Editorial illustration for Meta Pays Microsoft Hundreds of Millions Annually for AI
Meta Pays Microsoft Hundreds of Millions for AI
Meta burns through trillions of tokens a week on Microsoft's Azure cloud, and according to a Bloomberg report, that habit costs the company hundreds of millions of dollars annually. The arrangement makes Meta one of Microsoft's largest AI customers, tapping into models through Azure's Foundry marketplace to benchmark its own systems against the competition, including OpenAI's tools.
There's an obvious irony here. Meta is simultaneously building its own API service, one that could eventually go head-to-head with the very Foundry platform it currently relies on. Anyone who remembers Meta's search history has seen this movie before: the company leaned on Microsoft's Bing for years before cutting the cord and rolling out its own search tech.
Meta isn't alone in feeding Microsoft's AI business. ByteDance, Adobe, Perplexity, and Sierra all show up among Foundry's top customers. Still, the bulk of Microsoft's AI revenue doesn't come from any of them.
OpenAI alone accounts for roughly 70 percent of it, mostly through raw compute purchases on Azure rather than Foundry access. That leaves Foundry, and customers like Meta, as a smaller piece of a much larger machine.
Why this matters
Meta spending hundreds of millions a year to run OpenAI models on Microsoft's infrastructure, while simultaneously building Llama to compete with both companies, tells us something about where the real AI economy sits right now: on rented compute and borrowed benchmarks, not in any single lab's model weights. For developers and founders, the lesson is practical. Even a company with Meta's balance sheet and its own frontier models still finds it cheaper or faster to lean on Azure's Foundry marketplace for benchmarking than to build every comparison tool in-house.
That's a signal about how much friction remains in evaluating models at scale, trillions of tokens a week is not a side project. For researchers, it's worth watching whether this kind of cross-shopping becomes standard practice, companies buying rivals' models to test their own, because it changes how "competition" gets defined in AI. Microsoft, meanwhile, keeps winning regardless of who tops the leaderboard, since it collects the toll either way.
Watch whether Meta's usage numbers show up in Microsoft's next earnings call.
Related Reading
Browse all Meta news →
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic OpenAI hires 630 ex-Meta staff as ChatGPT memory may turn data into ads
Business & Startups
Related Topic Meta AI Update Pulls From Your Calendar for Daily Briefings
LLMs & Generative AI
From Archives Anthropic's Claude Expands Into Protein Design
From Archives OpenAI Revokes Access to Limited Cyber Program Daybreak Blue
From Archives Meta AI’s New Mac App Can See Your Screen and Answer Questions
Business & Startups
From Archives Meta Launches Dedicated Mac App for Its AI Chatbot
AI Tools & Apps
Common Questions Answered
How much does Meta spend annually on Microsoft's Azure cloud services for AI?
According to Bloomberg, Meta spends hundreds of millions of dollars annually on Microsoft's Azure cloud services. This makes Meta one of Microsoft's largest AI customers, burning through trillions of tokens per week to access AI models through Azure's Foundry marketplace.
Why does Meta use Microsoft Azure's AI models if it has its own Llama system?
Meta uses Azure to access and benchmark various AI models, including OpenAI's tools, to compare them against its own systems and stay competitive in the AI landscape. Even with Meta's substantial resources and its own frontier models, the company finds it more cost-effective or faster to leverage Azure's infrastructure rather than relying solely on internal capabilities.
What is the irony in Meta's relationship with Microsoft regarding AI development?
Meta is simultaneously one of Microsoft's largest AI customers while building its own API service that could eventually compete with both Microsoft and OpenAI. This demonstrates that Meta is investing heavily in external AI infrastructure through Azure while developing competing technologies in-house.
What does Meta's spending on Azure reveal about the current AI economy?
Meta's hundreds of millions in annual spending on Microsoft's cloud services shows that the real AI economy is built on rented compute and borrowed benchmarks rather than being concentrated in individual companies' proprietary model weights. This indicates that even well-resourced companies with their own frontier models still depend on external cloud infrastructure and third-party AI services.
Further Reading
- Meta Has Quietly Become One of Microsoft's Largest AI Customers - Bloomberg
- Meta emerges as major Microsoft Azure AI customer - Yahoo Finance
- Meta becomes one of Microsoft's largest AI customers | Report - Seeking Alpha
- Meta Quietly Becomes One of Microsoft's Biggest AI Customers - TradingView / GuruFocus
- Meta and Microsoft report ballooning AI expenses - Yahoo Finance
