Nvidia forms $500 billion memory pact with SK Hynix for long-term HBM supply to power AI infrastructure buildout.
Nvidia has locked in one of the largest supply agreements in semiconductor history, securing memory chips from South Korea's SK Hynix in a deal that could be worth up to $500 billion over several years. The stock barely reacted on Friday, closing down 0.80 percent at €182.00—a sign that even blockbuster announcements are failing to excite a market that had already priced them in.
The announcement coincided with a busy stretch for the chipmaker. At an AI summit in San Francisco attended by South Korean President Lee Jae Myung, Nvidia unveiled the SK Hynix deal while CEO Jensen Huang used a developer conference to push back against rumors that the company's next-generation "Vera Rubin" architecture was facing delays. Huang stated that Vera Rubin is "already in production" and promised "massive production volumes," directly contradicting research firms that had warned of potential setbacks pushing the full rollout of the Rubin Ultra platform into 2028.
Under the agreement, Nvidia has secured a multi-year supply of high-bandwidth memory (HBM) chips critical for its AI systems. The deal includes plans to build large-scale data centers expected to come online in 2027. SK Hynix's subsidiary, SK Telecom, will build a cloud business around Nvidia's Vera Rubin systems, with Nvidia planning capacity requiring 2 gigawatts of power—equivalent to several hundred thousand graphics processors.
The partnership extends beyond simple supply. Nvidia and SK Hynix will jointly develop the next generation of HBM chips, ensuring Nvidia has a stable pipeline given that SK Hynix controls roughly half the global HBM market, with Samsung holding most of the rest.
Nvidia is also investing $1 billion in Korean cloud firm Naver, which is building data centers around Nvidia GPUs. Together with Canadian investor Brookfield, the trio is tripling the capacity of South Korea's state-backed AI factory, GAK Sejong, from 55 megawatts to 200 megawatts by 2028. Brookfield has committed up to $9 billion as the exclusive capital partner, though the term sheet is non-binding. Nvidia's $1 billion contribution is contingent on Naver closing at least $9 billion in committed financing.
Samsung Electronics signed a parallel memorandum of understanding with chip designer Broadcom valued at roughly $200 billion to expand cooperation on memory and foundry technologies for next-generation AI infrastructure, highlighting how supply chain alliances in Asia have become the central battleground for AI hardware supremacy.
Nvidia shares have recovered 30.20 percent from their September 2025 low of €139.78, but remain 10.12 percent below the 52-week high of €202.50 reached in mid-May. The stock sits just above its 50-day moving average of €180.82, which has recently served as support. The 14-day relative strength index of 52.6 signals a neutral posture—neither overbought nor oversold.
Three of Nvidia's biggest customers—Microsoft, Meta, and Amazon—report quarterly earnings this week. For investors, the key focus will be capital expenditure plans for AI infrastructure rather than revenue growth at the hyperscalers. These figures have been the most reliable leading indicator for demand for Nvidia's H200 and Blackwell chips. Alphabet already sent positive signals in its most recent update.
UBS analysts see no threat to Nvidia from the trend toward open-source AI models, arguing instead that companies seeking to optimize internal AI applications will demand more powerful GPUs, not fewer.
Nvidia's next quarterly report isn't due until late August. In the meantime, the market will watch whether Brookfield's non-binding financing commitment becomes a signed contract and how quickly SK Hynix memory deliveries translate into firm purchase orders.