AI Infrastructure
Nvidia and SK Group Sign $500 Billion AI Infrastructure Partnership
Newaiera Desk · 2026-07-28 · 4 min read
SK Telecom will build a 2-gigawatt AI factory using Nvidia's DSX platform and Vera Rubin accelerated computing, powered by SK hynix HBM4 memory. The July 2
The AI infrastructure buildout just got another zero added to it. On July 25, Nvidia and SK Group announced a partnership valued at more than $500 billion spanning AI factory construction and AI memory supply — one of the largest infrastructure commitments of the year, in a year that has had no shortage of them.
What's actually being built
The centerpiece is a 2-gigawatt AI factory, to be built by SK Telecom, running on Nvidia's DSX platform and Vera Rubin accelerated computing. The facility will be powered by SK hynix HBM4 — next-generation high-bandwidth memory — with the first site targeted to come online in 2027.
The two companies signed letters of intent covering both halves of the deal: building the physical AI factories, and securing the memory supply chain that feeds them. SK hynix separately entered a long-term AI memory partnership with Nvidia to lock in supply of next-generation memory, with the companies committing to codevelop and optimize future HBM generations together.
Why memory is the other half of this story
It's easy to read AI infrastructure deals as being purely about GPUs, but this one makes the memory dependency explicit. High-bandwidth memory has been one of the tightest bottlenecks in AI accelerator supply chains — a top-end GPU is only as fast as the memory feeding it — and SK hynix is one of the two or three companies in the world that can produce HBM at the volume frontier labs and cloud providers need.
By tying the AI factory build directly to a codevelopment agreement on HBM4, Nvidia is securing supply for its own roadmap while SK hynix gets a committed, large-scale customer for a product category it has invested billions to scale. That's a different shape of deal than a simple GPU purchase order — it's a supply chain being built jointly rather than bought off a shelf.
The regional angle
The partnership is explicitly framed around Asia-Pacific demand, with South Korea as the anchor market. SK Telecom gets to invest in and control AI infrastructure capacity domestically, while making that capacity available to its own customers — effectively becoming a regional AI infrastructure operator built on Nvidia's stack, rather than a customer routing everything through US or other overseas cloud capacity.
That matters beyond South Korea. As AI infrastructure spend keeps climbing, national and regional operators building their own Nvidia-powered capacity — rather than depending entirely on the handful of US hyperscalers — is becoming a repeatable pattern, not a one-off.
The bigger picture
At this point, half-trillion-dollar AI infrastructure commitments have become frequent enough that the number itself barely functions as a headline anymore — what matters is the specifics. Here, the specifics are: a real facility, a real timeline (2027), a named next-generation memory chip (HBM4), and a supply chain partner rather than just a customer. That combination is what separates an infrastructure announcement with substance from one that's mostly a valuation signal.