THE global technology industry has committed roughly US$400 billion this year to artificial intelligence chips and data centres, but growing uncertainty suggests that such record spending may not be sustainable, analysts have indicated.
Concerns are increasingly focused on whether expectations about the useful life of specialised AI chips are overly optimistic, particularly as rapid technological advances shorten the window before newer hardware makes existing processors less competitive.
Against a backdrop of persistent fears about an AI bubble, and with a significant share of the US economy now linked to AI-driven growth, analysts have warned that any correction could prove both abrupt and costly. According to AFP, these worries are intensifying as companies double down on investments despite mounting questions over long-term returns.
Investor Michael Burry, known for his role in the film, The Big Short, has been cited as describing the situation as “fraud” in a post on X last month, reflecting scepticism among some high-profile market observers.
Prior to the surge in interest triggered by ChatGPT, cloud computing companies generally assumed their chips and servers would remain in service for about six years. However, Mihir Kshirsagar of Princeton University’s Centre for Information Technology Policy has argued that the combination of physical degradation and rapid technological obsolescence makes that assumption increasingly unrealistic.
The pace of innovation has been identified as a central factor. Chipmakers, led by Nvidia, are introducing more powerful processors at an unprecedented rate. Nvidia, for example, announced that its Rubin chip would follow its Blackwell processor in less than a year, with reported performance gains of more than sevenfold.
At this rate, analysts such as Gil Luria of D.A. Davidson have said that AI chips can lose as much as 85 to 90 per cent of their market value within three to four years. Nvidia chief executive Jensen Huang has acknowledged that demand for older chips drops sharply once new models are released, suggesting limited ongoing appetite for previous generations.
Reliability has also emerged as a concern. Luria has observed that modern AI processors tend to fail more often than earlier hardware because of the extreme heat they generate, while a Meta study on its Llama model has pointed to an annual failure rate of about nine per cent.
Both Kshirsagar and Burry have suggested that the realistic lifespan of AI chips may be closer to two or three years. Nvidia has rejected this view, stating in November that industry estimates of four to six years are supported by real-world usage patterns.
Even so, Kshirsagar has argued that optimistic lifespan assumptions keep costs artificially low and mask underlying financial risks. Jon Peddie of Jon Peddie Research has echoed this concern, warning that shorter depreciation schedules would quickly erode profits and expose companies to accounting pressures.
Analysts have cautioned that the consequences could ripple through an economy increasingly dependent on AI. While diversified technology giants such as Amazon, Google and Microsoft are seen as relatively insulated, companies more narrowly focused on AI—such as Oracle and CoreWeave—are viewed as more vulnerable due to high debt levels and heavy spending on hardware.
Raising capital could become more difficult if frequent equipment replacement undermines profitability, particularly as some loans are backed by the chips themselves. To soften the impact, companies are reportedly considering reselling older chips or redeploying them for less demanding tasks, with analysts noting that recent-generation processors may still be economically viable for secondary or backup uses.
-BTSMedia.my
