Nvidia and SK Group announced a $500 billion-plus strategic partnership on July 24-25, 2026, covering a 2-gigawatt AI factory in South Korea and a long-term memory co-development pact with SK Hynix that locks in supply at a scale no prior deal has matched.
The announcement came out of an AI Summit in San Francisco, and the numbers are not easy to contextualise at first. Five data centers of one gigawatt or more are expected to come online globally across all of 2026. SK Telecom is now committed to building a single 2-gigawatt facility. That's not incremental. It is a sovereign bet of the kind that, until this week, had been made only by hyperscalers with trillion-dollar market caps and the US federal government's blessing.
SK Telecom will construct the AI factory in South Korea on Nvidia's DSX platform, powered by Nvidia's Vera Rubin accelerated computing architecture and fed by SK Hynix HBM4 memory. The first facility targets operational status in 2027. Separately, Nvidia and SK Hynix have signed a long-term co-development agreement for next-generation AI memory solutions, a deal that formalises what the supply chain already knew: SK Hynix holds roughly 70% of Nvidia's HBM4 allocation for Vera Rubin, with Samsung and Micron sharing the remainder, according to industry analysis from KuCoin and TrendForce.
Memory is the bottleneck nobody talks about loudly enough. The processing side of AI infrastructure gets the headlines, but HBM supply has been the real constraint on how fast AI clusters scale. Nvidia's Vera Rubin platform entered full production in June 2026, with customer shipments scheduled for Q3. Locking SK Hynix into a multiyear co-development pact does two things at once: it secures Nvidia's own supply continuity through the most aggressive buildout period in the industry's history, and it makes it structurally harder for AMD, Intel, or a custom silicon rival to get the same memory at the same volume and cadence.
AMD has its own HBM relationships. Custom silicon shops at Google, Amazon, and Microsoft do too. But none of them just signed a deal at this scale with the company that controls 60-70% of leading-edge HBM capacity. The SK Hynix pact is not only a supply agreement. It is a moat, built in Korea, announced the same week Nvidia already signed a separate $1 billion-plus Naver investment in South Korean AI infrastructure. The timing is deliberate.
Frankly, the framing around "strategic partnership" undersells what is happening. This is Nvidia planting a flag in Northeast Asian AI infrastructure before anyone else can. South Korea produces the majority of the world's advanced memory. Nvidia just formalised a first-mover relationship with its dominant supplier and signed the country's largest telecom operator to build the biggest single AI compute facility outside the United States.
Why 2 gigawatts is the number to hold onto
To put 2GW in context: most large enterprise data centers operate at 50-100 megawatts. The AI-optimised hyperscale campuses that have been making headlines in 2026 run 100-750 MW per site. Gartner forecasts that AI servers will consume more power than all conventional data center hardware combined by 2027, and total data center consumption is projected to double from roughly 485 TWh in 2025 to 950 TWh in 2030. That is a city-scale power draw. A 2-gigawatt facility, running continuously, draws the equivalent of a mid-sized city's entire grid load.
SK Telecom is not a data center company. It is South Korea's largest telco. It is now building infrastructure that dwarfs anything its domestic competitors have attempted. The Naver announcement earlier this week - a $1 billion AI investment that drew considerable attention on its own - looks like a rounding error next to this. South Korea is positioning itself not as a chip supplier to the AI economy but as a compute hub in its own right, with sovereign infrastructure capable of serving Asia-Pacific demand at scale.
The letters of intent signed by both sides cover the AI factory construction and the memory supply framework. They are not yet final contracts, a distinction worth keeping in mind. But letters of intent at this scale, announced jointly by Nvidia's Jensen Huang and SK Group leadership at a major industry summit, do not quietly disappear. The machinery is already moving.
What gets built at 2GW by 2027, and what it runs, will tell you more about the actual trajectory of AI infrastructure in Asia than any forecast published this year.
Also read: Moody's warns that AI spending is turning the world's most profitable companies into capital-intensive ones • Texas is pulling the $1.3 billion annual tax break that built its AI data center empire • X removes 42,000 AI reply bot accounts and the warning to growth marketers is impossible to ignore
