Nvidia and SK Hynix sign a $500 billion deal to lock down AI memory supply
The agreement covers specialised memory chips that every powerful AI processor needs, and it comes with plans for massive new data centres in South Korea.
Key points
- Nvidia and South Korean chipmaker SK Hynix announced a deal on Friday, 18 July 2026 worth up to $500 billion over multiple years.
- The agreement includes large-scale data centres expected to come online in 2027, targeting 2 gigawatts of power capacity.
- SK Hynix affiliate SK Telecom will build a cloud business using Nvidia's Vera Rubin systems, Nvidia's next generation of AI computing hardware.
- Nvidia is separately investing $1 billion into Naver, a Korean cloud company building data centres around Nvidia's chips.
- SK Hynix listed on the Nasdaq stock exchange earlier in July 2026 to help finance its infrastructure expansion.
Nvidia has struck what could be a $500 billion supply agreement with South Korean memory chipmaker SK Hynix, securing a critical ingredient for its artificial intelligence processors. The deal was announced Friday evening in San Francisco at an AI summit attended by South Korean officials, including President Lee Jae Myung, as reported by CNBC Tech.
Why does Nvidia need this deal so badly?
The global AI boom has created a shortage of a specialised component called high-bandwidth memory, or HBM, the type of fast, high-capacity memory chip that sits directly on Nvidia's AI processors and lets them handle enormous amounts of data at speed. Without enough HBM, Nvidia cannot build enough of the chips that data centres, hospitals, banks and tech companies are all scrambling to buy.
SK Hynix is the world's leading producer of HBM, according to industry analysts. Locking in supply now gives Nvidia a significant advantage over rivals.
"The expansion will include a co-develop opportunity for us on the next-generation SK Hynix AI memory, and this will help us secure a stable supply of HBM memory," said Raj Mirpuri, Nvidia's enterprise vice president, on a call with reporters.
What is actually being built?
The deal goes well beyond a purchase order. It includes the construction of large-scale data centres, the giant warehouse-like facilities packed with computer hardware that power AI services, expected to come online in 2027.
Nvidia says it is targeting enough computing capacity to require 2 gigawatts of power, roughly the output of two large power stations, which would house hundreds of thousands of GPUs (graphics processing units, the specialised chips that do the heavy number-crunching AI needs).
SK Hynix's affiliate, the telecoms giant SK Telecom, will build a cloud computing business on top of Nvidia's Vera Rubin systems, the company's next generation of AI hardware platforms.
Separately, Nvidia said it will invest $1 billion into Naver, a major Korean internet and cloud company, to build additional data centre capacity. That project is targeting 200 megawatts of capacity and is open to customers in South Korea and globally who want to book AI computing time before the facility is finished.
What does this mean for ordinary people?
This deal matters beyond the headline figures. Until now, the biggest AI infrastructure projects have been driven almost entirely by a small group of US tech giants: Microsoft, Google, Amazon and Meta. South Korea's government and its corporate conglomerates stepping in as major builders signals that AI computing power is spreading to new regions and new players.
For consumers and businesses in South Korea, and eventually further afield, that means more places to access AI services and, over time, greater competition on price and availability.
SK Hynix also listed its shares on the Nasdaq stock exchange earlier in July 2026, a move designed to raise the capital needed to fund exactly these kinds of large-scale infrastructure projects.
