Meta is building enormous AI data centers and its own models. But it still needs Microsoft’s machines to keep the work moving.
According to Bloomberg, Meta is now one of Microsoft’s largest AI customers, spending hundreds of millions of dollars a year through Azure and consuming trillions of tokens each week. The company is using Microsoft’s AI platform to access outside models and computing capacity, illustrating how even one of the industry’s biggest AI investors cannot rely entirely on its own infrastructure.
And that dependency is becoming a business opportunity. The more companies spend on building and deploying AI, the more they need the computing, cloud services, and data-center capacity underlying those systems — shifting a growing share of the industry’s spending toward infrastructure providers.
Meta is not simply renting raw computing power from Azure. Part of the spending goes through Microsoft Foundry, Microsoft’s AI development platform, which provides developers with access to models from multiple providers.
Foundry essentially puts different AI models behind a common platform, allowing companies to select models rather than building their entire AI workflow around a single provider. Bloomberg reports that Foundry now offers several AI models to 100,000 companies.
That helps explain why Meta would use Microsoft’s platform even as it develops its own models. Bloomberg’s sources say Meta engineers have used OpenAI models through Foundry to assess the output of Meta’s own AI systems.
For Microsoft, that is the more interesting part of the deal. Foundry allows Azure to sit beneath competing AI companies rather than forcing Microsoft to win the model race itself. The same Foundry business has also been cited as the reason the company remains in China, where it serves Chinese companies like ByteDance.
Meta has used this playbook before
Long before Azure became an AI powerhouse, Facebook used Microsoft’s Bing technology to fill gaps in its own search capabilities, but dropped it in 2014. Meta later acquired Bing AI for its Meta AI platform and, according to Reuters, began developing an in-house alternative.
That history provides useful context for today’s Azure spending. Meta does not necessarily need to abandon its AI ownership plans. The company is using Microsoft’s technology while investing toward greater control of its own AI stack over time.
Why AI infrastructure may be the steadier bet
For cloud providers, infrastructure vendors, MSPs, and other channel companies, Meta’s Azure spending reinforces a potentially lucrative feature of the AI market: customers do not have to pick one AI ecosystem and stick with it.
An enterprise may build proprietary models, use OpenAI for certain workloads, rent Microsoft Azure capacity, and buy infrastructure from still other vendors. Every additional layer can create another opportunity for providers supplying cloud capacity, networking, storage, integration, security, and managed services.
The risk is that enterprise AI demand will not grow enough to justify the enormous infrastructure expansion now underway. Much of today’s AI infrastructure investment assumes companies outside the technology industry will eventually consume substantially more AI computing.
Until then, the industry will retain an unusual dynamic: some of AI’s fiercest competitors will also remain one another’s largest customers.
For the channel, that may be the more important race to watch. Model leaders can change quickly, but every contender still needs somewhere to run the models.
Related reading: For another sign of how quickly the AI infrastructure race is escalating, OpenAI has signed a 20-year lease for an 8GW Ohio data center campus with Nvidia supplying the computing infrastructure and financial backing.