The first batch of production-ready Vera Rubin has been deployed in Microsoft Azure data centers, with Microsoft getting the launch allocation—but behind this move is a bigger chess game by Nvidia: it’s turning synthetic hyperscale cloud into reality. Last week, Brad Gerstner described Nvidia with the term “synthetic hyperscaler”—a cloud giant that doesn’t own data centers but controls compute capacity. Nvidia has been following a clear path: it lends money to OpenAI to build an AI supercomputing center in Ohio, co-builds an AI data center with Cloverleaf, and today it even directly supplies cards to Microsoft. Nvidia’s quarterly revenue hit $81.6 billion, up 85% year over year. And its data center networking business is growing faster than the GPU business itself, indicating that when customers expand capacity, they’re effectively forced to pay a “tax” through InfiniBand. On Microsoft’s side, Azure just crossed $100 billion in annualized revenue, and Copilot’s paid seats number 30 million—but its own 300,000-unit order for Maia 300 at TSMC has become a kind of shackle that ends up ceding production capacity to Nvidia. Microsoft wants to build its own chips, but Vera Rubin’s initial release still goes to it, which suggests Azure customers’ demand is simply not waiting for Maia mass production. Google’s TPU 8 still doesn’t have clear external customers, while Nvidia has already locked in the deployment rhythm of top cloud providers through production-level delivery. This timing gap will directly translate into next year’s market share. Next, we’ll watch whether Nvidia’s pricing for renting out Ohio compute capacity to external customers ends up lower than Azure’s quotes for the same spec—if it does, the “synthetic hyperscale cloud” narrative will be confirmed. Keep an eye, too, on whether Microsoft’s next earnings report shows Azure AI gross margin and Maia orders being pushed back—that would be the most direct evidence of a shift in supply-chain bargaining power.