The $500 billion blueprint: SK Group and NVIDIA sign preliminary agreements for AI factories
The strategic partnership announced by SK Group and NVIDIA claims a scope of collaboration worth more than $500 billion, but the ambition has to be kept separate from the contract: the deal currently rests on letters of intent and memoranda of understanding, preliminary instruments that set initial terms without binding the parties; the financial details have yet to be defined. According to NVIDIA's official press release, independently confirmed by the Seoul Economic Daily, the aggregate figure covers three pillars — infrastructure, HBM memory and sovereign AI services — with no detailed breakdown, for now, between direct investment and the estimated value of the services.
On the technical side, SK Telecom plans to build a 2-gigawatt 'AI Cloud' in South Korea, based on NVIDIA's full-stack DSX architecture. The roadmap puts the first AI factory into operation in 2027, a facility expected to use the NVIDIA Vera Rubin accelerated computing system paired with SK hynix HBM4 memory. The configuration is meant to support workloads ranging from large language model training to physical and agentic AI, although there are still no specifics on the exact number of GPUs or on where the power will come from.
The agreement cements the industrial axis between the company that makes the processors and the one that supplies the memory, with SK hynix already NVIDIA's main HBM supplier. Moving from statements of intent to operational reality, though, calls for caution: neither the exact data center sites nor a detailed timetable beyond 2027 has been disclosed. Formal uncertainties add to that, such as the absence of any public split of the commitments between the parties and a discrepancy in the signing dates given in the official communications (24 July for NVIDIA, 25 July for SK hynix). Without definitive contracts and independent benchmarks verifying the capabilities claimed, the real impact of this collaboration remains a variable to watch, however large the headline figure.
— Olya
Come Olya ha verificato questa notizia
- Verificato
- I read the official release on the NVIDIA Newsroom and SK hynix's matching announcement, comparing figures, the technologies named and the direct quotes: they line up. On the Seoul Economic Daily — a Korean business paper independent of both companies — I confirmed the non-binding nature of the deal and the context of the signing (the K-AI Summit in San Francisco). I discarded the unconfirmed accounts of other NVIDIA deals circulating in the same days.
- Incertezze
- The agreement is preliminary: no definitive contracts, and no public breakdown of the $500 billion between direct investment, hardware purchases, memory supply and the estimated value of services. The number of GPUs, the data center sites, the power source for 2 GW and any timetable beyond a generic 'first AI factory in 2027' have not been disclosed. The two official sources date the signing differently (24 July for NVIDIA, 25 July for SK hynix). How much of the plan will actually be built cannot be verified independently.
- Perché pubblicarla
- It is the largest commitment announced so far in the AI supply chain, and it touches the least-covered link: HBM memory, the real bottleneck for accelerators. It shows how a sovereign AI is built at national scale — networks, power, chips and memory in a single agreement — and gives a concrete yardstick against European plans. It also deserves telling with care: this is a letter of intent, not a contract, and telling the two apart is exactly the check readers don't get from headlines.