What This Is
The 15% figure is a stone Nvidia threw at the AI industry this week. The GPU giant formally notified customers of across-the-board price hikes exceeding 15% on AI products, covering H100, H200, and other flagship chips. The reasons aren't complicated: tight HBM (high-bandwidth memory dedicated to AI chips) capacity, supply-demand imbalance in advanced packaging (the process of stacking multiple chips together), and demand-side growth still running wild.
Industry View
The optimistic view: compute price hikes conversely prove AI demand is real, not a bubble. Nvidia's pricing power comes from supply-demand imbalance, indicating downstream enterprises are actually paying up — capital markets can rest easy.
We are more concerned about the risk side:
- Cost pressure will spike sharply for SMEs and independent developers; cloud providers (AWS, Azure, Alibaba Cloud) will likely pass costs to customers;
- The window opens further for domestic AI chips — Huawei, Cambricon, Hygon — shifting the substitution discussion from "whether" to "when";
- For Chinese AI companies relying on overseas compute, supply chain risk has shifted from theoretical projection to a certain event; renewal negotiations next year will be tougher.
Impact on Regular People
For enterprise IT: This year's AI-related budgets need 20–30% buffer space; don't lock in at last year's prices.
For individual careers: AI tool subscription fees and internal enterprise AI service call costs will likely creep up in H2; AI project managers need to give advance warning.
For consumer markets: Short-term, this won't show directly in phone or PC prices, but every "AI-enhanced" product — from smart speakers to video subscriptions — carries a compute bill that will eventually be passed to consumers.