Meta became one of Microsoft's biggest AI customers, Bloomberg reports

Meta has become one of Microsoft's biggest AI customers, The Decoder reported, citing Bloomberg. The company spends hundreds of millions of dollars a year accessing AI models through Microsoft's cloud service Azure, and Meta burns through trillions of tokens on the platform every week.

Meta's engineers use OpenAI models available through Microsoft's Foundry marketplace, among other tools, to benchmark the performance of Meta's own models. Azure AI Foundry works as a marketplace layer on top of Azure. Developers and companies use it to access first-party and third-party AI models as a hosted service rather than running their own infrastructure for every model they want to test or deploy. Benchmarking against a rival's models before shipping your own is standard practice across the AI industry, so Meta running comparisons against OpenAI's models through Foundry fits a pattern most major labs already follow.

At the same time, Meta is building its own API service that could end up competing directly with Foundry. The Decoder frames this as a familiar pattern: Meta once relied on Microsoft's Bing search engine, then eventually replaced it with its own search technology. A customer buying heavily into a platform while building the tool that could replace it is common in cloud and platform businesses, where switching costs and integration depth often outlast any single competitive threat.

• Meta's AI spending on Azure: hundreds of millions of dollars per year
• Token usage: trillions of tokens burned weekly on the platform
• Top Foundry customer: ByteDance, followed by Adobe, Perplexity, and Sierra
• Share of Microsoft's AI revenue from OpenAI alone: about 70 percent
• OpenAI's compute purchases: mostly direct through Azure, not via Foundry

Most of Microsoft's biggest AI customers are other tech companies rather than smaller startups or enterprises outside the industry. ByteDance tops the list of Foundry customers, followed by Adobe, Perplexity, and Sierra. Foundry itself accounts for only a slice of Microsoft's total AI business, though. Roughly 70 percent of Microsoft's AI revenue comes from OpenAI alone, which mainly buys massive amounts of compute directly through Azure rather than routing purchases through the Foundry marketplace.

The stakes go beyond a single vendor relationship. Meta has poured billions into its own Llama models and data-center buildout, yet still finds it cheaper or faster in places to rent capacity and access rival models through Azure rather than build every piece of infrastructure itself. That calculus can shift company by company and quarter by quarter, depending on how fast in-house chips and models catch up to what a cloud provider already has running at scale. Other AI labs face the same tradeoff: building proprietary infrastructure takes years and capital that even the largest tech companies weigh carefully against renting access to whichever model performs best today.

The arrangement highlights how Microsoft's cloud business profits from AI regardless of which company's models ultimately win adoption. Azure sells the compute and the marketplace access either way, whether a customer ends up favoring OpenAI's models, building its own like Meta does with Llama, or running both in parallel to compare results.

Nothing here should be taken as financial advice — just information to consider.