Meta Platforms has quietly become one of Microsoft's biggest AI customers, according to a Bloomberg report published on August 20, 2026, spending hundreds of millions of dollars a year to access artificial intelligence models through Microsoft's Azure cloud. People familiar with the matter told Bloomberg that Meta consumes trillions of AI tokens each week via Microsoft's platform, underscoring how the current AI economy remains overwhelmingly concentrated within a handful of large technology companies.
Foundry: Microsoft's model marketplace
The vehicle for Meta's spending is Microsoft Foundry, an Azure marketplace that resells models from a wide range of providers and had roughly 100,000 customers as of July. While Microsoft's marketing highlights industrial and enterprise adopters, insiders say that most of Foundry's largest revenue contributors are other technology companies. ByteDance, the Chinese owner of TikTok, has historically been the biggest spender on Foundry, with Adobe, AI search company Perplexity and OpenAI-linked customer-service startup Sierra also among the top accounts.
Meta's dual-track AI strategy
Meta develops its own frontier Llama-family models and is investing heavily in first-party AI data centers, yet CTO Andrew Bosworth acknowledged on the Big Technology podcast in July that the company also rents leading external models based on availability and cost. Meta developers have used OpenAI technology through Foundry to help evaluate the output of their own models, sources told Bloomberg — a workflow that turns Microsoft into both a supplier and an indirect benchmark provider.
The disclosure lands as investors scrutinise so-called circular AI dealings, in which the same handful of tech giants buy compute and models from one another. Microsoft's most recent fiscal year showed that OpenAI alone contributed roughly 70% of the company's overall AI revenue, and Meta's rise on Foundry only deepens that concentration.
Meta wants to sell models too
Even as it buys, Meta is building an API service to sell access to a mix of internal and external AI models, a direct competitor to Foundry that could eventually reduce the company's reliance on Microsoft. Meta has form for this playbook: it used Microsoft's Bing to power Facebook web search in the late 2000s before replacing it in-house by late 2014, and it used Microsoft compute for early AI research before spinning up its own hyperscale data-center footprint. Combined with memory and foundry mega-deals like Samsung-Broadcom and Xiaomi's push into in-house 3nm AI silicon, the pattern is clear: the hyperscalers are both each other's biggest customers and each other's most direct competitors.
Reporting based on coverage from Bloomberg via Yahoo Finance.