Meta emerges as one of Microsoft's largest Azure AI customers, releases Muse Glimmer 30B

Bloomberg reported that Meta has quietly become one of Microsoft's largest Azure AI customers, consuming trillions of tokens weekly through Azure Foundry and spending hundreds of millions of dollars annually—much of it on OpenAI models—to benchmark and evaluate its own AI systems. The disclosure is striking: a frontier lab paying a rival's cloud to run a competitor's models, underscoring how evaluation and comparison have become a massive line item.

Logan Kilpatrick of Google AI Studio poked at exactly this dynamic on X, asking 'What % of global AI token spend do you think is people running evals?'—a question Meta's spending answers pointedly.

On the product side, Meta released Muse Glimmer, a 30B open-weight model designed for local multi-step agentic tasks on consumer GPUs, extending its open-weight strategy (Llama was rebranded to Muse earlier in 2026). At 30B it competes directly with Alibaba's Qwen 3.8-27B and NVIDIA's Nemotron 3.5 Lightning in the edge/local-deployment tier.

Financially, Meta raised 2026 capex guidance to $130–$145 billion, with Q2 ad revenue up 27% to $59.4B funding the AI buildout. Meta also shipped a dedicated Mac desktop app for Meta AI with thinking-mode switching, file attachments, and recurring briefings. The tension worth watching: Meta champions open weights publicly while spending heavily on closed rivals' models privately—a pragmatism that reveals where real capability gaps still sit.