Meta is spending hundreds of millions of dollars a year buying artificial-intelligence access from Microsoft, even as it commits extraordinary sums to building competing models, chips and data centers of its own.
The relationship makes Meta one of Microsoft’s largest customers for Azure AI Foundry, the cloud marketplace through which companies can access models from OpenAI and other developers. Meta consumes trillions of tokens through the service each week, according to a person familiar with the arrangement. Neither company has confirmed the figures.
A token is the small unit into which an AI system divides words, numbers and code before processing them. One trillion tokens can represent hundreds of billions of words. Meta’s reported weekly usage therefore points to industrial-scale use rather than employees occasionally asking a chatbot questions.
Meta developers use outside models for software development and to evaluate the output of the company’s own AI systems. Chief Technology Officer Andrew Bosworth has previously acknowledged that Meta rents leading models from outside providers when availability, cost or performance makes doing so useful.
The arrangement reveals how tangled the AI business has become. Meta competes with Microsoft for engineers, advertising customers and leadership in artificial intelligence. Yet it also pays Microsoft to access models and computing capacity that help it develop competing products.
For Microsoft, the revenue is real. The larger question is where the money ultimately originates. Microsoft says Foundry has reached 100,000 customers, but many of its largest users remain technology companies, including Meta, ByteDance, Adobe, Perplexity and customer-service AI company Sierra.
OpenAI alone generated $24.1 billion in commercial revenue for Microsoft during the fiscal year ended in June. Bloomberg estimated that this represented roughly 70% of Microsoft’s total AI-related sales.
That concentration matters because technology companies are simultaneously investing in one another, purchasing one another’s computing capacity and using one another’s models. A dollar can move from an AI developer to a cloud provider, then to a chipmaker or data-center operator, producing revenue at several companies before a customer outside the technology industry has paid for a finished service.
The arrangement does not mean the demand is artificial. Meta’s willingness to spend heavily on outside models suggests that AI computing remains constrained enough that even one of the world’s largest data-center builders cannot supply everything internally. Renting also allows Meta to compare competing models without waiting for its own infrastructure to be completed.
But it does complicate the investment case. The industry still must prove that factories, hospitals, retailers, banks and ordinary consumers will eventually generate enough economic value to support the hundreds of billions of dollars now circulating among technology companies.
Meta may eventually replace much of its Microsoft usage with its own models and an internal model marketplace, just as it previously used Microsoft’s Bing search technology before developing alternatives. For now, one of Microsoft’s biggest AI customers is also one of the companies working hardest to need Microsoft less.
JBizNews Desk | Redmond
© JBizNews.com All Rights Reserved. Reproduction or distribution without written permission is prohibited.
