Six-year-old Nvidia A100 AI accelerator chips still carry an estimated residual value of $4,956, according to Silicon Data, complicating Michael Burry’s warning that Big Tech may be overstating profits by depreciating AI hardware over five or six years. The estimate is based on future rental income, utilization and operating costs rather than a second-hand sale price. A100 GPUs can continue serving inference (running trained AI models), fine-tuning and lower-cost workloads after newer chips arrive. CoreWeave has signed a contract covering A100 GPUs through 2029, nearly a decade after the architecture debuted. Burry estimated that hyperscalers (the largest cloud providers) could understate depreciation by roughly $176 billion between 2026 and 2028 by extending equipment lives, while Meta said a 5.5-year schedule for certain servers and network assets would reduce 2025 depreciation by about $2.9 billion. Nvidia’s AI infrastructure expansion, including partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR targeting more than $500 billion, has made chip utilization, cash flow and residual value increasingly important. Kalshi’s year-end market assigns A100 compute a 74% chance of ending above $1.39 per GPU-hour, versus roughly $1.29 currently on Silicon Data’s A100 index. Kalshi Research said the A100 forward curve remained broadly flat into 2027, while rental prices for four-year-old H100 chips have recently risen sharply.