Investor Michael Burry compares Nvidia's GPU depreciation claims to a 1960s market bubble, questioning the real value of AI infrastructure investments.

The debate over the sustainability of the artificial intelligence gold rush has shifted from vague predictions about productivity to the granular details of corporate accounting. At the center of this friction is Nvidia and a specific line item in financial disclosures: depreciation schedules. Michael Burry, the investor famous for forecasting the 2008 housing collapse, has entered the fray, suggesting that the current justification for AI infrastructure spending is a ghost from the past.

The dispute over GPU residual value

The tension ignited following a September 27 presentation by Nvidia. In a slide titled 'NVIDIA AI Infrastructure Retains Value Beyond Accelerated Depreciation Schedules', the company plotted the retained value of its A100, H100, and B200 chips against a standard five-year accelerated depreciation curve. The core message was clear: companies are over-depreciating their hardware because these chips maintain significantly higher values across generations than traditional accounting assumes.

For the bulls, this was a definitive victory. If the hardware does not lose value as quickly as accountants predict, the return on investment for the trillions of dollars flowing into AI data centers looks far more attractive. However, Burry views this not as a technological breakthrough, but as a psychological pattern. He argues that we have all been here before, drawing a direct line between today's GPU valuations and the market bubbles of the 1960s.

Lessons from the 1960s leasing mania

Burry's critique, detailed in his Substack newsletter 'Cassandra Unchained', focuses on the structural similarities between the current AI build-out and the computer-leasing bubble of the late 1960s. During that era, the aggressive deployment of capital into early computing infrastructure was fueled by unsustainable leasing models that ignored the rapid obsolescence of the hardware.

According to Burry, the current market is repeating this mistake by relying on long-term discounted cash flow models to justify the residual value of GPUs. He asserts that these models are theoretical and do not reflect actual secondary market prices. While Nvidia suggests the hardware is a durable asset, Burry warns that the financial engineering used to support this expansion is precarious.

The role of private credit and GPU-backed debt

One of the most concerning elements for Burry is the emergence of GPU-backed debt financing. In a bid to scale AI capacity rapidly, some firms are using their hardware as collateral for private credit. This creates a feedback loop where the perceived value of the chips supports the loans used to buy more chips.

Burry suggests that recent rebounds in rental rates for older GPUs are not evidence of long-term hardware durability. Instead, he attributes these price spikes to temporary global bottlenecks in energy and memory. By mistaking a supply-chain glitch for a fundamental value floor, investors may be ignoring the risk of a sudden price collapse once those bottlenecks clear.

The comparison Burry makes says less about computers than it does about human nature repeating itself.

Portfolio shifts and tactical hedging

Burry is not merely shouting from the sidelines; he is adjusting his financial exposure to match his skepticism. Recent filings show a significant repositioning of his portfolio. Burry covered all his short common-stock positions in several key players, including Nvidia, Palantir, Micron, and the iShares Semiconductor ETF (SOXX), as well as CoreWeave and Nebius.

However, this was not a surrender. Burry replaced these short positions with put options. By switching from shorting common stock to using options, the investor is maintaining a bearish bet on the sector while altering his risk profile. This tactical move suggests he still expects a correction, even if he is no longer betting against the stocks in a linear fashion.

The clash of narratives: Bulls vs Bears

The divide in the market currently rests on two opposing views of AI hardware. On one side, the Nvidia narrative posits that AI chips are a new class of asset—infrastructure that remains productive and valuable even as newer versions arrive. This view suggests that the AI revolution is a fundamental shift in computing that renders old depreciation rules obsolete.

On the other side, Burry and other critics, such as Jim Chanos, question the logic of these valuations. Chanos has previously challenged CEO Jensen Huang on why Nvidia does not rent out its own chips if they are as highly rentable and valuable as the company claims. The core of the bear case is that the AI trade is built on a foundation of accounting assumptions that will eventually crumble when the secondary market for hardware fails to meet the theoretical projections.

Global implications for AI enterprises

For entrepreneurs and business leaders in the USA, UK, and global markets, this debate is more than a Wall Street squabble; it is a warning about capital expenditure (CapEx) strategy. The risk highlighted by Burry is that companies may be overestimating the terminal value of their AI investments, leading to future balance sheet shocks when assets must be written down.

In the US and UK, where private credit is increasingly used to fund tech expansion, the reliance on GPU-backed loans could create systemic vulnerabilities. While the US lacks a centralized AI regulatory body similar to the EU's AI Act, the financial oversight of how these assets are valued on corporate balance sheets remains a critical point of failure. Businesses should be cautious about treating AI hardware as a long-term store of value and instead plan for a faster cycle of obsolescence. If Burry's 1960s analogy holds true, the winners will not be those who accumulated the most hardware, but those who generated actual revenue before the depreciation curve caught up with the hype.

FAQ

---

Sources: Finance (2), Newsbreak ·