Google has limited how much Meta can use its Gemini AI model, according to a Financial Times report cited by Bloomberg on 28 June 2026. The move is notable because Meta — one of the world's largest AI developers — and Google are in a tricky position: they are simultaneously cloud partners, customers of each other's services, and direct competitors in the generative AI market.

The details of the restriction are sparse. Google's Gemini is available to enterprise customers, and Meta was among them. Apparently Google has imposed some kind of ceiling on that access, but whether it comes in the form of usage limits, access to fewer capabilities, contractual language changes, or an outright ban on certain uses remains unconfirmed. Neither company has issued an official statement explaining the move as of this writing.

Why This Matters

Meta's approach to AI makes this notable. The company is aggressively developing and releasing open-source foundation models — think of these as the large, general-purpose AI systems that other companies and developers can build on top of. Meta's Llama 3.1 405B, released in July 2024, is freely available and competes directly with the proprietary models from Google, OpenAI, and Anthropic. At the same time, Meta develops its own advanced models and also buys access to competitors' models through their APIs — a dual approach that large companies with serious AI programs commonly use.

The Google move is structurally interesting precisely because of this split strategy. If a company as large and technically skilled as Meta is using Gemini, it suggests one of two things: either Gemini has capabilities or integrations with Google's other services that Meta finds hard to build itself, or Meta values speed and redundancy over relying solely on its own models. Both explanations are reasonable. Neither has been confirmed.

Google's side of this is equally layered. Gemini is a commercial product, and the company can legitimately use contract terms to restrict how customers use it. The real question is whether Google thinks Meta crossed a line — specifically, whether Meta was using Gemini outputs to test, retrain, or otherwise learn from Google's models to improve its own. This practice, sometimes called "model distillation," is a recognised concern among AI companies. Over the past two years, several of them have explicitly banned it in their service agreements.

The verified facts here are limited. A single source — the Financial Times — reported this, and it was relayed through Bloomberg. Google and Meta have not confirmed anything on the record. The scope of the restriction, how long it lasts, and Google's exact reasoning all remain unclear. Treat this as a reported development that needs more confirmation rather than a fully established policy change.

The Bigger Picture

The real dynamic here is one that's become genuinely tricky to manage. The line between being someone's cloud customer, API user, and direct rival has blurred. Google Cloud hosts numerous AI companies — many of which compete directly against Google's own AI services. Microsoft Azure has a similar setup with OpenAI's commercial offerings. The AI infrastructure layer has become so valuable that companies cannot afford to turn away customers, even when those customers are also their competitors.

Meta's position is especially unusual. Through its open-source Llama models, Meta is deliberately trying to make advanced AI foundation models cheap and widely available — essentially commodifying them. This strategy benefits every company trying to reduce dependence on proprietary APIs, but it directly conflicts with how Google, OpenAI, and Anthropic make money. For Google to restrict Meta's access to Gemini, if this holds up and if it is indeed a competitive move, would be an acknowledgment that this tension has real business consequences.

Whether this turns out to be a one-off dispute or the start of a pattern — where AI providers systematically restrict access for rivals — should become clearer as more information surfaces. For now, the question of who gets to use which AI models, and on what terms, has shifted from a theoretical discussion to a real business and regulatory issue.