GPU Rental Data Undercuts Burry's $176 Billion Depreciation Warning
GPU rental prices for older NVIDIA chips are rising, not collapsing, undercutting Michael Burry's $176 billion depreciation-understatement thesis — but the accounting fight isn't settled.

Why it matters
- Michael Burry estimated AI companies could understate depreciation by $176 billion between 2026 and 2028 via extended server useful-life schedules.
- Silicon Data CEO Carmen Li said rental prices for certain older GPUs are rising rather than collapsing as newer generations arrive.
- NVIDIA shares closed at $162.85 on November 13, down 26.41% over the past month; Jensen Huang resigned as CEO and CFO Colette Kress was named interim chief.
- NVIDIA's fiscal Q2 revenue reached $96.221 billion, with Data Center revenue of $89.023 billion, up 117% year over year.
- NVIDIA supply commitments reached $279 billion, AI cloud guarantee obligations are capped at $108.5 billion, and days sales outstanding stretched to 60 days from 45.
Michael Burry estimated that AI companies could understate depreciation by $176 billion between 2026 and 2028 by stretching the assumed useful lives of servers and GPUs. New evidence from the GPU rental market now cuts against a core assumption behind that number: older chips are not collapsing in value. Some of their rental prices are rising.
The stakes land squarely on NVIDIA (NASDAQ:NVDA), whose shares closed at $162.85 on November 13, down 26.41% over the past month. The company is also navigating an abrupt leadership change: Jensen Huang resigned as CEO amid disclosures about his personal relationships, and the board appointed its finance chief, Colette Kress, interim CEO this week. Investors are asking whether AI capital spending can hold up without him. The Burry debate and the leadership shock converge on one question: how long does an AI chip actually stay productive?
Burry made his case on his Substack, Cassandra Unchained. He called the practice of extending server and GPU depreciation schedules "one of the more common frauds of the modern era." His target is accounting, not engineering. The question is whether reported earnings flatter reality.
By James Calloway
What is Burry's actual argument?
Companies that buy NVIDIA processors spread the cost of each chip over its estimated useful life. A longer schedule means a smaller yearly expense and higher reported profit. Microsoft (NASDAQ:MSFT) and Alphabet (NASDAQ:GOOGL) have moved certain server classes to four-to-six-year depreciation schedules. Burry argues that NVIDIA's 12-to-18-month product cycle shortens the real economic life of each generation, so today's reported earnings borrow from tomorrow's losses.
There is a nuance in his thesis that critics often skip. Burry's argument concerns accounting lives and reported earnings, and it leaves room for older GPUs to retain commercial value. A company can overstate profits on paper while its aging hardware still earns rent. The two claims are separable, and the rental data tests only one of them.
Takeaway: Burry attacked the earnings numbers, not necessarily the hardware's lifespan.
What does the rental data show?
In a Yahoo Finance interview with Deirdre Bosa, Silicon Data founder and CEO Carmen Li said older GPUs are holding their value better than expected, with rental prices for certain older chips rising rather than collapsing as newer generations arrive. Silicon Data publishes daily GPU benchmarks for on-demand rental rates, giving it a direct, market-based view of what compute actually costs.
That is a meaningful data source. Most AI infrastructure arguments rest on company filings or analyst models. Rental indexes reflect what buyers pay for compute today, chip generation by chip generation.
NVIDIA's own numbers point the same direction. On the August 26 earnings call, Kress said NVIDIA compute is "fully utilized across every cloud we serve" and that supply should remain a constraint "at least through the end of fiscal year '28." The company still ships Hopper parts, now two generations back.
The fiscal second-quarter release quantifies the demand behind those statements:
- Total revenue: $96.221 billion
- Data Center revenue: $89.023 billion, up 117% year over year
- Full utilization of compute across every cloud NVIDIA serves, per Kress
Takeaway: the secondary market for older chips is stronger than Burry's framework implied, at least for now.
Why do older chips keep earning?
Frontier model training grabs headlines, but it is not the whole demand base. Inference (running trained models for users), fine-tuning, research, and smaller workloads can all run economically on older hardware. Not every task needs the newest silicon.
Huang, speaking on the same call before his resignation, put a number on the demand shift. An AI agent, he said, needs 15 to 100 times the compute of a human user, and a deployed system "will be preserved and useful and productive for a lot longer." When demand runs ahead of supply, last-generation chips stay in service rather than getting retired.
Takeaway: inference-heavy workloads, not frontier training, are what keep old GPUs rented out.
Where does Burry still have a point?
Rental rates answer one question and leave several open. Rents measure current earning power. Resale value, book value, and lifetime profitability are separate questions that rising rents today do not resolve.
There is also a cyclical caveat. Today's pricing reflects a shortage, and shortages ease. The same supply squeeze is driving a broader expansion in power, cooling, and networking infrastructure across the industry.
The next product cycle could reset the rental market. Kress said the upcoming Vera Rubin platform delivers 35 times lower token costs relative to Grace Blackwell Ultra. Once that supply arrives in volume, demand for older chips could weaken, and the rental data that currently undercuts Burry could turn in his favor.
NVIDIA has also taken on more balance-sheet risk to keep the expansion moving:
- Supply commitments: $279 billion
- Guarantee obligations for AI cloud partners: capped at $108.5 billion
- Days sales outstanding: stretched to 60 days, from 45 days
That last figure deserves attention. When a company takes longer to collect revenue from customers, it can signal loosening credit terms or customer strain. Neither explanation is comforting at this scale.
Takeaway: strong rents under current shortage conditions prove little about lifetime chip economics.
What should investors watch next?
Burry may have underestimated how long older AI chips stay commercially productive. Rising rents and full utilization suggest the hardware earns money well into its life. But the accounting debate will be settled by data that has not arrived yet.
Three disclosures will matter most:
- Rental indexes: whether older-chip rents hold once Vera Rubin supply catches up with demand.
- Hyperscaler disclosures: any change to server useful-life estimates in Microsoft's and Alphabet's quarterly filings.
- NVIDIA's next report: results against guidance of $108.0 billion plus or minus 2% in revenue and gross margins of 74.0%, plus any comment on utilization under interim leadership.
The market is not waiting for that data neutrally. NVIDIA trades at a price-to-earnings ratio of 33, meaning investors already price in durable demand for AI compute. If rents fall sharply after the next product ramp, Burry's $176 billion warning gains weight. If they hold, the case for longer depreciation schedules gets stronger — and the accounting label he attached to the practice gets harder to defend.
Source: 247wallst.com
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
604 articles
Related articles
- Michael Burry Says Nvidia's Real Problem Is a $176 Billion Accounting Gap
- Burry: Nvidia's Real Risk Is $3 Trillion in AI Spending, Not Huang
- Super Micro Falls 7% as NVIDIA Leadership Shock Drowns Out Gartner's $957 Billion Forecast
- Jensen Huang Resigns as NVIDIA CEO, Congress Seeks Testimony
- Burry Called Nvidia's Key-Man Risk Hours Before Jensen Huang Resigned