The same FT account said Meta told staff to use AI tokens more efficiently and that several other Google clients were affected to a lesser degree. Neither Google nor Meta has publicly confirmed the cap. The visible pressure beneath it is less disputed: both companies have told investors that AI infrastructure is now a central capital-spending priority, and Google Cloud presents its AI stack as a platform for model building, deployment and enterprise use.
The distinction between model access and cloud compute matters. A cap on Gemini usage is a limit on a specific model service; a cap on cloud capacity would reach deeper into the data-centre stack. The FT report should therefore not be inflated into a claim that Google cut Meta off from all AI infrastructure. The business point is narrower and more revealing: when capacity is tight, the supplier can decide which form of access expands and which slows.
For Google, rationing access would be commercially awkward but not irrational. Alphabet's investor materials frame AI investment as both a product opportunity and an infrastructure commitment, with cloud demand tied to specialised chips, data-centre capacity and model services. If the same infrastructure serves Google products, Google Cloud customers and outside developers, capacity allocation becomes a business decision rather than an engineering footnote.
For Meta, the reported constraint cuts the other way. Meta's investor materials emphasise heavy AI infrastructure spending and model development, but the company still operates in an ecosystem where cloud, chips and frontier-model access are concentrated among a small number of suppliers and rivals. A dependence on a competitor's model stack, even at the margin, is a strategic weakness when access can be narrowed.
Meta is not a small buyer bargaining at the edge of the market. It has its own models, its own data-centre programme and the balance sheet to buy large amounts of compute. That is why the reported cap is useful evidence of the market's shape. If a company with Meta's resources still hits a supplier constraint, smaller AI firms have less room to treat cloud and model access as an elastic input.