Navigating AI's Uncertain Future: The GPU Depreciation Debate Heats Up

Investors and tech giants are grappling with the critical question of how long graphics processing units (GPUs) will remain valuable, as the industry braces for a $1 trillion spend on AI data centers over the next five years. Google, Oracle, and Microsoft estimate their AI servers could last up to six years, but short seller Michael Burry warns that these projections may be overly optimistic.

Industry Giants Set Long Lifespans, Skeptics Disagree

Google, Oracle, and Microsoft have pegged the lifespan of their AI computers at up to six years. However, Microsoft's latest annual filing suggests a more conservative range of two to six years. This wide disparity is causing concern among investors and lenders, who need to factor in depreciation to gauge the financial health of these companies.

Depreciation and Financial Impact

Depreciation, the allocation of the cost of a hard asset over its useful life, is a crucial concept in the tech industry. The longer equipment remains valuable, the more years a company can stretch out depreciation, reducing the impact on profits. For instance, if a GPU lasts six years instead of three, it can significantly improve a company's financial outlook.

Data-Driven Insights from CoreWeave

CoreWeave, a company that buys and rents out GPUs, has been using a six-year depreciation cycle since 2023. CEO Michael Intrator told CNBC that his company is being 'data driven' about GPU shelf life. He noted that CoreWeave's Nvidia A100 chips, announced in 2020, are all fully booked. Additionally, a batch of Nvidia H100 chips from 2022 became available due to an expired contract and were immediately rebooked at 95% of their original price.

Market Reactions and Concerns

Despite positive data points, CoreWeave's shares plunged 16% after its earnings report, due to delays at a third-party data center developer. The stock is down 57% from its high in June, reflecting broader concerns about overspending in AI. Oracle shares have also plummeted 34% from their record high in September.

Vocal Skepticism from Michael Burry

Short seller Michael Burry, known for his bets against Nvidia and Palantir, is one of the most vocal skeptics. He recently suggested that companies may be overestimating the longevity of their AI infrastructure. This skepticism adds to the uncertainty surrounding the true lifespan of GPUs and the financial risks associated with large-scale investments in AI hardware.

References

• The question everyone in AI is asking: How long before a GPU depreciates?