The tech industry’s substantial investment in artificial intelligence has sparked concerns about the longevity of specialized chips. This year, the industry has allocated approximately $400 billion to these chips and data centers, but doubts are emerging about the wisdom of such unprecedented investment levels. At the core of these doubts are overly optimistic estimates regarding the lifespan of these specialized chips before they become obsolete.

Typically, cloud computing giants assumed that their chips and servers would last about six years. However, the rapid pace of technological advancements, particularly with chip makers like Nvidia releasing new, more powerful processors, is challenging this assumption. Mihir Kshirsagar of Princeton University’s Center for Information Technology Policy notes that the combination of wear and tear, along with technological obsolescence, makes the six-year assumption difficult to sustain.

The accelerated release of new chips is significantly reducing the market value of existing ones. For instance, Nvidia’s announcement of the Rubin chip, which is expected to have 7.5 times greater performance than its predecessor, less than a year after launching the Blackwell chip, underscores this point. As a result, chips are losing 85 to 90 percent of their market value within three to four years, according to Gil Luria of D.A. Davidson.

Furthermore, AI processors are failing more often due to running at high temperatures, which can cause equipment to burn out. A recent Meta study found an annual failure rate of 9 percent for its Llama AI model. These factors suggest that the realistic lifespan of AI chips may be just two or three years, rather than the industry’s estimated four to six years.

The potential consequences of these overly optimistic assumptions are significant. If companies are forced to shorten their depreciation timelines, it could immediately impact their bottom line and slash profits. This could have a ripple effect throughout the economy, which is increasingly dependent on AI. While giants like Amazon, Google, and Microsoft may be less affected due to their diverse revenue streams, AI specialists like Oracle and CoreWeave, which are already heavily indebted, may face significant challenges in raising capital if their equipment needs to be replaced more frequently.

As the industry navigates these challenges, some companies are exploring options like reselling older chips or using them for less demanding tasks. However, the situation remains precarious, particularly for companies that have used chips as collateral for loans. The potential fallout from the AI boom’s reliance on rapidly obsolete chips serves as a reminder of the importance of realistic assessments and planning in the tech industry.