Nvidia and the Short-Sellers: Can AI's Infrastructure Resilience Defy Bearish Bets?

The current debate over Nvidia's valuation and the sustainability of its AI-driven growth has drawn sharp contrasts between bullish optimism and bearish skepticism. At the heart of this tension lies the short thesis articulated by Michael Burry, the "Big Short" investor, who argues that the AI infrastructure boom is built on fragile accounting assumptions and unsustainable demand. To evaluate this thesis, we must dissect the interplay of supply-demand dynamics, valuation realism, and historical parallels.

The Accounting Mismatch: Depreciation and Earnings Inflation

Burry's core argument hinges on the mismatch between the economic reality of AI chip obsolescence and the accounting practices of hyperscalers. He contends that companies like Microsoft, Google, Amazon, and Meta are depreciating AI GPUs over 5–6 years, despite the fact that these chips become technologically obsolete in 2–3 years due to rapid advancements in AI models according to analysis. For instance, Microsoft's $17 billion in GPU purchases in 2024, when depreciated over 6 years instead of 3, results in a $2.9 billion annual overstatement of earnings-3.3% of its FY2024 net income according to data. Extrapolating this logic, Burry estimates that cumulative earnings inflation across the sector could reach $10–12 billion annually, or 11–14% of reported profits according to research.

This accounting discrepancy, if exposed, could trigger a reassessment of the financial health of hyperscalers and, by extension, Nvidia, which supplies the bulk of their AI infrastructure. However, Nvidia's CFO has pushed back, asserting that older GPU models like the A100 remain economically viable for extended periods according to CNBC. The company's recent Blackwell and Rubin chips, set for 2026, are also marketed as durable solutions for multi-year AI workloads according to Morningstar.

Supply Constraints and Demand Resilience

Nvidia's dominance in the AI chip market-accounting for ~90% of data-center GPU sales according to Fool-is underpinned by its technological lead and the CUDA ecosystem. Yet, supply constraints are emerging as a critical bottleneck. For example, the H200 AI chip, designed for Chinese hyperscalers, faces a demand-to-inventory gap of 2 million units in 2026, with TSMC struggling to expand CoWoS packaging capacity to meet this need according to Reuters. Similarly, HBM (High-Bandwidth Memory) and 3nm/2nm wafer production are fully oversubscribed, with TSMC's capacity expected to remain strained through mid-2026.

Despite these challenges, demand for AI infrastructure remains robust. Nvidia reported Q4 2025 revenue of $39.3 billion, a 78% year-over-year increase, driven by its data-center segment. Analysts project that AI compute demand will outpace supply in 2026, with the rollout of Blackwell and Rubin chips potentially reinforcing Nvidia's leadership according to Morningstar. However, Burry warns that this demand could collapse if hyperscalers tighten capital controls, given their reliance on Nvidia as a "pure play" on AI infrastructure spending according to Reddit analysis.

Valuation Realism: A Dot-Com Bubble Analogy?

Nvidia's trailing P/E ratio of 45.8 and forward P/E of 39.5 according to Fool appear elevated but are far below the peak valuations of the dot-com era. During the 2000 bubble, the unweighted average P/E of the "Magnificent 7" tech stocks reached 276x, with Cisco trading at a P/E of 472x according to research. By contrast, today's tech giants, including Alphabet (P/E 24.94) and Amazon (P/E 32.78), are valued more conservatively.

Burry, however, draws parallels to the dot-com bubble by comparing Nvidia to Cisco, arguing that its valuation assumes indefinite exponential growth according to Fortune. Yet, unlike the speculative tech firms of 2000, Nvidia generates substantial free cash flow and operates in a sector with real, near-term demand. The company's gross margin of ~75% according to Morningstar and its role in training next-generation models like OpenAI's GPT-6 according to Morningstar suggest a durable competitive advantage.

Competitor Threats and Market Share Shifts

While Google's TPUs and Amazon's Tranium3 chips offer cost efficiencies and performance improvements, they remain niche solutions. Custom silicon programs by Meta, Apple, and Broadcom are also limited in scope, with Nvidia's merchant silicon still dominating general-purpose AI workloads according to Morningstar. Analysts project that while market share may shift, Nvidia's ecosystem and R&D pipeline will sustain its leadership through 2026 according to Fool.

Conclusion: A Structural Shift or a Bubble?

Burry's thesis rests on the assumption that the AI infrastructure boom is a speculative bubble, driven by accounting distortions and unsustainable demand. While supply constraints and valuation risks are real, the current AI rally is anchored in tangible demand, technological innovation, and strong earnings growth. Unlike the dot-com era, where many companies lacked profitability, today's hyperscalers and their suppliers are generating cash flow and reinvesting in long-term infrastructure according to iShares analysis.

That said, the risks Burry highlights-overstated earnings, energy bottlenecks, and the rise of custom silicon-cannot be ignored. If hyperscalers begin to prioritize cost efficiency over raw compute power, or if regulatory shifts disrupt chip exports to China, Nvidia's growth trajectory could face headwinds. For now, however, the data suggests that AI's infrastructure resilience may yet defy the bearish bets.