AI chip depreciation risks amount to $176 billion. Tech giants often claim a 5-7 year depreciation period for a $40,000 AI chip, which misrepresents the true 24-month timeline. This discrepancy between reported profits and actual expenses arises from this depreciation method. Companies buy billions of dollars in hardware for a 2-3 year product cycle but depreciate it over 5-6 years, artificially inflating earnings. This practice misrepresents asset longevity and profits. AI chips typically last 24 months. Physical failures can occur within 1-3 years due to continuous use. The unrealistic 5-7 year schedule leads to inflated financial expectations. Value falls quickly as new chip versions with better features hit the market. For example, the shift from the A100 to H100 shows how faster performance can make older models obsolete almost overnight. However, obsolete chips still retain some value. Even though they may be outdated, chips like the A100 from 2020 still hold some worth, typically worth 50-70% less than their peak price. With earnings likely overstated and adjustments from financial institutions on the horizon, the AI boom mirrors past overinvestment trends. The reality contradicts the extended depreciation periods, echoing previous economic bubbles. The market is overlooking the rapid obsolescence of AI hardware. This is a significant multi-billion dollar earnings risk hiding in the balance sheets of major tech companies. The 24-month hardware cliff is real, and accounting schedules are dangerously outdated. Recent Q3 2025 earnings from major cloud providers highlighted this spending spree, with total capital expenditures hitting over $60 billion for the quarter. A new report from Gartner estimates that over 75% of this expenditure is on AI servers with a lifespan of less than three years. This massive investment could soon become a liability, contrary to investor beliefs. This scenario suggests considering long-dated put options on major cloud providers like Microsoft, Amazon, and Meta. The market hasn’t fully accounted for the significant earnings writedowns anticipated in 2026 and 2027. Options that expire in late 2026 could provide a direct opportunity to prepare for this accounting adjustment. The catalyst for this market shift is likely the upcoming release of NVIDIA’s next-generation chip architecture, named after another renowned scientist, expected in late 2026. Previously, the launch of the Blackwell B200 in late 2024 made the H100 virtually obsolete for high-level training overnight. The same situation is looming for Blackwell chips, and balance sheets are unprepared for this potential value drop. This scenario echoes events from the telecommunications boom of the late 1990s. Companies invested billions in fiber optic cables, depreciating them over decades, but much became “dark fiber” after the 2000 crash. These asset writedowns were a blow to shareholder returns. A more balanced approach involves a pairs trade: going long on chip designers like NVIDIA while shorting a basket of hyperscalers. This strategy benefits from the “shovel seller” who continues to earn money, while the “miners” face the consequences of depreciating equipment. It shields the accounting risk from wider market fluctuations.

here to set up a live account on VT Markets now