Why a 28x Rise in Korean Enterprise ChatGPT Use Matters Globally

The number Western AI reports never catch early

OpenAI Korea announced on September 9 that enterprise ChatGPT users in South Korea grew roughly 28 times between August 2024 and August 2025. That figure — cited by Kim Kyung-hoon, head of OpenAI Korea — is not a press-release flourish. It is a leading indicator that most English-language AI coverage is structured to miss until months later.

The dominant Western narrative around enterprise AI adoption moves slowly. Analyst firms publish quarterly surveys. Tech media report on flagship deals. Most American and European companies appear in those stories only after their pilots mature into production contracts. By the time the data lands, the adoption curve in parts of Asia has already flattened into routine operations.

Korea is now showing a velocity signal. And the shape of that signal matters more than the raw multiplier.

What the 28x actually measures

Kim’s briefing at the Joseon Palace Hotel in Gangnam described a company actively widening the scope of what ChatGPT does inside organizations, not just increasing the number of logins. Weekly active users of ChatGPT Workspace and Codex — two products that go beyond chat interfaces — grew approximately 14 times over a six-month period from March. Workspace applies agent-like workflows to research, analysis, and document production. Codex is an autonomous coding agent. Both require a different kind of organizational fit than a freemium consumer product.

Usage is expanding across functional lines — legal, sales, finance, marketing, operations — and moving deeper into actual work outputs: drafting analyses, writing software, automating processes. OpenAI Korea pointed to Samsung Electronics and Seoul National University as cases where AI is no longer a departmental proof of concept but a shared infrastructure layer.

That transition — from isolated pilot to organizational default — is the part most Western coverage treats as background. In Korea it is the main event.

The missing East Asian variable in global AI strategy debates

There is a persistent framing question in AI policy circles: is East Asia falling behind the United States? The answer depends almost entirely on which data you look at. If you measure frontier model training, capital raised by generative-AI startups, or raw compute procurement, the United States remains far ahead. But if you measure how quickly firms integrate usable AI tools into operational workflows, the picture changes. Korea’s 28x enterprise growth, paired with a 35 percent year-over-year rise in domestic weekly active consumer users, suggests adoption velocity that Western enterprise benchmarks have not yet matched.

Japan is running a different experiment. Instead of racing to adopt global models at scale, Japanese firms have increasingly favored domestic model development and localization. That path offers data-sovereignty comfort and language tailoring, but it trades off the speed and capability head start that comes from deploying the most capable frontier models first. Korea is choosing the latter — deploying widely while building guardrails.

The risk in the Korean approach is dependency. The risk in the Japanese approach is fragmentation. Neither is obviously superior over a multiyear horizon. What is clear is that both are happening faster than most international reports track.

The infrastructure play behind the growth figures

OpenAI Korea used its first-anniversary briefing to outline a full-spectrum expansion, not just more licenses sold. The company is preparing domestic data-center capacity to support inference residency — meaning model computation for Korean customers would run inside Korea rather than overseas. Data residency for stored content already exists for ChatGPT Enterprise, ChatGPT Edu, and API customers. Inference residency is the harder ask and the more strategic one.

Security is being addressed through DaveLake, a vulnerability-detection service, with OpenAI Korea hiring a dedicated partner and selecting at least five local partners to help enterprises find, verify, and respond to AI-related security gaps. Distribution is being widened through platforms like Amazon Bedrock so firms can access OpenAI models without abandoning their existing cloud environments.

These moves target the two objections that most slow enterprise procurement: where data lives and who controls the security stack. Solving them is how a product transitions from convenient side tool to mission-critical infrastructure.

Who wins, who loses, and what happens next

The immediate winners are Korean enterprises that moved early and can now point to production results, along with the local system-integration and cloud-operations partners OpenAI is recruiting into its partner network. The secondary winners are the global model makers — OpenAI included — that gain a deepening geographic footprint before competitors can replicate the local partnerships.

The losers in this dynamic are firms that bet on domestic models as a sufficient substitute for frontier capability. Japan’s ecosystem will feel the pressure most directly, because its strategy prioritizes localization over speed. Korean buyers who can deploy the best available tooling tomorrow face less incentive to wait for homegrown alternatives that may take years to converge on equivalent performance.

For Western readers, the practical takeaway is simple: assume that enterprise AI adoption in Korea and much of East Asia is further along than global coverage suggests. The 28x figure is a snapshot of a trajectory, not a stationary milestone. OpenAI’s data-center and inference-residency plans indicate the company expects that trajectory to persist. If Korean firms continue integrating AI across legal, sales, finance, marketing, and operations at this pace, the next measurable outcome will not be more adoption headlines — it will be competitive output. And that is where the global implications become real.