Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
The AI Industry’s Burning Question: How Long Before a GPU Depreciates?
Published November 14, 2025 – Updated November 14, 2025
The AI boom moves fast. Big tech firms invest large sums. They build new AI centers. A key money question sits on many minds: When does a GPU lose value? Spreading the cost over time helps show profit growth in the fast AI hardware field.
Industry Leaders Take Slightly Different Stances on GPU Lifespan
Top tech firms like Google, Oracle, and Microsoft say their AI gear may last up to six years. Microsoft’s report shows computer parts losing value in two to six years. This gap shows doubt in a field where AI parts stay new. Nvidia began making data center GPUs for AI in 2018. Its yearly data center money grew from $15 billion to $115 billion. The field grows fast while the history is short. The AI surge grew after ChatGPT started in late 2022. Today, money experts and lawyers have little past work to guide them.
The Financial Implications of GPU Depreciation
If GPUs last longer, firms can spread out depreciation over many years. This plan cuts yearly costs and lifts profit numbers. If GPUs have short lives, each year brings high costs that cut earnings. Haim Zaltzman from Latham & Watkins asks, "Is it three, five, or seven years?" His point shows that lenders must judge how safe a GPU purchase is.
Evidence from GPU Leasing and Usage Markets
Some GPU renters feel good. CoreWeave rents out GPUs and uses a six-year schedule since 2023. CEO Michael Intrator notes that Nvidia’s 2020 A100 chips still fill up fast. When free, the 2022 H100 chips rent again at 95% of their first price. This result hints that these parts keep worth. Still, CoreWeave’s stock dropped 57% since June. This dip reflects worry over high AI spending and slow data center projects. Oracle’s stock fell 34% from early record highs. This fall shows that investors fear lower profit and tight money use in AI.
Skeptics Highlight Risks of Overestimating Useful Life
Some voices, like short-seller Michael Burry, worry that top firms such as Meta, Oracle, Microsoft, Google, and Amazon may count on GPUs to last too long. Burry says that server GPUs might work well only for two to three years. This short span could shake current profit plans and stock views. Amazon and Microsoft did not comment on these views. Meta, Google, and Oracle also did not reply when asked.
The Challenge of Rapid Technological Obsolescence
GPUs face extra risks. They wear down and may also fall out of use when new parts come in. Nvidia CEO Jensen Huang joked at an AI event that a new "Blackwell" chip could make the previous "Hopper" part nearly worthless, even if it still runs. Nvidia now updates its AI chips every year instead of every two years. AMD also works in this way. This fast pace makes older GPUs lose use and worth quickly. Amazon even reduced its GPU life guess from six to five years after studying new AI trends.
Balancing Risk and Investment Strategy
Microsoft CEO Satya Nadella spoke of a big test for firms. "The biggest rival for any new Nvidia AI chip is the one that came before it," he said. Microsoft seeks not to spend too much on one type of AI part. This plan aims to cut long cost schedules on parts that soon seem old.
The Auditor’s Perspective on Depreciation Estimates
Dustin Madsen from the Society of Depreciation Professionals says that depreciation plans are money guesses that auditors check. He explains, "You must win over an auditor by showing that your guess for a part’s life is sound. They check tech data, past use, and trends of loss in use." His view shows that firms face hard work and doubt when planning a part’s life in a fast-changing tech scene.
As AI needs grow, the question "How long before a GPU loses value?" stays at the heart of money plans, firm accounts, and market steps. With fast tech shifts and market change, companies must balance hope for long use with care for tech shifts.
Reporting by Kif Leswing. Additional contributions by Jordan Novet.
