Nvidia booms as debate over AI chip value divides the market

Nvidia’s dominance in the artificial intelligence sector has reached new heights as its latest processor, the Blackwell Ultra, is rapidly reshaping the landscape of data centre investment. Earlier this year, Jensen Huang, Nvidia’s founder, announced the arrival of the superpowered chip with characteristic showmanship, driving demand for Nvidia hardware to levels previously unseen. Tech giants including Amazon, Meta and OpenAI are investing trillions to secure the company’s most advanced technology, sending Nvidia’s quarterly revenues surging to 54 billion dollars and boosting its valuation to 4 point 7 trillion dollars.

The rising fortunes of Nvidia, however, have given rise to critical questions about the true longevity and depreciative pace of AI chips. Corporate accounting practices have come under fresh scrutiny as large technology companies extend the expected lifespan of graphics processors from three to five years. Such adjustments have the potential to inflate apparent profits, since hardware costs are allocated over longer periods. If these optimistic forecasts prove inaccurate, firms could be forced into substantial write downs as assets wear out or become obsolete sooner than projected.

Influential investors and short sellers are voicing concerns. Jim Chanos, known for predicting Enron’s collapse, pointed out the significant impact of depreciation estimates on company profits, arguing that if GPUs only remain economically viable for two or three years, declared profits may be substantially overstated. Michael Burry, famous for his role in predicting the 2008 financial crisis, has similarly accused major companies of overstating their financial health through extended asset lifespans. Burry’s analysis suggests the industry could understate depreciation by 176 billion dollars by 2028, potentially setting the scene for a financial reckoning.

The discussion is not limited to industry heavyweights. Emerging data centre firms, sometimes labelled neocloud providers, are adopting even more ambitious depreciation timelines. For example, CoreWeave estimates its chips will last up to six years, a position criticised by short sellers for flattering financial statements. Its rival Nebius, in contrast, uses a four year depreciation cycle. Although CoreWeave claims its estimates are grounded in operational data and strong contract renewals, its shares recently fell 16 percent after disappointing results, demonstrating the volatility associated with asset valuation in the sector.

Amidst this debate, some analysts maintain that the risks may be overstated. Older AI accelerators are often repurposed for less demanding workloads or resold, extending their utility and revenue potential. Google continues to employ in house chips developed seven years ago, indicating that nascent obsolescence may not pose a universal threat. As demand for AI remains robust, older components are expected to maintain value for lower intensity applications, supporting ongoing returns for large technology firms.

Yet the tension between aggressive growth and asset write down risk remains significant. While established companies may possess the resources to weather revaluations, smaller start ups face greater exposure. The fate of Nvidia’s advanced processors is therefore intricately linked to both the highs and lows of the AI investment cycle, deciding not only company fortunes but potentially influencing the future trajectory of the sector as a whole.