A figure of $500 billion has just shifted the global battle for artificial intelligence into a new dimension. Nvidia and the South Korean conglomerate SK Group have unveiled an initiative worth over $500 billion, encompassing giant data centers, next-generation memory, and long-term industrial cooperation. Behind the announcement, a clear ambition emerges: to secure the components that will power the models, agents, and robots of tomorrow.
The project combines three complementary powers. Nvidia brings its accelerators and computing platform. SK Hynix supplies the high-bandwidth memory that has become essential for AI systems. SK Telecom aims to build the infrastructure capable of hosting this power. Together, they position South Korea at the center of a value chain dominated by the United States in software and processors, but whose industrial heart remains in Asia.
A 2-Gigawatt AI ‘Factory’ by 2027
The most spectacular aspect is a data center with a projected capacity of 2 gigawatts. At this scale, it is no longer just a building filled with servers, but an infrastructure comparable to a vast energy and industrial complex. The first facility is expected to come online in 2027, according to Nvidia and reports from Reuters.
This âAI factoryâ will utilize Nvidia’s upcoming Vera Rubin architecture and SK Hynix’s HBM4 memory. Nvidia’s DSX platform is designed to integrate accelerated computing, networking, systems, and software to produce greater AI capacity while reducing the cost per token. This terminology reveals a paradigm shift: AI is now conceived as a measurable industrial production, complete with factories, energy, critical materials, and supply contracts.
SK Telecom does not hide an even broader ambition. The company has presented a roadmap that could reach 15 gigawatts of capacity in the long term, aiming to transform Korea into a major Asian hub for AI infrastructure. The 2-gigawatt project with Nvidia thus constitutes a significant piece of a much larger national plan.
HBM4 Memory Becomes the Strategic Gold of the Digital Age
Nvidia processors take center stage, but they cannot operate alone. The most powerful models must move vast amounts of data at very high speeds. This is the role of HBM memory, stacked as close as possible to the accelerators. Without a sufficient quantity of this memory, even the best chips lose some of their appeal.
The agreement thus establishes a long-term relationship between Nvidia and SK Hynix to secure supply, as well as to develop and optimize future generations together. The playing field extends beyond training large language models. The partners cite autonomous agents, physical AI, robotics, and enterprise services, all of which are likely to multiply the demand for computing.
This alliance gives Nvidia greater visibility over a critical resource and provides SK Hynix with guaranteed demand over several years. It may also tighten the market for competitors. If the largest players reserve advanced memory capacities in advance, smaller companies may face higher costs, longer wait times, or reduced ambitions.
$950 Billion in Agreements Surrounding South Korea
The Nvidia-SK announcement does not occur in isolation. Reuters reports that South Korea has presented nearly $950 billion in new initiatives involving Samsung Electronics, SK Group, and several American tech groups. SK is said to have signed approximately $750 billion in agreements, including the over $500 billion initiative with Nvidia. Samsung, for its part, has announced a protocol with Broadcom that could cover up to $200 billion in memory, foundry, and advanced packaging.
These amounts encompass projects, partnerships, and commitments spread over time: they should not be confused with an amount immediately spent. Part of it relies on letters of intent and future objectives. However, their magnitude indicates the direction the industry is taking. Tech groups are seeking to simultaneously lock in chips, memory, electricity, data centers, and partners capable of assembling everything.
South Korean President Lee Jae-myung took advantage of a summit in San Francisco to advocate for new cooperation with major American companies. Leaders from Nvidia, OpenAI, Anthropic, and Broadcom participated alongside key Korean industrialists. South Korea no longer wants to merely sell components: it aims to become a global platform where artificial intelligence is built and operated.
The Real Issue: Who Will Control Computing Power?
Competition in AI is no longer limited to which model achieves the best score. It revolves around the ability to operate millions of agents, train increasingly complex systems, and connect AI to physical machines. This capability depends on a very concrete chain: semiconductors, memory, networks, cooling, land, and electrical production.
With Vera Rubin, Nvidia is preparing the generation that will succeed current platforms. With HBM4, SK Hynix aims to remain one of the most difficult suppliers to replace. With its data centers, SK Telecom wants to become the operator capable of transforming these components into accessible services for businesses. The pact thus combines three levels of the same power: designing, powering, and operating.
The risk is commensurate with the ambition. Building a capacity of 2 gigawatts requires gigantic investments, stable power supply, and sustainable demand. The companies themselves acknowledge that financing costs, construction timelines, public regulations, and equipment availability can alter the schedule. The agreement remains an industrial promise to be fulfilled, not an automatic guarantee of profitability.
France and Europe Face a Power Gap
For France and Europe, the signal is stark. The continent has laboratories, startups, mathematical expertise, and champions like Mistral AI. It also has significant players in semiconductor equipment and energy. However, it remains dependent on foreign suppliers for the most sought-after accelerators and for a large part of advanced memory.
One single Asian project of 2 gigawatts illustrates the scale at which the next phase is being prepared. Europe can respond with computing centers, low-carbon electricity, faster financing, and industrial partnerships. It will also need to decide where it wants to be sovereign and where it accepts interdependence. Attempting to replicate everything would be costly; failing to secure anything would be dangerous.
The question also concerns French companies. If computing capacity and memory are reserved by a few giant alliances, access to infrastructure could become a decisive competitive advantage. Groups capable of signing contracts early or developing more efficient models will be better positioned than those relying solely on a future drop in prices.
An Announcement That Changes the Measure of the Possible
The Nvidia-SK Group alliance is not only significant for its staggering amount. It outlines the industrial model of AI for the coming years: very long-term contracts, electrical capacities comparable to those of major national infrastructures, and close integration between chip designers, memory manufacturers, and data center operators.
South Korea thus sends a message to the entire world. It does not want to be a mere workshop for components in a revolution directed elsewhere. It wants to control a decisive share of the power that makes this revolution possible. If the partners execute their plan, 2027 could mark the moment when AI will no longer be perceived as merely a software industry but will become a strategic infrastructure on par with energy, telecommunications, or transportation.
Reliable Sources
- Reuters via Investing.com â Nvidia-SK initiative of over $500 billion, July 24, 2026
- Reuters via Investing.com â $950 billion in AI agreements announced around South Korea, July 25, 2026
- Nvidia Korea â official details on the 2 GW AI factory, Vera Rubin and HBM4, July 25, 2026
- SK Telecom Newsroom â roadmap to 15 GW of AI capacity in South Korea
