As the demand for AI computing power continues to double every year, SK Group and Nvidia have chosen to lock in the future with a $500 billion agreement. On July 25th, the two sides officially signed a letter of intent for cooperation at the San Francisco AI Summit, covering two core areas: AI factory construction and joint research and development of next-generation memory. SK Group Chairman Choi Tae won described this cooperation as a strategic leap from a computing power consuming country to an AI exporting country.

2 gigawatt AI factory

The core project of the cooperation is SK Telecom's construction of a 2-gigawatt AI cloud factory in South Korea, which adopts Nvidia DSX full stack architecture and Vera Rubin accelerated computing platform, matched with SK Hynix HBM4 high bandwidth memory. The first AI factory is planned to be put into operation in 2027.

Nvidia CEO Huang Renxun commented on this: "South Korea has world-class networks and data centers, leading chip technology, and a huge industrial scale - these elements come together to provide all the raw materials needed for a global AI powerhouse. ”SK Telecom plans to gradually expand similar large-scale computing facilities to Zhongqing and Hunan regions starting from Ulsan in the southeast.

From HBM supply to joint research and development

The binding of storage dimensions is equally deep. NVIDIA and SK Hynix have established a long-term AI memory partnership, with NVIDIA obtaining stable production capacity for high-end HBM and SK Hynix locking in long-term shipping demand. Both parties are simultaneously promoting joint development and optimization of HBM4 and subsequent generations. Huang Renxun previously revealed in an interview that the semiconductor industry may expand tenfold in the next decade, and HBM is the biggest bottleneck of the memory wall for AI chips. Keywords: semiconductor industry, chip

500 billion US dollars is the total amount of two-way transactions, not one-way procurement commitments. But a certain fact is that SK Group is binding Hynix's storage advantages with its telecommunications infrastructure capabilities, creating a complete closed loop from chips to clusters. When Huang Renxun shook hands with Cui Taiyuan in San Francisco, they were betting that the demand for AI computing power would not cool down in the next decade.Editor/Cheng Liting