Nvidia Secures Half-Trillion-Dollar Memory Deal With SK Hynix to Dominate AI Hardware

A Deal That Reshapes the AI Supply Chain

As global demand for AI compute continues to outpace available hardware, Nvidia and South Korea's SK Group have announced an expanded, multi-faceted partnership valued at over $500 billion. The announcement, made on Friday, July 25, at an AI Summit in San Francisco, signals that the race to control the physical building blocks of artificial intelligence has entered a new and far more aggressive phase.

The deal was announced at an AI summit in San Francisco with South Korean officials, including President Lee Jae Myung. The two sides signed letters of intent to formalize the agreement, which spans from AI factory construction to AI memory supply. The sheer scale of the commitment — half a trillion dollars — dwarfs virtually anything previously seen in the semiconductor industry.

Locking Down the Most Critical Component in AI

While Nvidia has dominated headlines with its GPU architectures, the quiet crisis has been high-bandwidth memory (HBM) supply. These memory modules, which sit directly on the GPU package and deliver dramatically faster data throughput than traditional DRAM, have become the gating factor for how many AI accelerators can actually ship.

Nvidia enterprise vice president Raj Mirpuri confirmed the strategic intent, saying the deal "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." SK Hynix, which currently controls roughly 60 percent of the global HBM market, will enter a long-term supply arrangement with Nvidia and co-develop next-generation high-bandwidth memory covering applications from large language model training to physical AI.

By securing $500 billion worth of supply — a figure that dwarfs typical semiconductor supply agreements by orders of magnitude — Nvidia is essentially cornering the market on the component its competitors desperately need. AMD, which has been positioning its MI300 series as a credible alternative to Nvidia's data center GPUs, relies on the same HBM supply chain. The competitive squeeze could be felt for years.

A Massive Physical Buildout to Match

Under the partnership, SK Telecom will build a 2-gigawatt AI Cloud in Korea using the Nvidia DSX platform, deploying Nvidia Vera Rubin accelerated computing powered by SK Hynix HBM4 memory, with the first AI factory planned to come online in 2027. To put that power figure in perspective, 2 gigawatts is enough electricity to power roughly 1.5 million average American homes — all dedicated to running AI workloads.

The agreement is also a sign that massive AI infrastructure buildouts are moving beyond a handful of hyperscalers, with foreign governments and massive conglomerates starting to get involved. Nvidia also announced Friday that it would invest $1 billion into Naver, a Korean cloud company building data centers around its GPUs.

Broader Ripples Across the Industry

In a separate deal, South Korean giant Samsung Electronics said it signed a memorandum of understanding with chip designer Broadcom to expand their collaboration across memory and foundry technologies. That deal, worth an estimated $200 billion, would help support the next generation of AI infrastructure. Together, the two announcements represent nearly $700 billion in AI-related commitments tied to South Korean chipmakers in a single weekend.

AI startups and mid-tier cloud providers without long-term supply agreements face compute rationing or substantially higher hardware costs through the end of 2027, because long-term forward contracts with hyperscalers take priority over spot-market buyers. The $500 billion doesn't just represent a large order — it represents years of forward commitment, potentially locking up a substantial portion of SK Hynix's total HBM production capacity through the end of the decade. For anyone building AI systems outside of these mega-deals, the road ahead just got considerably steeper.