Meta is quietly one of Microsoft Azure's biggest AI customers

Meta reportedly spends hundreds of millions of dollars each year renting AI models from rival labs, including OpenAI, through Microsoft Azure and Foundry. The reported use is substantial, with Meta consuming trillions of tokens a week to benchmark those models against its own.

That matters because Meta is widely seen as one of the strongest advocates for open models. Yet this report suggests it is also willing to pay heavily for access to closed, competing models when it needs an external benchmark. That is a practical reminder for any business owner weighing an open versus closed AI strategy. The question isn’t which camp you belong to. It’s whether the model you’re using performs best for the job, at a cost and level of control you can live with.

For operators, benchmarking should become a normal discipline, not a one-off project. Test the same customer support query, sales research task, document workflow, or reporting prompt across the models you’re considering. Track quality, speed, cost, error rates, and how much human checking remains. The cheapest model can become expensive if staff spend hours correcting its work.

This is also a sign that even the biggest AI builders don’t rely on one model or one provider. They compare, switch, and keep options open. This is the kind of thing we build into an AI command centre, giving teams a practical way to monitor model performance and decide where each tool fits.