Nvidia and SK Group signed letters of intent this week valuing their strategic relationship at more than $500 billion, a figure that bundles a multi-year memory supply deal with SK hynix, a planned 2-gigawatt AI data center from SK Telecom, and unspecified future expansions into a single headline number. The agreement commits real hardware purchases, Nvidia GPUs, networking, systems, and SK hynix HBM4 memory, but the half-trillion-dollar valuation represents an aggregate of expected commercial activity over several years, not a signed contract or a capital commitment.
The centerpiece is SK Telecom’s 2-gigawatt AI data center in South Korea, slated to enter service in 2027 and built on Nvidia’s DSX AI factory platform with Vera Rubin accelerated computing systems. The facility is intended to serve sovereign AI, enterprise AI, physical AI, and agentic AI workloads across South Korea and the Asia-Pacific region. Nvidia describes DSX as a complete blueprint combining accelerated compute, networking, software, and partner technologies into a data-center-scale platform optimized for low-cost token generation and energy efficiency.
Breaking down the $500 billion requires assumptions the announcement does not support. The figure rolls together SK Telecom’s hardware procurement, SK hynix memory supplies to Nvidia, revenue for Nvidia’s ecosystem partners building the DSX factories, OEMs, ODMs, networking, storage, cooling, power, and vague “potential future expansion.” The 2-gigawatt deployment alone implies thousands of NVL72 VR200 racks and hundreds of thousands of Vera CPUs and Rubin GPUs, but the exact count is unknowable: NVL72 VR200 racks draw 166 to 240 kilowatts each, and the facility’s power usage effectiveness is undisclosed.
The memory agreement with SK hynix, unveiled in June, underpins the hardware side. HBM4 supply is critical for the Vera Rubin architecture, and locking in long-term volume gives Nvidia visibility while giving SK hynix a demand anchor. But the partnership’s value is only as durable as the build-out schedule and the willingness of end customers to fill the capacity with paying workloads.
Skepticism is warranted. Letters of intent carry no enforcement mechanism, and the 2027 target looks aggressive against a backdrop of delayed hyperscale projects, Microsoft’s announced data centers from several years ago remain incomplete. The $500 billion figure serves a signaling function: it tells investors and competitors that Nvidia’s platform pull extends from silicon to memory to full-stack infrastructure, and that SK Group is betting its telecom and memory franchises on that ecosystem.
What matters next is whether SK Telecom secures the power, permitting, and supply chain slots to hit 2027, and whether the memory agreement translates into firm purchase orders with pricing and volume schedules. Until capital expenditure turns into deployed racks and revenue-bearing tokens, the half-trillion-dollar number is a narrative device, useful for market perception, indifferent to execution risk.