A new front is opening in app store regulation, and this time the finger is pointed squarely at Google. The Korea Fair Trade Commission has formally accused the Mountain View giant of abusing its dominant position in the Android market, signaling its intention to adopt corrective measures and a financial penalty. Figures leaked by the authority point to conduct that would have affected revenues of 14.16 trillion won, or roughly $9.1 billion.

The case revolves around the so-called “Project Hug,” an initiative that the South Korean antitrust body claims was used to retain top developers—especially in gaming—by offering financial incentives in exchange for exclusive presence on the Play Store. A strategy that, if confirmed, would represent a textbook example of market power abuse aimed at locking down the ecosystem and suffocating alternatives.

Project Hug: not just gaming, but a dependency model

What makes the South Korean case a watershed is not the dollar amount itself, but the underlying mechanism. Offering money to prevent an app from being published on third-party stores or distributed via sideloading means leveraging dominance to erect invisible barriers against competition. For software developers, and especially those building applications that embed AI models, control over the distribution channel turns into a tight constraint: every update, every feature based on local inference or proprietary APIs must pass through rules imposed by the store owner.

At a time when AI is increasingly moving to devices—with LLMs optimized to run locally thanks to quantization and frameworks like llama.cpp—the ability to freely distribute software packages becomes a competitive factor. The South Korean case shows how lock-in dynamics are not limited to cloud contracts but can extend straight into the devices we carry in our pockets.

The stakes for the on-device AI ecosystem

For those working with large language models, the Seoul verdict has implications that go well beyond a simple fine. On-device AI is one of the most promising arenas for combining performance, privacy, and control. Running a model locally eliminates dependency on remote servers, slashes network latency, and, not least, ensures user data never leaves the device. But if access to end users is mediated by a gatekeeper that can decide which inference runtimes are allowed, or which models can be downloaded and updated, then technological sovereignty crumbles.

This is not a theoretical concern. Several companies are evaluating on-premise architectures for inference, even in enterprise settings where model patching and software audit are regulatory requirements. If the only channel to reach employee mobile devices is a marketplace governed by a single actor, control over the software supply chain weakens.

Beyond the store: sovereignty and open architectures

The South Korean case fits into a broader movement pushing for transparency in digital ecosystems. The European Union’s Digital Markets Act has already introduced openness obligations for gatekeepers, but the distribution knot remains sensitive. For those evaluating self-hosted AI deployment—on bare metal servers or edge devices—the availability of open channels is a prerequisite for maintaining operational independence.

In this sense, the dispute ignited by Seoul draws attention back to the need to preserve the freedom to install software without forced mediation. A principle that applies not just to games but also to professional AI tools, often distributed as containerized packages or executables optimized for specific hardware configurations, starting from available VRAM.

AI-RADAR analysis: implications for on-premise choosers

From our observatory focused on on-premise deployments, the case takes on familiar contours. Every time an infrastructure becomes dependent on a single supplier—be it a cloud provider or a mobile store—the risk of being trapped by top-down choices rises. The South Korean affair demonstrates that even the app market, seemingly distant from data center logics, can become a weak link in the digital sovereignty chain.

For those migrating AI workloads to owned hardware, the lesson is clear: distribution freedom is an integral part of the control strategy. And when a project like Project Hug threatens to turn an open marketplace into an exclusive enclosure, the casualties are not just game developers, but anyone betting on an independent architecture for their artificial intelligence models.