SK Group and NVIDIA announced on July 25 that they have signed letters of intent on a partnership worth more than 500 billion dollars, covering both the construction of AI factories and the supply of next generation AI memory. It is one of the largest infrastructure commitments the industry has produced, and it is aimed squarely at the compute shortage that has defined this summer.
The centerpiece is a facility. SK Telecom will build a 2 gigawatt NVIDIA Vera Rubin DSX AI factory to serve global compute demand, running NVIDIA's Vera Rubin accelerated computing platform on SK hynix HBM4 memory, with the first factory scheduled to come online in 2027. Two gigawatts is roughly twice the scale of the domestic chip data center Z.ai switched on in China this week, and it is being built for customers worldwide rather than for one lab.
The memory half of the deal is the part worth reading twice. NVIDIA and SK hynix have entered a long term partnership to secure and co-develop next generation AI memory, including high bandwidth memory. HBM has been the genuine bottleneck in AI hardware for two years, harder to expand than logic and dominated by a small number of suppliers. By tying up supply and joint development with one of the two companies that can make it at scale, NVIDIA is defending the input its own accelerators cannot ship without.
The geography is deliberate too. Both companies frame the partnership as accelerating large scale AI infrastructure across the Asia Pacific region, including sovereign, physical, agentic and enterprise AI services, with South Korea as the anchor. NVIDIA's chief executive has been calling this the golden age of Korea and pitching a national K-AI vision spanning chips, frontier models and infrastructure, which lands in a country that already supplies much of the world's advanced memory.
The caution is in the fine print. These are letters of intent rather than executed contracts, and the 500 billion dollar figure is a multi year headline spanning construction, equipment and memory supply, not money changing hands. Even discounted, it belongs to a pattern that has become the defining fact of the year, with Alphabet lifting its own 2026 spending to as much as 205 billion dollars days ago and Microsoft committing to AMD racks at scale. The capital being committed to AI infrastructure has stopped being measured against past technology cycles and started being measured against national budgets.