The debate over how long AI chips and GPUs can realistically last has heated up in recent weeks, with some investors warning that extended depreciation schedules could artificially inflate tech company profits. But according to a new report from Bernstein, the controversy may be overblown.
The Bear Case
Skeptics argue that setting depreciation periods of five to six years for GPUs is too optimistic. Michael Burry, the investor famous for shorting the subprime mortgage bubble, has warned that large tech firms may be underestimating depreciation by stretching the lifespan of computing equipment.
By lowering book costs, companies can make profits appear stronger than they really are, Burry cautioned. His comments have fueled broader concerns about whether the AI boom is masking accounting risks in the sector.
Bernstein’s Response
Bernstein’s research counters that these fears are "exaggerated." The firm notes that major cloud service providers generally apply accurate depreciation methods for accelerated computing equipment and have not misrepresented GPU lifespans.
The report acknowledges that GPU value does decline sharply in the first year, largely due to burn-in damage and the migration of high-performance workloads to newer chips. But after that initial drop, resale prices stabilize, and older GPUs remain useful for low-intensity computing tasks. This makes a five- to six-year lifespan both reasonable and common practice.
Economics of Older Hardware
Bernstein also points out that the operating costs of older GPUs are still far lower than leasing comparable equipment, allowing suppliers to maintain profitability even with hardware that is five years old.
Additionally, long-term contracts between cloud providers and customers can help transfer depreciation risk, ensuring that faster-than-expected hardware declines do not fall entirely on suppliers.
Industry Growth Still Intact
Despite the debate, Bernstein concludes that GPU depreciation assumptions are not undermining industry growth. Demand for accelerated computing remains massive, and older hardware continues to play a role in meeting that demand.
The report stresses that there has been no structural decline in replacement demand, meaning the upgrade cycle remains intact even as companies extend depreciation schedules.
Market Reaction
The controversy briefly weighed on Nvidia (NVDA-US), which dipped to around $184 on Monday before bargain hunters stepped in. The stock closed at $186.60, down 1.88%. Broader markets also softened, with the Nasdaq Composite falling 0.84% and the Philadelphia Semiconductor Index down 1.55%.
The Bottom Line
While bears like Michael Burry warn of accounting risks, Bernstein argues that GPU depreciation practices are largely sound. The firm’s view is that extended lifespans reflect real-world usage patterns and do not pose a major threat to industry profitability.
For investors, the takeaway is clear: depreciation debates may stir headlines, but the AI hardware cycle remains robust, with demand far outweighing concerns about accounting assumptions.