The Flaws in Michael Burry's 2026 AI Bubble Thesis

When Michael Burry predicted the 2008 financial crisis, he wasn’t just right—he became a legend. The contrarian investor made roughly $100 million in personal profit and $700 million for his investors at Scion Capital by correctly forecasting the subprime mortgage collapse. That prescient call, immortalized in the film “The Big Short,” cemented his reputation as a financial oracle. Today, however, Michael Burry’s latest bearish crusade against artificial intelligence stocks appears to ignore the very real profits and operational improvements already materializing across the AI industry.

Who Is Michael Burry and Why His Track Record Cuts Both Ways

Michael Burry’s legend rests on one extraordinary prediction. Yet the years following his 2008 windfall tell a different story. As U.S. equity markets surged over the past decade-plus, Burry has repeatedly issued premature bearish warnings that failed to materialize. In fact, he shuttered his hedge fund in late 2025, citing misalignment with market trajectories. This pattern suggests something crucial: being right once doesn’t guarantee accuracy twice, especially when market conditions fundamentally shift.

The real issue isn’t that Burry lacks analytical rigor—it’s that he appears to be fighting the last war. His 2008 success relied on identifying structural dysfunction in the financial system. Today’s AI landscape operates under entirely different dynamics, something his current thesis seems to discount.

AI Infrastructure Demand Shatters Burry’s Depreciation Argument

One of Michael Burry’s core contentions centers on accounting manipulation. He claims that tech giants like Meta, Microsoft, and Alphabet artificially inflate earnings by using overly aggressive depreciation schedules. As an example, he notes that Alphabet depreciates its servers over “four to six years.”

The problem with this argument? It misunderstands modern AI infrastructure economics. While GPUs do depreciate faster than traditional servers—some wearing out in accelerated timeframes—most AI computing infrastructure actually maintains useful economic life extending 15 to 20 years. Crucially, older GPU models don’t become worthless the moment newer chips arrive. Dated GPUs retain substantial value for inference tasks, where companies run pre-trained models for end users rather than training new models from scratch.

This overlooked reality means the total economic value extraction from AI hardware significantly exceeds what Burry’s framework assumes. The math simply doesn’t support his fraud allegation.

Cash Flow Explodes Even as Michael Burry Warns of Strain

Michael Burry’s second pillar rests on a cautionary tale about capital expenditure strain. He argues that massive CAPEX spending by hyperscalers will crush cash flows and create an unsustainable financial burden.

Yet the actual data moves in the opposite direction entirely. Alphabet’s cash from operations (trailing twelve months) has surged from under $100 billion to $164 billion as of 2026—precisely the period when AI infrastructure spending reached historic highs. This isn’t theoretical speculation; it’s documented financial performance.

Moreover, margins continue expanding dramatically across the industry. Companies that successfully deployed AI-trained infrastructure report returns exceeding $3 for every $1 invested. The newest wave—agentic AI systems that autonomously execute human-like tasks—appears poised to deliver even more efficiency, with early reports suggesting 25% or greater cost reductions in enterprise operations.

When the world’s largest tech companies are simultaneously scaling AI investments AND experiencing accelerating cash generation, it suggests Michael Burry’s doom scenario misses the operational reality.

Why Comparing NVIDIA to Cisco in 2000 Collapses Under Scrutiny

Michael Burry’s third argument invokes the dotcom crash by drawing a line between NVIDIA today and Cisco at its bubble peak in 2000. The comparison sounds superficially reasonable—both were infrastructure plays during periods of exuberance. The valuation evidence, however, demolishes this equivalence.

When Cisco peaked in March 2000, its price-to-earnings multiple exceeded 200x—an astronomical figure that reflected pure speculation untethered from earnings power. NVIDIA’s current P/E multiple sits at 47x. That’s conservative by historical tech standards, and it reflects actual earnings growth driving the stock price upward, not frothy valuation expansion.

The gap between a 200x and 47x multiple represents the difference between unsustainable bubble valuations and earnings-justified growth. Michael Burry’s historical parallel doesn’t account for this critical distinction.

GPU Scarcity Intensifies Amid Agentic AI Surge

The market doesn’t reward contrarian calls made against powerful structural tailwinds. The NVIDIA H100 GPU—the powerful data center processor powering AI model training—has seen rental prices balloon by approximately 17% since mid-December 2025. This surge reflects the relentless demand for compute capacity as agentic AI workloads accelerate.

Rental price appreciation directly signals ongoing GPU scarcity and robust underlying demand. This dynamics benefits not only NVIDIA but the entire AI infrastructure ecosystem including GPU cloud platforms, specialty energy solutions, and computing infrastructure providers.

Separately, companies like Bloom Energy—which solve the energy bottleneck constraining hyperscaler expansion—have attracted substantial bullish positioning from sophisticated traders. Major option positions and strong technical patterns suggest market participants expect continued infrastructure spending to drive demand for power solutions.

The Options Market Whispers What the Bull Case Believes

Sophisticated options traders don’t typically deploy $9 million positions without conviction. Late into the week of early March 2026, a major trader made exactly such a bet on NVIDIA March call options with a $205 strike price. Bloom Energy similarly attracted substantial call buying exceeding $1 million as traders positioned for continued strength.

These weren’t small retail positions—they represent the allocation decisions of experienced traders deploying serious capital. When combined with actual operating performance data, they suggest market professionals see further runway in AI infrastructure stocks despite Michael Burry’s warnings.

What Michael Burry’s Thesis Fundamentally Overlooks

Michael Burry built his reputation on identifying moments when market consensus collides with mathematical reality. His instinct for contrarian positioning remains sharp. However, his current AI bearishness appears to project 2000-era dynamics onto an entirely different technological and financial landscape.

The AI infrastructure buildout differs fundamentally from the dotcom era because it generates immediate, measurable returns. Hyperscalers aren’t burning cash on speculative bets—they’re harvesting $3+ of value from each dollar invested. GPU prices strengthen rather than collapse. Margins expand. Cash generation accelerates. These aren’t signs of an unsustainable bubble; they’re markers of sustainable infrastructure buildout.

Perhaps the lesson isn’t that Michael Burry has lost his analytical edge, but rather that successfully timing one crisis doesn’t guarantee predictive success during technological transitions. Even legendary investors occasionally confuse the echoes of past bubbles with present realities.