As South Korea builds its data centers, commercial cloud computing is a necessary stopgap to fuel its AI boom.

South Korea wants to become an artificial intelligence powerhouse alongside the United States. But Seoul’s cloud rules could make that harder than it needs to be. In September 2025, President Lee Jae-myung launched the National AI Strategy Committee to make Korea a top AI power alongside the United States. In June, he backed that ambition with a 10-year, $880 billion plan for semiconductors and AI data centers. South Korea has also set a goal of achieving 95 percent AI adoption across the public sector by 2030.

Those ambitions require more than GPUs. They require the ability to put those chips to work securely, which means giving government agencies access to scalable cloud infrastructure.

Approval to provide it has been slow, though. Providers handling government data must clear a review process requiring physical separation—dedicated hardware for government data rather than software isolation on shared servers. If maintained, that requirement would exclude providers unwilling to duplicate infrastructure they already operate in South Korea, slow public-sector cloud and AI adoption, and increase service costs.

Every government has the right and responsibility to secure its public-sector data, and Seoul takes this obligation seriously. The United States does the same, certifying cloud providers for federal use through FedRAMP. The question is whether security rules should be based on where infrastructure sits or how effectively it protects data.

To train public-sector AI models, South Korea needs two things: compute and government data. For compute, Seoul has tripled its AI budget to nearly $7 billion for 2026, secured some 260,000 GPUs across public and private buyers, and backed projects like Solasido, a data center targeting roughly 15,000 GPUs by 2028. That capacity will come, but construction and chip supply take time. Commercial cloud can bridge the gap, putting compute to work now rather than waiting for new facilities to come online.

The bigger obstacle is access to government data. South Korean public-sector systems are graded high, medium, or low; separation applies to the top two, which cover anything holding personal information. AWS, Microsoft, and Google already operate cloud regions inside South Korea. They do not run dedicated infrastructure reserved for government data, because building that for a market South Korea’s size is difficult to justify. The providers are already there. Rules keep public institutions from using them. As of the end of 2025, no foreign cloud provider had obtained South Korea’s mid-tier or higher CSAP certification.

Physical localization can also work against resilience when it limits geographic redundancy. South Korea learned the cost of concentrated infrastructure last year. A fire at the National Information Resources Service disrupted 647 government systems and destroyed the government’s G-Drive, including 858 terabytes of data used by roughly 125,000 officials. The G-Drive had no backup. That is precisely the kind of vulnerability a resilient cloud architecture should prevent.

The good news is that Seoul and Washington have already agreed on the direction of travel. The US-Korea Technology Prosperity Deal calls for removing barriers to innovative, trusted, and privacy-preserving data-hosting architectures and reducing operational burdens on technology companies. A subsequent Trump-Lee agreement stated Seoul’s commitment to ensure American companies do not face discriminatory or unnecessary barriers in digital services. 

Now the two governments need to turn those commitments into practice. Seoul should judge providers on security rather than location, with measures such as encryption keys the Korean government holds and providers cannot use, audited access, and independent verification. These safeguards address the concerns behind localization—including concerns about the CLOUD Act—while avoiding the need to duplicate expensive infrastructure.

Washington’s task is the mirror image. FedRAMP kept most providers out of the federal market for a decade, and June’s FedRAMP 20x rules replaced it with continuous, automated certification that lets providers reuse security work they have already done rather than rebuilding it in a government format. That same principle should extend to trusted allies. Mutual recognition of rigorous cloud-security standards, measured against a common benchmark such as the Krach Institute’s Global Trusted Tech Standard, would turn trust into a process rather than a promise.

US-Korean AI cooperation is already physical. SK hynix is investing nearly $4 billion in an advanced AI-memory packaging and R&D facility in West Lafayette, Indiana. In comparison, Amazon has committed at least $5 billion to AI and cloud infrastructure in Korea through 2031. Capital is crossing the Pacific in both directions because companies on both sides believe in the alliance.

The governments have made the same bet. Now they need the architecture to support it. Mutual recognition of rigorous cloud-security standards would let Korea protect its sovereignty without isolating its AI ecosystem and let the United States demonstrate that strong allies deserve more than assurances of trust. They deserve rules that reflect it. 

About the Authors: Paul Stahle

Paul Stahle is the Korea Trusted Tech & Trade fellow at the Krach Institute for Tech Diplomacy at Purdue, focusing on US-Korea trade, competition, and emerging technology policy. A 19-year Foreign Service veteran, he advanced US economic, energy, and technology interests across Washington, China, Europe, and the Indo-Pacific, with expertise spanning Korea, China, India, critical and emerging technologies, energy security, and critical minerals.

The post To Lead in AI, the US and South Korea Need to Trust Each Other’s Clouds appeared first on The National Interest.