its rival SK Hynix.
Notebookcheck reports that if Samsung, in trouble, can return to 2018 and come back to another, there may be different situations now. Because at that time, Samsung refused to recommend NVIDIA cooperation.
NVIDIA executive president Jensen Huang visited Samsung in 2018, hoping to expand HBM cooperation. NVIDIA hopes that the three areas will have deeper cooperation, for the development of HBM memory, new processes, and CUDA platform cooperation. If it had been concluded at that time, Samsung would have long been an important partner of NVIDIA.
Samsung refused to reject NVIDIA's proposal, and Huang Ren later expressed his frustration that Samsung has no one to discuss long-term strategies with me. Market insiders believe that this not only implies that NVIDIA is looking forward to cooperation, but also says that Samsung was arrogant at that time and decided on difficulties or communication problems. Samsung is unable to discuss long-term strategies with NVIDIA and is likely to face multiple criminal investigations with President Lee Jae-kyo. Due to high-level legal difficulties, it has a significant impact on the company's strategic decision-making process, making it difficult to make decisive and forward-looking decisions.
After this cooperation, it has had a profound impact on the global semiconductor structure. In recent years, NVIDIA has become a hot beneficiary of AI gold rush, and the demand for AI chips has reached an unprecedented level. HBM is the most important part. In order to achieve AI recommendations, H200 GPU must have sufficient high-volume memory to increase the strategic importance of HBM in the AI era.
Since Samsung failed to seize the opportunity to cooperate with NVIDIA, most of the HBM chips were supplied by SK Hynix. This transaction is a huge benefit for SK Hynix. The stock price has grown by more than 220% in five years, fully demonstrating the value of HBM's big market, and SK Hynix has also received heavy returns.
Samsung's failure to cooperate with NVIDIA will also have an impact on Samsung's crystalline OEM. Telco produces NVIDIA AI chips and also enjoys the exclusive position of NVIDIA AI GPUs. Although market news pointed out that NVIDIA may cooperate with Samsung to produce game chips in 2 nanometers, there is no news yet Samsung will produce key AI chips, representing a high-profit and high-demand AI chip market. Telco is still NVIDIA's first choice.
Although the impact of short-sight decisions is difficult, Samsung has also actively tried to win the NVIDIA and Ultramicro (AMD) part of HBM orders. Samsung has planned to supply two HBM4 memory devices in early 2026. After Samsung realized HBM's key to AI chips, it has devoted itself to R&D and production, striving to recover its leading position in the memory market and repair its relationship with NVIDIA.
Conclusion: Samsung and NVIDIA failed to cooperate in 2018, which had a significant negative impact on Samsung's HBM market status and crystalline foundry development. It not only allowed SK Hynix and Taiwan Power to achieve advantages, but also highlighted the importance of rapidly changing technology industry, high-level decision-making and long-term vision. Although Samsung is chasing it out loud, the mistakes made in the past seem not so easy to save.
