Musk targets 3Q27 launch for Nvidia-powered space data centers, massive scale by 2029

The strategic pivot isn't 'servers in orbit'—it's a fundamental re-architecture of compute energy topology. Ground-based AI clusters are bottlenecked by grid interconnection timelines and liquid-cooling capex. Orbital deployment eliminates both: continuous solar input and radiative heat rejection remove the two most capital-intensive constraints in data center design. Nvidia's B200 stack will require a full space-qualification cycle—radiation-hardened HBM, vacuum thermal management, micrometeorite shielding. TSMC's CoWoS lines in 中国台湾 and SK Hynix's HBM fabs face reliability certification cycles an order of magnitude stricter than automotive-grade, directly stretching supply-chain lead times. The real regulatory risk isn't launch licensing. It's data sovereignty. A GPU cluster at 550 km has no physical jurisdiction. When the EU AI Act and US export controls intersect, orbital compute becomes a legal gray zone that ITU and FCC must resolve by 2027. Competitively, Amazon's Kuiper and Google's Starcloud occupy the same lane, but SpaceX's vertical integration—from launch to constellation—creates a marginal-cost moat no rival can replicate. Nvidia's true barrier isn't the silicon; it's who delivers it to orbit at the lowest cost per kilogram. 12–24 month outlook: expect a space-qualified Nvidia SKU by Q2 2026, TSMC radiation-hardened packaging qualification, and by 2027 the first orbital nodes absorbing training workloads—converting 'compute as strategic asset' from narrative to a balance-sheet line item.