Google has placed restrictions on Meta's use of its enterprise-grade Gemini AI models, according to a Financial Times report cited by Bloomberg on 28 June 2026. The move limits how Meta — one of the largest AI developers in the world — can access Google's flagship foundation model offering, drawing a sharper line between two companies that are simultaneously partners, cloud customers, and direct competitors in the generative AI market.

The specifics of the restriction have not been fully disclosed publicly. What is known is that Google's enterprise Gemini model is deployed across multiple corporate customers, and that Meta was among those users. Google appears to have placed a ceiling on that access, though the precise mechanism — whether rate limits, capability tiers, contractual clauses, or an outright cutoff for particular use cases — has not been confirmed through official statements from either company as of this writing.

Why This Matters

Meta's own AI posture makes the restriction notable. The company is one of the most active developers and distributors of open foundation models: its Llama 3.1 405B, released in July 2024, is a large openly available model that competes directly with proprietary offerings from Google, OpenAI, and Anthropic. Meta simultaneously builds its own frontier models and, evidently, evaluates or uses commercial model APIs from competitors — a dual-track approach common in large organisations running internal AI programmes alongside procurement of external capabilities.

That dual-track reality is what makes the Google decision structurally interesting. If a company of Meta's size and technical depth is drawing on Gemini for any portion of its internal or product work, it suggests either that Gemini offers specific capabilities or infrastructure integrations Meta finds difficult to replicate internally, or that procurement speed and risk distribution are driving the decision rather than raw model performance. Both explanations are plausible. Neither has been confirmed.

Google's position is equally layered. Gemini is a commercial product, and enterprise access is governed by terms of service and business agreements that can legitimately restrict competitive use. The core question is whether Meta's usage crossed a threshold Google considers damaging to its competitive position — for instance, using Gemini outputs to benchmark, fine-tune, or otherwise inform Meta's own model development. That kind of usage, sometimes called "model distillation" or competitive intelligence gathering, is a recognised concern among frontier model providers, several of which have added explicit prohibitions to their terms of service over the past two years.

Worth flagging: the verified facts here are thin. The restriction is reported via a single Financial Times account, relayed through Bloomberg, with no on-record confirmation from Google or Meta. The scope, duration, and precise rationale remain opaque. Readers should treat this as a reported development requiring further confirmation rather than a fully documented policy change.

The Competitive Terrain

The broader landscape this sits within is one where the boundaries between cloud customer, API consumer, and direct competitor have become genuinely difficult to manage. Google Cloud counts numerous AI companies — including some that directly compete with Google's own AI products — among its customers. Microsoft Azure has a similar dynamic with OpenAI's commercial deployments. The enterprise AI stack has created a tier of providers whose infrastructure is too valuable to refuse, even when the customer at the table is also a rival.

Meta, specifically, occupies an unusual position. Its open-model strategy through the Llama series is explicitly designed to commoditise the foundation model layer — a strategic pressure that benefits every company trying to reduce dependence on proprietary APIs, and one that cuts directly against the commercial model that Google, OpenAI, and Anthropic are building. For Google to limit Meta's Gemini access, if confirmed and if commercially motivated, would be a direct acknowledgment that this tension has reached the point of operational consequence.

Whether this is a one-off contractual dispute or the leading edge of a broader policy shift — frontier model providers systematically restricting access for direct competitors — will likely become clearer as more reporting surfaces. Either way, the question of who gets to sit at the foundation-model table, and on what terms, is now an active business and potentially regulatory issue, not a theoretical one.