SK Group and NVIDIA Announce $500 Billion-Plus Global AI Infrastructure and Memory Partnership

Imagine the AI tool you use at work — maybe to draft emails, analyse data, or manage appointments. Somewhere behind that tool sits a vast warehouse of computing hardware, consuming enormous amounts of electricity, processing your requests in milliseconds. That infrastructure is about to get a great deal bigger. SK Group and NVIDIA have announced one of the largest AI infrastructure deals ever put to paper, and its effects will ripple through global technology for years to come.

The two companies announced a partnership valued at more than 500 billion US dollars — roughly around £395 billion at current exchange rates — covering AI data centres and the next generation of memory chips that power them. The announcement was made through NVIDIA’s corporate communications and reported by Reuters, though the companies have been clear that the detailed financial breakdown of that figure hasn’t yet been fully disclosed publicly.

What’s Actually Been Agreed

The partnership is built on letters of intent between SK Group — South Korea’s major conglomerate whose affiliates include telecommunications giant SK Telecom and chipmaker SK hynix — and NVIDIA, the US company whose graphics processing units have become the engine of the global AI industry.

SK Telecom plans to build what NVIDIA is calling a Vera Rubin DSX AI Factory in South Korea. It’s a colossal facility, designed to run at 2 gigawatts of power capacity. To put that in context, a single gigawatt is enough to power around 700,000 average UK homes. A 2-gigawatt data centre sits at the very top end of anything currently planned globally. The first facility is targeted to begin operating in 2027.

The factory will be powered by NVIDIA’s Vera Rubin processors — the company’s latest generation of data centre AI chips, designed for training and running large-scale AI models — alongside SK hynix’s HBM4 high-bandwidth memory. High-bandwidth memory, or HBM, is a specialist type of chip that sits close to the processor and moves data at exceptional speed. It’s what makes running enormous AI models practical rather than painfully slow.

A Partnership That Already Existed — Now Made Formal

SK hynix isn’t new to NVIDIA. The South Korean chipmaker is already one of the leading suppliers of HBM used in NVIDIA’s current AI GPUs, the H100 and H200 chips that data centres around the world have been scrambling to get hold of. This new agreement formalises and extends that relationship into future generations of memory technology.

Under the terms announced, NVIDIA and SK hynix will jointly develop high-bandwidth memory for AI training, AI agents, and what the industry calls physical AI — systems that operate in the real world, such as robotics and autonomous vehicles.

Jensen Huang, NVIDIA’s chief executive, has described the partnership as a response to the surging global demand for AI compute, framing it as a long-term strategic collaboration rather than a one-off procurement deal.

The Bigger Picture

South Korea has been positioning itself as a global hub for AI and semiconductor manufacturing, and this initiative fits squarely into that national strategy. NVIDIA has also been expanding its footprint in the country separately, with plans involving partners such as Naver and Brookfield to develop additional AI data centres there.

But not everyone is simply celebrating the scale of the announcement. Some commentators have raised questions about the environmental cost of running multi-gigawatt data centres — facilities that consume extraordinary amounts of electricity and require significant cooling infrastructure. There are also questions about the concentration of AI computing power in the hands of a small number of very large technology companies.

The “$500-billion-plus” figure itself has attracted scrutiny. Critics and analysts have noted that it isn’t yet clear how much of that total represents committed capital expenditure, how much is projected operational spending over many years, and how much remains subject to further negotiation. The letters of intent signal serious intent, but they’re not yet fully binding contracts on every component.

That said, industry analysts broadly see the announcement as confirmation of the extraordinary scale of investment now flowing into AI infrastructure — and of NVIDIA’s continued central role in supplying the hardware that makes it possible.

Energy, Environment, and the Questions Being Asked

A 2-gigawatt AI factory is a striking thing to contemplate. Data centres of this scale require dedicated power infrastructure, significant land, and complex cooling systems. As AI workloads grow — driven by generative AI, large language models, and increasingly capable AI agents — the energy demands of the global computing industry are rising sharply.

Those concerns aren’t unique to South Korea. The UK government and energy regulators have been grappling with how to plan electricity grids around data centre growth. It’s a debate that’s very much alive here in Britain too.

What This Means for Kent Residents

Kent residents won’t see any of this infrastructure built locally — the AI factory and memory facilities are in South Korea. But the cloud-based AI tools used by local businesses, NHS Kent and Medway, and universities in the county are ultimately powered by exactly this kind of hardware, so improvements in NVIDIA and SK hynix’s next-generation chips could make those services faster and more capable over time. The scale of this deal also feeds into UK-wide policy discussions about data centre energy use and digital infrastructure investment, debates that will shape how Britain — and the South East — plans its own digital future.