Nvidia Stock: Is the AI Boom Hiding A Hardware Time Bomb?

Michael Burry, the famed investor who predicted the 2008 housing collapse, is once again betting against a seemingly invincible market. This time, his target is the AI industry, specifically the valuations of Nvidia (NASDAQ:NVDA) and the hyperscalers buying its chips. While the market sees a once-in-a-lifetime revolution in artificial intelligence, Burry sees a bubble that is inflated not just by hype, but by a relatively mundane accounting metric: Depreciation.  The core of the debate isn’t about whether AI is real—it’s about whether the “shovels” used to dig for this gold will turn into rust faster than accountants are willing to admit. In an industry burning cash at unprecedented levels, small accounting choices can create the illusion of profitability —and even small shifts in how these giants allocate their massive AI budgets could bring the current narrative crashing down.

When markets turn, single-asset exposure hurts. Smart financial advisors protect client wealth by working with partners who allocate across multiple asset classes – including the High Quality (HQ) Portfolio.

Image by Pete Linforth from Pixabay

An Accounting Tweak

In recent years, tech giants like Microsoft, Meta, and Google made a subtle change to their balance sheets: lengthening the estimated “useful life” of their server hardware from roughly four years to nearly six. By extending this period, they spread the massive cost of equipment (like the roughly $30,000 Nvidia H100 GPU) over a longer duration, reducing annual expenses and boosting net income. Microsoft reportedly extended its server / network-equipment depreciation horizon from four years to six years (effective from  2022). For its part, Google shifted from a roughly four-year lifespan to six years in 2023. Meta adopted a more gradual approach – moving from four years up to five and a half years by early 2025. This maneuver postpones the financial pain, but the profit picture could deteriorate quickly when faster obsolescence or depreciation eventually catches up. This is critical, as the biggest spenders are projected to boost their combined capital expenditures to about $460 billion in the next 12 months.

But this accounting maneuver raises a critical question:

Can a GPU Survive A 6-Year Marathon?

Unlike CPUs, which often run at variable loads in cool environments, AI GPUs are the workhorses of the data center. They are subjected to “thermal cycling” – heating up to $80\; degrees\; Celsius\; (over\; 170\; degrees\; Fahrenheit)$ under massive loads and cooling down, over and over again. This expands and contracts the solder, eventually leading to physical failure. If these chips physically degrade or “burn out” their voltage regulators in year four, the billions of dollars of “value” booked for years five and six will evaporate instantly. While modern servers incorporate advanced hardware and software features designed to improve reliability, these resilience mechanisms are largely based on older or general-purpose workloads rather than the extreme continuous demands of AI workloads running on GPUs.

Even if the hardware physically survives, Burry’s thesis suggests it will die an economic death long before 2030. Nvidia is currently releasing considerably faster chips every 18 months. The Blackwell architecture introduced in Q1 this year offers massive efficiency gains over the current Hopper (H100) line. The next gen Rubin chips are expected by around the first half of 2026. If a new chip can generate five times the AI tokens for the same electricity cost, the old chips become liabilities, as it becomes cheaper to move on and buy new hardware. If innovation moves faster than depreciation, we could be headed toward a massive correction. Burry predicts that hyperscalers will eventually be forced to admit their stockpiles of H100s are economically worthless. This would trigger billions in “write-downs” – sudden losses that could shatter the perceived profitability of the AI trade.

The Real Risk: Slower Capex

What Could Be The Trigger Event? 

What specific events would force a major hyperscaler (Meta, Microsoft, Google) to accelerate their depreciation schedule or announce a significant impairment charge (write-down)?  The timing is everything.

• Competitor Pressure: A rival cloud provider successfully uses a shorter, more conservative 3-4 year depreciation schedule to gain trust and signal efficiency.
• Auditor Scrutiny: External auditors force an acceleration due to accumulating evidence of physical or economic failure.
• CEO Admission: If heads of these companies publicly admits the faster pace of AI innovation or chip degradation warrants a shorter life.  This could trigger a reckoning within the industry, leading to write-downs, impacting Net Income and Book Value across the sector.