Meta has emerged as one of Microsoft's top AI customers, pouring hundreds of millions of dollars annually into Azure. According to Bloomberg reporting cited by The Decoder, Mark Zuckerberg's engineers are torching trillions of tokens every week on the platform.
Rather than building purely in isolation, Meta actively buys OpenAI compute via Microsoft's Foundry marketplace to benchmark and calibrate its open-weight Llama family against closed proprietary leaders. It exposes an amusing paradox in modern enterprise AI: the foremost champion of open-source models is directly financing the compute infrastructure of its fiercest closed-ecosystem competitors.
This dynamic mirrors Meta's historical playbook in search, where it leaned heavily on Microsoft Bing before engineering its own proprietary search stack. As Meta quietly develops an internal API service to eventually rival Foundry, Microsoft reaps the windfall. Foundry's broader roster features heavyweights like ByteDance, Adobe, Perplexity, and Sierra—though the marketplace remains a sliver of Microsoft's AI top line, roughly 70 percent of which comes directly from OpenAI purchasing Azure compute.
Audit your external AI evaluation spend this quarter. Map out exactly how many compute dollars your engineering teams route to direct market rivals for calibration before committing capital to bring those evaluation pipelines entirely in-house.