Jensen and Musk both think aging GPUs are worth more than the books show

The Jensen-Musk convergence on GPU asset life is not an accounting footnote—it is a valuation paradigm break. Reclassifying GPUs from three-year consumables to five-to-seven-year capital assets rewrites the competitive physics of the entire stack. Upstream, this raises the bar for advanced packaging. CoWoS yield is no longer "good enough"; it becomes the single variable governing multi-year reliability. TSMC in Taiwan, China gains pricing leverage: the less frequently customers refresh, the more punishing any long-tail defect becomes. On the software side, CUDA's moat deepens mechanically. Every additional year of hardware life dilutes migration cost, compressing the window for rivals. On the regulatory front, the risk calculus shifts. BIS export controls are calibrated around technology-diffusion windows. Longer GPU lifespans extend the effective radiation period of any single authorization, making policymakers more likely to tighten rather than relax. For AMD and Intel, the picture worsens. Lower refresh cadence kills the next-gen performance narrative. Lock-in becomes Nvidia's most durable moat. Over the next 18 months, expect three shifts: a secondary GPU market matures into its own ecosystem; accounting standards re-examine AI hardware depreciation cycles; and "GPU as infrastructure" fully displaces "GPU as commodity." Nvidia's valuation anchor is migrating from cyclical hardware to platform infrastructure.