Meta now ranks among Microsoft’s largest AI customers, spending hundreds of millions of dollars annually to access models via the Azure cloud platform. The company consumes trillions of tokens on the service every week, primarily using OpenAI models available through the Foundry marketplace to benchmark its own research. This arrangement mirrors Meta’s earlier reliance on Bing before it developed its own search technology. While Meta engineers depend on these external tools for evaluation, the company simultaneously builds an API service that could eventually compete with Foundry. Most of Microsoft’s major AI clients are other tech firms, yet OpenAI alone generates roughly 70 percent of the company’s AI revenue. The remainder comes from diverse buyers like ByteDance, Adobe, Perplexity, and Sierra, with Foundry accounting for only a small fraction of that total business.
The situation highlights a complex dynamic where competitors share infrastructure while preparing to replace it. Meta uses Microsoft’s compute to validate its own models, creating a dependency that may shift as its proprietary tools mature. Microsoft benefits from immediate revenue while betting that its platform remains essential even as clients build alternatives. This pattern suggests that cloud providers will continue to profit from the training and testing phases of AI development.
* Meta spends hundreds of millions annually on Azure for model benchmarking.
* OpenAI generates approximately 70 percent of Microsoft’s AI revenue.
* Meta is developing its own API to potentially replace Foundry services.
