The market narrative says Michael Burry is betting against the king of AI. But a closer look at the trade's anatomy reveals something more interesting: a hedge against his own conviction.
Scarcely a week passes without a new declaration of a top in AI. The latest shot comes from the legendary investor, Michael Burry, who has reportedly opened a short position against Nvidia (NVDA) while simultaneously purchasing call options as a hedge. The headline is predictable. The narrative is a mixture of “brief monopoly” and “capital expenditure bubble.”
But this is a trap. The market is reading a trade, not the thesis. The real analysis lies beneath the surface, in the architecture of the business and the history of its narrative cycles. We should not be asking if the short is justified. The question is whether the reason for the short is structurally sound. Based on the data, it is not.
The Mechanical Heart of the Monster
To understand the short, you must first understand the revenue engine. Nvidia is not just a chip designer. It is a platform company with a software moat that is often underestimated. As of 2024, the CUDA software ecosystem boasts over 4 million developers. This is not a convenience; it is a handcuff.
When a developer writes code in CUDA, that code is not portable to AMD or Google TPUs without significant rewrites. The cost of switching is not just the price of silicon. It is the entire stack of infrastructure, the networking, the software libraries, and the expertise of the team. This creates a dependency cycle that is far more durable than the physical hardware itself.
Burry's public thesis acknowledges a “brief monopoly” power. He sees the GPU demand as cyclical and the competition as inevitable. He looks at the technical specs of AMD's MI300 series, which are indeed competitive in raw benchmark tests. He sees Google's TPU and the ASIC chips from OpenAI and Broadcom. He sees the customer self-engineering efforts at Microsoft and Amazon. He sees the writing on the wall.
But he is reading the wrong wall.
The market is currently a bull market. In this environment, investors are chasing the highest growth. They are FOMOing into AI. The narrative is about Nvidia's dominance. Burry is betting on the displacement of that dominance. Yet, based on my analysis of the fundamental structure, the “brief” in that monopoly is likely to last longer than the market cycle can hold a short.
The “Brief Monopoly” is a Structural Monopoly
The core of the bearish argument is the threat of self-designed ASIC chips. It is a valid threat. Amazon's Trainium, Microsoft's Maia, and Google's TPU are all designed to reduce dependency on Nvidia. They are the main risk to NVDA's dominance. The hidden information here is that these chips are not being deployed to replace Nvidia. They are being used to replace some Nvidia workloads. They are not generic. They are specialized.
Nvidia's true defense is the system-level architecture. They do not sell a chip; they sell the DGX system. They sell NVLink and InfiniBand networking. They sell the CUDA software stack. This is not a single component. It is an integrated architecture. It is incredibly difficult to replicate.
If a customer buys a self-engineered chip, they still need the network. They still need the software. They still need the interconnect. Nvidia's platform is a sticky ecosystem. The short thesis treats the GPU as a commodity. It is not. It is a foundational layer. Even if a customer builds a custom ASIC for inference, they will still require the general-purpose GPUs for the training and the data processing. This makes the threat more of a complementary challenge than a substitution threat.
Nvidia's pricing power is not a function of hardware alone. It is a function of the return on investment (ROI). Even if the H100 costs $30,000, it is a bargain if it saves the customer $50,000 in compute time. As long as the AI ROI is positive, the price elasticity is low. The demand is inelastic to the price of the chip. It is elastic to the cost of the output.
The Contrarian Angle: The Capital Expenditure is the Real Trap
Let's examine the contrarian angle. The bearish thesis hinges on “capital expenditure entering the bubble top.” Burry is not wrong. The hyperscalers (Microsoft, Meta, Amazon, Google) are spending astronomical amounts on data centers. There will be a hangover. But the market is conflating the customer's capital expenditure with Nvidia's revenue.
If the hyperscalers cut their spending in 2026, Nvidia's revenue growth will slow. The PE ratio of 60-70 times will look expensive. That is the fundamental risk. It is not the competition. It is the demand cycle. This is a real risk. The data suggests it's a medium probability event. But the short position is a bet on a very high probability of a severe decline.
My experience from 2020 DeFi Summer taught me that narratives often diverge from the mechanics. During the DeFi bull run, the yield narratives were wild, but the fundamental usage data was a better predictor of value than the sentiment. The same applies here. The narrative is about competition. The data is about the cyclicality of the customer's capital expenditures.
If you look at the historical data, the top five customers (Microsoft, Meta, Amazon, Google, Oracle) account for over 40% of Nvidia's revenue. They are the primary source of the boom. The contradiction is that the same customers are the ones building their own chips. They are the ones reducing the dependency. This is a double-edged sword. If they build their own chips, they reduce their CapEx on Nvidia. But if they do not, they are stuck. The most likely scenario is a hybrid: they use their own chips for the high-volume, standardized inference workloads, and they use Nvidia for the complex training workloads. This will erode Nvidia's unit share but not its value share. The margins will remain high for the complex workloads.
The Regulatory Ceiling
There is also the regulatory layer. Nvidia's export controls to China are a political risk, not a technical one. But the market is not pricing in the impact of AI being used in defense. The ethical concerns are not going to reduce the revenue. They might increase the barriers to entry. The "military" narrative is a regulatory risk that could actually strengthen the pricing power of the incumbents who can navigate the compliance.
The market is looking at the old measure of chip share. The future is the measure of software share. Nvidia's AI Enterprise software subscription is the new wave. It's a recurring revenue model. It's the tool that moves the market to a lower cost. It's a hedge against the hardware cycle. This is the angle that most short sellers overlook. They look at the hardware as a cycle, and the software as a value multiplier. That is the same pattern as the PayPal USDC case: the utility of the underlying asset changes the regulatory and market posture.
## The Verdict: A Hedge, Not a Call The specifics of the trade: the short, plus the call hedge with the strike price in the mid-200s. This is not a confident bet. This is a protection. It is a way to participate in a decline without being exposed to the short squeeze. It's a bet on the volatility. The market needs to recognize that Burry is not betting on the death of the AI king. He is betting on the timing of the "overinvestment" correction.
The real risk to the short is that the "bubble top" is not yet. The market's capital expenditure is still rising. The demand is still outstripping supply. The NVIDIA of the future is not the chip. It's the platform. If the market sees the AI GPU as a pure commodity, then the short is a trade. But if the market sees it as a critical piece of the tech infrastructure, the short is a trap.
The end of the cycle is not defined by the competition. It is defined by the customer's capital allocation. Watch the spending of the hyperscalers. If they pause, the short is right. If they continue to spend, the short will be squeezed. History has not seen a time when the infrastructure vendor was displaced while the industry was still expanding. Not yet.
