SpaceX Needs $40 Billion To Buy Chips From Nvidia — Here's Why - BGR

This is not a procurement headline—it is the moment aerospace formally becomes the largest consumer of AI compute. The $40 billion Nvidia order is, in substance, a conversion of orbital infrastructure into a distributed supercomputing substrate. The technical cascade is immediate: this volume of GPU demand will compress CoWoS advanced-packaging capacity at TSMC (Taiwan, China) and squeeze HBM output from SK Hynix and Samsung, forcing wafer allocation to tilt toward hyperscalers and space-sector buyers. On the compliance front, concentrating advanced-silicon consumption in a single buyer creates a single-point-of-failure vulnerability in the US supply chain; any tightening of export controls on fabs in Taiwan, China or mainland China could trigger cascading disruptions. Competitively, AMD and Intel will accelerate MI-series and Gaudi roadmaps, while Amazon's Trainium and Google's TPU teams will reposition around "sovereign compute" narratives to capture non-Nvidia workloads. Over the next 12–24 months, expect custom ASICs to erode Nvidia's share in inference-heavy orbital workloads, and the boundary between "space company" and "AI infrastructure provider" to dissolve entirely. The real story is not who buys chips—it is that the compute bottleneck has migrated from data centers to low-Earth orbit.