Nvidia and SK Have Defined the Technology in Their $500 Billion AI Plan, Not the Price

That leaves an important comparison unresolved: an initiative valuation spanning potential equipment, supply and services cannot be treated as equivalent to a cash investment, capital-expenditure budget or booked revenue without those definitions. The reported announcement likewise identifies the factory and memory partnership, rather than a spending schedule.

The figure also sits inside a wider set of Korean-US technology discussions. Kim Yong-beom, South Korea's presidential chief of staff for policy, said participants had agreed to pursue $950 billion of cooperation projects, including a proposed $750 billion of high-performance-semiconductor supply by SK Group to global technology companies, including Nvidia. He also cited joint plans for AI data centers totaling 5 gigawatts and 2 million GPUs. The briefing does not say how those totals overlap with the Nvidia-SK initiative, so the $500 billion, $750 billion and $950 billion figures should not be added together or treated as competing like-for-like commitments.

Two projects make the partnership concrete

The factory is the clearest testable milestone. The companies say SK Telecom will use the DSX architecture and Vera Rubin systems powered by HBM4, with the first AI factory planned to come online in 2027. “Up to 2 gigawatts” describes the intended maximum capacity, not capacity already installed; progress will therefore be visible in power, construction and equipment deployment rather than in the partnership's aggregate value.

The second project binds Nvidia's system roadmap to SK hynix's next-generation AI-memory development and supply. That is strategically relevant because the parties are pairing a factory plan with a component partnership, but the announced scope is still a plan: the release names the technologies and intended workloads—from large-language-model training to agentic and physical AI—without commercial terms.

This expands an existing network of deals

The announcement is not the start of the relationship. It builds on a decades-long technology partnership, and the same account of the June agreement says SK hynix had already entered a multiyear technology arrangement to co-develop memory for Nvidia products while SK Telecom agreed to build a DSX-based AI cloud aimed at gigawatt-scale factories. Huang did not establish in that account whether the $500 billion is incremental to that framework or a new valuation placed on it.

Nor is SK's infrastructure strategy limited to Nvidia. SK hynix is pursuing a long-term memory-supply collaboration with Microsoft; SK Telecom has signed a memorandum with Anthropic on a Korean gigawatt-scale data-center project; and SK Telecom and AWS plan to expand their infrastructure cooperation from an Ulsan AI data-center project. The SK account describes all three as planned collaborations, not completed capacity. Together they show that the Nvidia arrangement is part of a multi-partner buildout, not evidence of an exclusive infrastructure stack.

What would make the value auditable

The next meaningful disclosures are mundane but decisive: a time horizon for the $500 billion, a definition of what it counts, each party's committed capital and the factory's power, site, equipment and customer arrangements. Memory volumes, duration and pricing would show whether the supply partnership is principally technical cooperation or a commercial procurement commitment.

Until then, the evidence supports a substantial technical and strategic expansion: Nvidia systems, SK Telecom infrastructure and SK hynix memory are tied to a 2027 first-facility target. It does not support treating the $500-billion figure as a single, fully financed project or a directly comparable investment total.