SpaceX’s quarterly numbers put a new shape on the AI economy: compute doesn’t just live in data centers, it ships. In its third-quarter report, the company posted its first operating profit as a public company, while AI-related revenue in the “Super Intelligence” (AI-compute leasing) segment climbed to $7.56 billion—nearly tripling from the prior quarter and slightly outpacing Starlink. Even the fine print matters. Management pointed to higher spending and additional debt tied to both AI compute and connectivity buildout, and Starlink added 1.6 million consumer subscribers, reinforcing the same theme: capacity is a supply-chain problem.
That financial momentum collided with another ecosystem battle today. Google unveiled TorchTPU, a specialized PyTorch variant aimed at easing developers’ move to Google’s TPUs instead of Nvidia GPUs. The pitch is practical: PyTorch is used for 85%+ of AI applications, and Google is pairing TorchTPU with migration tooling (including a TPU agent suite) plus LiteRT updates so TPU deployments better support PyTorch-based models.
Meanwhile, corporate structure headlines hovered over the background. Musk said Tesla and SpaceX boards are evaluating a potential $3.5 trillion merger, but he’s barred from the talks as a conflicted party—no guaranteed deal, and no merger vote proposed. For executives tracking AI investment, the message is clear: the companies winning near-term scale are the ones turning hardware scarcity, developer friction, and capital intensity into a single plan.