In the high-altitude air of the current semiconductor boom, where Nvidia’s market capitalization has ascended to heights once reserved for entire national economies, a distinct sound of alarm is beginning to emanate from some of Wall Street’s most seasoned contrarians. While retail investors and institutional bulls continue to pile into the artificial intelligence trade, driven by the belief that the demand for computing power is infinite, a more structural skepticism is forming in the quiet corners of major hedge funds. The concern is no longer just about valuation multiples or interest rates; it is about the physical nature of the assets underpinning this rally. The argument posits that the most valuable commodity on earth right now—the H100 GPU—is rapidly transforming from a capital asset into a perishable good, akin to fresh produce that spoils if not consumed immediately.

Leading this charge of skepticism are two heavyweights of the investing world: Greg Jensen, the co-chief investment officer of Bridgewater Associates, and Michael Burry, the scion of distressed value investing famous for predicting the 2008 housing collapse. Their thesis strikes at the heart of the hardware cycle. They argue that the industry is mispricing the longevity of these chips, treating them like long-term infrastructure such as railways or fiber optic cables, when they are actually rapidly depreciating consumer electronics. According to a recent report by Business Insider, Jensen explicitly compares the hoarding of Nvidia chips to stockpiling iPhones—a strategy that guarantees losses as newer, faster models render current inventory obsolete.

Greg Jensen’s warning centers on the dangerous misconception that stockpiling high-performance silicon is equivalent to holding a store of value, whereas the rapid pace of technological innovation ensures that today’s gold standard is tomorrow’s electronic waste.

Jensen’s analogy of the iPhone is particularly bruising for the bullish narrative surrounding the hyperscalers—Amazon, Microsoft, and Google—who are currently engaged in an arms race to secure as much compute capacity as possible. In a traditional capital expenditure cycle, a company builds a factory or lays a pipeline, amortizing that cost over twenty or thirty years. However, in the AI compute cycle, the “factory” is composed of silicon that may lose half its performance-per-dollar value within eighteen months. Jensen notes that if you are not utilizing the chip the moment it arrives, you are essentially burning capital. This creates a terrifying dynamic for companies building reserves: the inventory on their balance sheets is depreciating not on a standard accounting schedule, but on a merciless technological curve defined by Nvidia’s own release cadence.

This view suggests that the massive capital expenditures we are seeing in 2024 and 2025 are not building a permanent moat, but rather a temporary stock of capability that must be constantly replenished at higher costs. As noted in coverage by Bloomberg, Jensen has previously highlighted that while the AI phenomenon is real, the market is “priced for perfection,” ignoring the inevitable commoditization of the hardware. If compute power becomes abundant and cheap due to massive oversupply and rapid obsolescence, the pricing power of the chipmakers—and the margins of the cloud providers renting those chips—could collapse simultaneously.

Michael Burry has moved beyond theoretical criticism and is actively positioning his portfolio to profit from a semiconductor correction, utilizing options strategies that suggest he sees a significant downside in the near term.

While Jensen provides the macro-economic theory, Michael Burry is providing the market mechanics for the bear case. Through his firm, Scion Asset Management, Burry has reportedly taken bearish positions against the semiconductor sector. His strategy involves purchasing put options against the iShares Semiconductor ETF (SOXX), effectively betting that the entire chip sector is due for a violent repricing. Burry’s skepticism is often rooted in the disconnect between revenue growth and sustainability. In the current environment, his bet implies that the market has overestimated the duration of the shortage and underestimated the speed at which the cycle turns from famine to feast, and eventually to glut.

Burry’s history of identifying bubbles before they burst adds a layer of gravity to these positions. His involvement suggests he sees a “Big Short” style setup where the consensus view—that AI spending will continue to double indefinitely—is mathematically impossible. As reported by Reuters regarding his portfolio adjustments, Burry often oscillates between sectors, but his targeting of semiconductors during a historic bull run signals a conviction that the “picks and shovels” trade has become overcrowded. If the end-users of AI applications do not monetize the technology fast enough to justify the hardware expense, the orders for new chips will evaporate, leaving Nvidia and its peers with bloated inventories and crashing average selling prices.

The introduction of Nvidia’s Blackwell architecture serves as the catalyst for this depreciation crisis, proving that the company’s own relentless innovation is the greatest threat to the value of its previous generation of hardware.

The irony of Nvidia’s dominance is that its success is built on destroying the value of its own installed base. The imminent arrival of the Blackwell chip architecture, which promises massive performance gains over the current Hopper (H100) series, illustrates Jensen’s point vividly. For a hyperscaler that just spent billions filling data centers with H100s, the Blackwell release is a double-edged sword. It offers new capabilities, but it instantly devalues the billions of dollars of hardware they just installed. In a rational market, the rental price for H100 compute should plummet once Blackwell comes online. This rapid obsolescence cycle forces companies into a “Red Queen” race where they must keep running (spending) just to stay in the same place.

This dynamic creates a precarious situation for the secondary market and the financing of these chips. During the peak of the shortage, H100s were being used as collateral for loans, treated as appreciating assets. If the value of an H100 drops by 50% upon the release of the next generation, the collateral base for these loans evaporates. The Wall Street Journal has previously documented the rise of “GPU-backed lending,” a financial innovation that looks increasingly risky if one accepts Jensen’s depreciation thesis. If the underlying asset behaves more like a decaying option than real estate, the financial leverage built on top of these chips could unwind rapidly.

The disconnect between the massive capital expenditures of the hyperscalers and the uncertain revenue streams from AI software creates an ‘ROI Gap’ that threatens to pull the rug out from under the hardware manufacturers.

The skeptics argue that the current market is witnessing a classic mesmerizing disconnect between Capex and revenue. The “build it and they will come” mentality has gripped Silicon Valley, but the revenue from generative AI applications—while growing—is nowhere near the level required to service the depreciation costs of the hardware being deployed. If a cloud provider spends $50,000 on a server cluster that becomes obsolete in three years, they need to generate massive daily profits just to break even. Currently, many AI startups are subsidized by venture capital, meaning the true end-user demand is distorted. When the VC subsidy dries up, the demand for raw compute may soften, exposing the overcapacity.

This leads to the “Prisoner’s Dilemma” faced by Big Tech CEOs. Even if they agree with Jensen that the chips are depreciating assets, they cannot stop buying them because doing so risks ceding dominance to a rival. However, this collective behavior leads to aggregate oversupply. The industry insiders watching this dynamic fear a repeat of the 2001 fiber optic bubble, where companies buried millions of miles of cable that went dark for a decade. The difference this time is that fiber optic cable doesn’t rot; silicon chips, functionally speaking, do.

Despite the compelling arguments for a bearish turn, the counter-narrative remains strong, supported by the possibility that AI represents a platform shift so profound that standard depreciation models simply do not apply.

To present a balanced view, one must acknowledge the bull case that defies Burry and Jensen. Proponents argue that the demand for intelligence is unlike the demand for bandwidth or storage; it is potentially uncapped. If AI models continue to scale according to scaling laws, older chips like the H100 may not become obsolete but simply move down the stack to handle inference tasks (running the models) while the new Blackwell chips handle training (building the models). This cascading utility could extend the useful life of the hardware, mitigating the depreciation cliff that the bears fear. Furthermore, the sheer complexity of the supply chain—specifically advanced packaging and memory—may keep supply constrained for years, artificially propping up the value of older chips.

However, history is on the side of the cyclicalists. Every semiconductor boom has ended in a glut. The specific mechanism of the bust is always different, but the result is the same: capacity overshoots demand. Greg Jensen and Michael Burry are betting that the laws of economics have not been suspended for the AI era. They foresee a moment where the depreciation bill comes due, and the market realizes that it has capitalized perishable inventory as permanent infrastructure. When that realization hits, the repricing of the entire sector will likely be swift, brutal, and rational.