When a US federal court preliminarily approves a $7.85 million antitrust settlement, the tech world takes notice. The Sony case isn’t just about PlayStation consoles: it’s a wake-up call on the concentration of power in digital storefronts — and a reminder for anyone who relies on a closed ecosystem, including those who build and deploy applications powered by Large Language Models.

The Sony case at a glance

The class action, known as Caccuri v. Sony Interactive Entertainment, alleges that the company monopolized the digital PlayStation games market. At the heart of the complaint: since 2019, Sony stopped allowing third-party retailers to sell download codes for digital games, forcing users to go exclusively through the PlayStation Store. The result, according to the plaintiffs, was full control over pricing and supply, harming millions of consumers. The preliminary settlement, approved on April 8, covers around 4.4 million console owners in the United States.

The lock-in trap

Sony’s move is far from unique. The “walled garden” approach runs deep: Apple with its App Store, Epic Games’ legal battle, Amazon’s Kindle ecosystem. In every case, the platform owner enforces a single distribution and payment channel, extracts high commissions, and stifles competition. For developers and users, it means dependency on one vendor, fewer alternatives, and rising costs.

What about AI?

The same dynamics of control and lock-in are now surging into the AI space. The big cloud platforms — AWS, Azure, Google Cloud — offer “managed” LLM services that simplify inference and fine-tuning, but simultaneously tether data and decision-making to a proprietary ecosystem. Switching providers becomes complex and expensive, much like having a digital game library locked to a single store. Organizations evaluating on-premise LLM deployment often do so precisely to retain data sovereignty, ensure GDPR compliance, and negotiate from a position of strength rather than accepting the terms of a de facto monopolist.

It’s no accident that frameworks like vLLM, TGI, and Ollama are gaining momentum: they let teams serve models on their own hardware, keeping full control of the pipeline. With the right hardware — GPUs with adequate VRAM, quantization techniques to reduce footprint — even mid-size organizations can consider a self-hosted infrastructure without depending on a single cloud provider. For those weighing such options, there are non-trivial trade-offs around TCO, in-house skills, and operational overhead, but the strategic independence is often the deciding factor.

Looking ahead

The Sony settlement adds to a growing list of regulatory moves — from Europe’s Digital Markets Act to the FTC’s actions in the US — aimed at curbing dominant platforms. For the AI world, the lesson is clear: entrusting the full lifecycle of an LLM to a single cloud vendor exposes businesses to the same risks PlayStation gamers have faced. Those investing in on-premise or hybrid architectures today aren’t just after performance: they want independence, cost predictability, and assurance that their data won’t become yet another asset trapped in someone else’s garden.

This small settlement won’t shake the cloud giants, but it’s a piece of a shifting landscape. Just as with digital games, the sovereignty battle for AI will be fought on the antitrust front and on the ability to choose where and how to run one’s models.