Nvidia faces questions over AI chip values and investor risks

Nvidia, led by Jensen Huang, recently unveiled the Blackwell Ultra, a highly advanced graphics processor designed for artificial intelligence workloads. Tech giants such as Amazon, Meta, and OpenAI have committed vast resources to secure these cutting-edge chips, fuelling unprecedented demand and propelling Nvidia’s quarterly revenues to $54 billion dollars, with a soaring valuation now at $4.7 trillion dollars.

This surge in spending on AI hardware has prompted scrutiny about the long term value of Nvidia’s chips. Several analysts view the rapid depreciation of these assets as a potential risk, raising parallels with past technology investment bubbles. The debate hinges on how long these chips remain effective and profitable. Traditionally, companies amortise IT hardware over several years. Recently, expectations for the useful life of these processors have shifted upwards from three to five years for some of the world’s biggest technology firms. If these estimates prove accurate, reported profits will be higher. If not, there is a risk of significant write downs in future reporting periods.

Prominent investors, such as Jim Chanos and Michael Burry, have expressed concern about extending the lifespan of Nvidia’s chips in corporate accounts. Chanos likens the current environment to the late 1990s telecoms bubble, warning that if the effective usage of graphics processors is closer to two or three years, reported profits may be substantially overstated. Burry has gone further, recently placing a billion dollar bet against the AI boom and alleging that these lifespan extensions could understate industry wide depreciation by $176 billion dollars by 2028. Companies have started to revise their estimates, with Amazon shortening its chip depreciation cycle from six to five years, at a cost of $700 million dollars.

The challenge is not limited to technology giants. A wave of emerging data centre providers, often referred to as neocloud companies, are forecasting chip lifespans of up to six years. This optimistic outlook has met scepticism. Kerrisdale Capital recently criticised CoreWeave, a prominent neocloud operator, for adopting what it considers overly aggressive asset accounting. In contrast, a competitor, Nebius, operates with a four year depreciation cycle. CoreWeave, for its part, claims that its estimates draw upon extensive real world usage data and points to contract renewals for older GPUs still valued at 95 percent of their original price.

Despite these arguments, some industry analysts downplay claims that rapid chip depreciation poses a systemic financial risk. Richard Windsor notes that Google’s locally developed chips, now up to seven years old, continue to be used in data centres. He contends that as long as demand for AI workloads remains high, even older processors will find productive roles within the technology ecosystem. David Harold of Jon Peddie Research adds that older accelerators are often repurposed or resold rather than retired, thus extending their economic contribution beyond their primary service life.

Investor sentiment, nevertheless, remains unsettled. Large scale technology enterprises such as Meta and Amazon are likely to absorb any financial shocks connected to depreciation forecast errors. Smaller, newer companies may be more vulnerable. Recent market reactions highlight the risks; CoreWeave, for example, saw shares fall sixteen percent after underwhelming results, wiping out a third of its valuation during a period of tech sector volatility.

The durability and value of Nvidia’s AI chips now serve as bellwethers for the broader industry. As capital investment in AI infrastructure accelerates, the ultimate viability of these hardware assets could be the decisive factor between sustained growth and a disruptive reversal for the sector.