Jensen Huang Just Delivered Incredible News for Nvidia Stock Investors - Yahoo Finance

The core insight here is not simply a projected doubling of chip shipments, but the structural shift in AI economics that Vera Rubin represents. A 97% reduction in inference token costs is a paradigm-... Read original →

The 97% inference cost collapse in Vera Rubin is the story that matters—not the shipment doubling. It re-prices AI from a capital experiment into a utility, and the demand curve bends from linear to exponential. Technology cascade: The Vera CPU–Rubin GPU co-design dissolves the traditional CPU-GPU boundary. NVLink 6.0 and Spectrum-X networking must evolve in lockstep, and liquid cooling shifts from optional to mandatory. A 50x performance jump in four years means per-chip power crossing 1,000W, forcing a rethink of data-center electrical architecture. Supply-chain exposure: Leading-edge fabrication remains tethered to Taiwan, China capacity. Export-control tightening has converted geopolitical risk from a tail event into a standing variable in every P&L. Competitive dynamics: AMD's MI400 and hyperscaler custom silicon (TPU, Trainium) form a pincer. Yet CUDA's developer lock-in remains the moat—silicon is no longer the differentiator; the software stack is. 12–24 month outlook: Inference commoditizes first; training migrates to the next architecture. Power and thermal constraints, not silicon, become the binding constraint on AI capex. The ~$700B FY2028 revenue guide embeds two fragile assumptions—sustained hyperscaler spend and no architectural disruption. Break either, and the valuation anchor resets downward, fast.