South Korea's chip giants strike $700 billion in AI deals, and the fine print comes later
SK Group and Samsung used a single San Francisco summit to line up on opposite sides of the AI silicon race, though both stopped short of binding contracts.
In the span of a few days, South Korea's two largest chip powers each unveiled a sprawling tie-up with an American technology giant, and between them they attached price tags approaching US$700 billion. Both announcements came out of the same event, an AI summit in San Francisco held during President Lee Jae Myung's visit to the United States, where semiconductors anchored a wider push for Korea-US industrial cooperation.
The bigger of the two paired SK Group with Nvidia. On July 24 the companies said they had signed letters of intent covering a partnership valued at more than $500 billion over five years. Under the plan, SK Telecom will build a two-gigawatt AI factory in Korea using Nvidia’s Vera Rubin systems, with the first phase due to come online in 2027. The hardware will run on SK hynix's HBM4 memory, and SK hynix separately agreed to a long-term arrangement to supply and help develop next-generation memory for Nvidia.
"By leveraging SK hynix's AI memory and SK Telecom's AI infrastructure capabilities, SK will collaborate with NVIDIA to build a world-class AI factory," SK Group Chairman Chey Tae-won said in a statement, adding that the goal was to help Korea "become a global hub that drives AI innovation."
The second deal brought together Samsung Electronics and Broadcom. The pair signed a memorandum of understanding they expect to be worth more than US$200 billion through 2030, covering memory, chip manufacturing and packaging. Samsung will supply high-bandwidth memory for Broadcom's AI accelerators and produce Broadcom chips, including networking silicon, on its two-nanometre process. The two also plan to work on advanced packaging that stacks memory and logic more closely together.
The Broadcom agreement matters to Samsung well beyond the headline number. Its contract manufacturing arm has struggled to win flagship customers and holds an estimated 7% of the global foundry market, far behind Taiwan's TSMC. Winning a designer of Broadcom's stature is a public endorsement of Samsung's 2nm roadmap, which has been dogged by questions over production yields. Analysts have pointed to the pull of Samsung's one-stop model, which keeps logic manufacturing, memory and packaging under a single roof, at a moment when TSMC's leading-edge capacity is fully booked and its prices are climbing.
Read side by side, the two deals map onto a rivalry that has been building for years. SK hynix, the current front-runner in high-bandwidth memory, is deepening its ties to Nvidia and its graphics processors. Samsung, still working to close the memory gap, is casting its lot with Broadcom, whose custom accelerators sit inside the data centres of hyperscalers such as Google and Meta. Korea's two memory makers have, in effect, each attached themselves to one of the two dominant camps in AI computing. Memory is the common thread by design: demand for the latest HBM is running ahead of what the industry can make, and Samsung's output for this year is already spoken for.
The setting was as much diplomatic as commercial. The San Francisco announcements formed part of a broader Korea-US semiconductor package tied to President Lee's trip, and sat alongside parallel commitments from other Korean names, among them NAVER and Hyundai. The signal from Seoul was that its industrial base means to supply the American AI buildout rather than watch it from the sidelines.
The figures, though, reward a careful read. The SK-NVIDIA partnership rests on letters of intent and the Samsung-Broadcom pact on a memorandum of understanding. Neither is a binding contract, and the companies disclosed no committed investment, project timeline or commercial terms. The dollar totals describe an expected value of business over five years, not money spent or orders firmly placed. Investors seemed to note the difference: NVIDIA's shares were little changed after the news.
Execution is the other open question, most of all for Samsung. Industry watchers say the Broadcom collaboration will only convert into revenue once Samsung clears customer qualification, lifts its 2nm yields to stable levels and shows it can mass-produce the packaging the deal calls for. Its yields have reportedly improved but still hover near the level the industry treats as the floor for dependable production.
What the week delivered, then, was two of the largest chip announcements of the year and an unmistakable statement of intent. Whether the headline totals hold will come down to contracts, capacity and yields that are still being settled.