Whalesbook
### The Depreciation Dilemma
The escalating investment in AI infrastructure is undeniable, with hyperscalers projecting hundreds of billions in capital expenditures for 2026. However, a growing technical debate, spearheaded by investor Michael Burry, suggests that the financial accounting for these assets may not align with their rapid economic decay, potentially creating a substantial valuation illusion.
Burry's core thesis centers on the depreciation of AI GPUs. While standard IT equipment might be depreciated over 3-6 years [48], he argues AI chips, due to rapid technological evolution, have an economic lifespan closer to 2-3 years [14, 40, 48]. Analyst Aakash Gupta estimates this gap could understate depreciation by $176 billion between 2026-2028, inflating reported income by over 20% at companies like Oracle and Meta [Source A, 42, 43]. This practice, Burry contends, is a "depreciation trick" that artificially boosts earnings [40, 43]. While some, like CoreWeave, assume around 6 years of useful life for GPUs, others warn of 3 years or less under heavy use [31]. The distinction between training and inference workloads also plays a role, with older chips potentially repurposed for inference, extending their economic life to 6-7 years or more [39]. However, Burry's argument posits that current depreciation schedules do not reflect the speed of obsolescence in high-performance computing hardware.
### CapEx Surge & Debt Burden
The demand for AI infrastructure is driving unprecedented spending. The top five U.S. hyperscalers are projected to spend approximately $600 billion in 2026 [13, 16], a significant increase from prior years, with roughly 75% of this earmarked for AI [16]. This massive capital outlay is increasingly financed through debt. Technology sector debt issuance reached record highs in 2025, comprising about 16.7% of global non-financial corporate bond issuance [13]. JPMorgan projects $1.5 trillion in future tech debt issuance [21], with $108.7 billion issued by tech companies in Q4 2025 alone [17]. Companies like Amazon have committed $200 billion in CapEx [Source A], projecting negative free cash flow in 2026, while Alphabet's free cash flow is expected to decline significantly [Source A]. This reliance on debt for capital-intensive, potentially volatile projects raises concerns about financial stability.
### Nvidia's Purity & Peer Valuation
Michael Burry specifically targets Nvidia due to its position as a "purest play" on the AI buildout [Source A]. Unlike diversified tech giants, Nvidia lacks significant non-AI revenue streams, making it more exposed if AI infrastructure demand falters. Nvidia's market capitalization stands at approximately $4.66 trillion [49]. As of February 2026, its P/E ratio hovers around 60-75 [49, 50], with an average analyst price target suggesting an upside of roughly 25% to $254.54 [23]. In contrast, AMD, another key player, has a market cap around $320 billion [49] and a P/E ratio near 75 [30]. Analysts see an average upside of about 45% to $283.69 for AMD [12]. While Nvidia commands a premium, the debate is whether its current valuation adequately discounts the potential risks of hardware obsolescence and the sustainability of the CapEx cycle.
### Historical Echoes: The RCA Parallel
Burry draws a striking parallel between the current AI boom and the 1920s radio craze, epitomized by Radio Corporation of America (RCA). RCA's stock surged approximately 200-fold before collapsing 98% between 1929 and 1932 [Source A, 20, 33, 36, 44]. At its peak in 1929, RCA traded at a P/E ratio of 72x [33, 44], demonstrating a market valuation far detached from economic fundamentals. RCA's stock price did not recover its 1929 levels until the 1960s [33, 36]. The RCA narrative serves as a cautionary tale: a revolutionary technology (radio) did not fail, but the stock valuation, driven by speculative fervor, imploded. The AI infrastructure buildout, while technologically profound, faces a similar risk if market optimism outpaces sustainable economic returns from the capital deployed.
### The Forensic Bear Case
Nvidia's dependence on the AI infrastructure buildout is its primary vulnerability. If hyperscaler spending slows, or if hardware obsolescence outpaces the projected useful life, Nvidia's revenue and earnings could be severely impacted. While some analysts argue that older GPUs can be repurposed, extending their economic life [39], Burry's thesis challenges this adaptability in the face of continuous, rapid innovation cycles. The risk is that the "economic obsolescence gap" will widen, leading to significant write-downs and earnings restatements. Furthermore, the sheer scale of capital expenditure, often debt-funded by hyperscalers, creates a fragile ecosystem. If the projected returns on these colossal investments do not materialize swiftly, the financial strain could trigger a sharp correction across the AI supply chain. Unlike diversified tech giants, Nvidia has less buffer against a downturn in AI hardware demand.
### Future Outlook
Despite concerns, the semiconductor market is projected for robust growth, expected to approach $1 trillion in 2026, driven by AI [2, 3, 4, 8]. Gartner forecasts worldwide AI spending to reach $2.52 trillion in 2026, with AI infrastructure consuming $1.37 trillion [19]. Analyst sentiment for Nvidia remains largely positive, with a consensus "Strong Buy" rating and price targets suggesting significant upside [6, 11, 23, 24]. Similarly, AMD garners "Buy" ratings, indicating a belief in continued sector strength. However, the scale of investment, combined with the underlying accounting questions raised by Burry and others, suggests that the narrative of unchecked AI infrastructure growth may face increasing scrutiny. The market's ability to absorb this debt and sustain these capital expenditures hinges on the real-world economic returns generated by AI.