Meta spends hundreds of millions on Azure, top Microsoft AI customer

Meta Platforms has become one of Microsoft's biggest AI customers, spending hundreds of millions of dollars annually to access AI models through Azure, according to a person familiar with the matter.

Demand for AI remains concentrated within the technology industry itself, Bloomberg reported, even as the sector pours record capital into data centers and model development.

Meta has used trillions of tokens through Microsoft's Azure platform, buying access to models across multiple providers based on availability and cost, according to people familiar with the matter. Microsoft's Foundry marketplace, which offers models from a variety of vendors, counted 100,000 customers as of July, including ByteDance, Adobe, Perplexity and Sierra.

OpenAI accounted for 70 percent of Microsoft's overall AI revenue in its most recent fiscal year, a concentration that shows the risk if AI adoption fails to spread beyond the tech sector. Meta's spending adds a second major revenue stream for Azure's AI business while the social media company simultaneously builds its own model distribution channel.

Meta's emergence as a major Azure customer is notable because the company has invested heavily in its own AI infrastructure. The Facebook and Instagram parent operates one of the largest compute fleets in the industry, yet still relies on external cloud services for certain model workloads, according to people familiar with the matter. The spending covers access to frontier models that Meta does not build itself, including those from OpenAI and other vendors distributed through Azure.

The relationship is not one-sided. Meta is planning to sell access to its own AI models through an API service that could compete directly with Microsoft Foundry, potentially reducing its reliance on external providers. The company recently announced plans to introduce a range of new AI models designed to run directly on laptops and personal devices, a move that could shift some inference workloads off centralized cloud infrastructure and onto consumer hardware.

Open-Weight Strategy Puts Meta in Competition With Its Suppliers

Unlike OpenAI and Anthropic, Meta has released several of its models as open-weight systems, allowing developers to download, modify and run them independently rather than accessing them exclusively through Meta's services. Chinese companies including Alibaba, DeepSeek and Moonshot have pursued the same approach, and their models have gained traction among developers seeking alternatives to US providers.

Mark Zuckerberg has argued that open-weight models could give these companies an advantage, particularly as US AI developers face greater restrictions in areas such as the use of training data. The strategy could help Meta capture developer mindshare even as it pays Microsoft for cloud capacity, creating a dynamic where the company is simultaneously a customer, competitor and potential supplier to the same market.

What the Spending Means for Azure's Growth

For Microsoft, Meta's spending provides evidence that its multi-vendor marketplace strategy is working. Rather than relying solely on OpenAI's models, Azure has established itself as a neutral distribution layer for AI, attracting customers who want flexibility in model selection. The 100,000-customer Foundry base and the addition of Meta as a major spender suggest the approach is gaining traction beyond the initial wave of AI startups.

The concentration risk remains, however. OpenAI's 70 percent share of Microsoft's AI revenue means the relationship with its most important model provider is also its biggest vulnerability. If OpenAI were to shift more of its compute to other providers or develop its own distribution channels, Microsoft would need to replace a substantial portion of its AI revenue.

For investors, the relationship highlights the circular nature of AI spending. Microsoft shares rose 0.56 percent while Meta gained 0.43 percent on the day the Bloomberg report was published, suggesting the market had already priced in the arrangement. The bigger question is whether AI demand will broaden beyond the handful of technology giants that currently account for most of the spending, or whether the sector's infrastructure buildout will continue to feed itself.

This article is for informational purposes only and does not constitute investment advice.