Was Michael Burry Right About AI Stocks for the Wrong Reason?

Last November, Michael Burry published a pointed critique arguing that major AI players were artificially inflating profits by stretching depreciation schedules on GPUs and servers far beyond their real-world usefulness. In his view, hyperscalers are treating cutting-edge chips — designed by Nvidia for 12- to 18-month obsolescence cycles — as if they would last four to six years. 

This accounting maneuver, Burry claimed, understated expenses by up to $176 billion between 2026 and 2028, boosting sector earnings by roughly 20% and creating a temporary tailwind that would reverse sharply once the hype cooled.

Both Nvidia (NASDAQ:NVDA | NVDA Price Prediction) CEO Jensen Huang and, just days ago, CoreWeave (NASDAQ:CRWV) CEO Michael Intrator have pushed back hard. Huang has repeatedly defended the durability of AI infrastructure, while CoreWeave’s chief executive said that the 18-month obsolescence narrative ignores reality: most customer contracts run five years, and older GPUs like A100 and H100 units still command 95% of their original value as new use cases emerge. 

Are Nvidia and CoreWeave actually right about depreciation timelines — or is Burry still correct overall that the AI profit story rests on shaky foundations?

Burry’s Big Short Delivers Early Pain

When Burry first disclosed sizable short positions in Nvidia and Palantir Technologies (NYSE:PLTR) through Scion Capital in late 2025, the market barely flinched. Four months later, the picture looks very different. Nvidia’s shares have slipped 14% and Palantir is down 24%. CoreWeave — the specialized AI-cloud provider that has become a poster child for hyperscale GPU demand — has tumbled roughly 40%.

To be fair, Burry’s critique is only one factor. Broader concerns about AI capex sustainability, slowing hyperscaler spending, and valuation compression have weighed on the sector. Yet his public broadside has undeniably amplified selling pressure. Short interest has ticked higher across the names he targeted, and every fresh whisper of margin pressure or delayed AI ROI now gets filtered through the “Burry was right” lens. 

CoreWeave Strikes Back

CoreWeave’s CEO emphasized that enterprise buyers do not treat GPUs like disposable smartphones. Five-year commitments dominate the order book, and secondary markets for last-generation hardware remain robust. “Older GPUs still retain value as new use cases emerge,” he noted, directly challenging Burry’s planned-obsolescence thesis. The argument aligns with Nvidia’s long-standing position that AI workloads are evolving so rapidly that hardware stays relevant far longer than skeptics assume.

This defense carries weight for CoreWeave in particular. As a leading GPU-cloud operator, CoreWeave’s business model depends on high utilization rates and predictable cash flows from long-term leases. If GPUs truly hold value across multiple generations, CoreWeave’s fleet retains book value and supports higher margins than a strict two-to-three-year depreciation cycle would allow. The company’s bullish stance also reassures hyperscaler partners that their massive capex is not about to trigger a wave of impairment charges.

Whose Argument Carries Greater Weight?

Burry’s case is focused on accounting: even if demand holds, the timing of expense recognition matters. Extending useful lives from three to six years front-loads profits today while back-loading costs tomorrow. If the music stops in 2026 or 2027, hyperscalers could face simultaneous writedowns that erase much of the incremental AI earnings investors are currently celebrating. Burry compares the setup to Cisco’s (NASDAQ:CSCO) dot-com-era overcapacity, where explosive growth masked an eventual demand cliff.

Nvidia and CoreWeave counter with operational reality: customer behavior and secondary-market pricing prove GPUs are not obsolete on Nvidia’s 18-month design cadence. Long-term contracts and emerging applications (inference, edge AI, enterprise fine-tuning) keep older silicon productive. If true, the depreciation schedules are not fraudulent — they reflect genuine economic lives.

Key Takeaways

The greater weight probably tilts toward the tech executives in the near term. Real-world utilization data and contract backlogs provide tangible evidence that Burry’s obsolescence timeline may be too aggressive. Yet Burry’s broader point about borrowed profitability remains relevant. 

Even if GPUs last longer, the sheer scale of industry capex — nearly $3 trillion across hyperscalers — still requires end-user ROI that has yet to materialize at promised levels. Burry may have been right about betting against AI stocks, at least so far, even if his depreciation timing proves off. Nvidia, Palantir, and especially CoreWeave have all declined materially since his disclosure, validating the short-term pressure he highlighted. 

The real test lies ahead: if CoreWeave and Nvidia are correct, AI infrastructure demand will absorb the capex glut and render depreciation debates academic. If Burry’s Cisco analogy holds, a painful normalization awaits regardless of accounting tweaks.