Nvidia's Chip Depreciation Debate: AI Hardware Value and Market Uncertainty

Nvidia CFO's Stance on Chip Lifespan

• 💡 Nvidia's CFO stated that chips shipped six years ago are still fully utilized today, attributing this longevity to their CUDA software.
• ⚠️ This claim, intended to settle the depreciation debate, has instead raised further questions about the true economic value of older AI chips.

Distinguishing Utilization from Value Creation

• 🧠 Michael Bur argues that physical utilization is being confused with economic value creation.
• ✈️ He likens older AI chips to airlines keeping old planes for overflow: they fly and are full, but are only marginally profitable and have low resale value.
• 📉 Jensen Huang himself has indicated that older GPUs become difficult to sell once a new generation, like Blackwell, is released, stating previous models like Hopper are "fine" in only a few circumstances.

Investor Uncertainty and Market Implications

• 📊 The core risk for investors lies in the uncertainty surrounding the economic lifespan of AI chips.
• 📈 This ambiguity makes it difficult to accurately model capital expenditures (capex), profit margins, and the overall AI cycle with confidence.
• ⚠️ If supply imbalances shift, the value of older chips could drop rapidly.

Vertical Integration and Alternatives

• 🏢 Google's strategy is to build its own hardware, exemplified by the release of Nano Banana Pro, powered by their Gemini 3 model on TPUs.
• 📈 This vertical integration strategy is gaining traction with investors, as seen by Google's stock performance.

Innovation and Cost Efficiency in China

• 🇨🇳 Questions arise about how Chinese companies are developing comparable open-source AI models with significantly lower spending on infrastructure compared to US counterparts.
• ❓ This disparity raises further concerns about GPU usage and contributes to broader anxieties about market bubbles.