To The Moon

$NVIDIA(NVDA)$

Few clashes in modern financial discourse attract as much attention as when Michael Burry—the legendary investor behind The Big Short—takes aim at a market darling. His latest critique targets the crown jewel of the AI boom: NVIDIA. But unlike past macro warnings, Burry’s concern this time is more technical, more nuanced, and arguably more provocative. He’s challenging the accounting foundations of the AI investment cycle, especially the depreciation policies of hyperscalers whose AI infrastructure spending drives NVIDIA’s explosive revenue growth.

NVIDIA, in an unusual public move, has responded.

This sets the stage for one of the most consequential debates in the current market cycle: Are AI accounting practices a real risk to NVIDIA and the broader AI trade, or is this an overextended fear that misses how modern AI businesses operate?

This article dives deep into the accounting mechanics, the business incentives, NVIDIA’s responses, and the broader market implications—ultimately arriving at a balanced verdict on whether Burry’s concerns warrant investor action.

I. The Origins of the Debate: What Exactly Is Michael Burry Warning About?

Michael Burry’s critiques have always leaned toward structural vulnerabilities—housing leverage before 2008, passive indexing distortions in the late 2010s, and speculative tech cycles in recent years. His newest focus is AI accounting, specifically how:

1. Hyperscalers recognize depreciation on AI hardware, such as NVIDIA GPUs.

2. These policies influence reported earnings, cash flows, and demand sustainability.

3. The broader AI ecosystem may be misleading investors regarding the true profitability of AI-driven growth.

Burry’s Core Claim

Burry argues that AI-related capex may be overstated in current profit calculations, creating a mismatch between:

• the useful life of AI chips (often shorter due to rapid obsolescence), and

• the depreciation schedules used by companies to expense these chips (often longer than realistic).

If hardware becomes obsolete faster than expected—due to NVIDIA’s rapid innovation cycles (H100 → H200 → B100 → B200…)—then hyperscalers may be understating expenses, thereby overstating margins.

This, in turn, would imply that:

• AI-driven earnings may be inflated,

• return on invested capital may be misrepresented, and

• demand for AI chips may not be as durable as it appears.

In Burry’s framing, hyperscalers may be pulling forward capex and smoothing losses using overly favorable accounting assumptions.

Why This Matters for NVIDIA

Hyperscalers (Microsoft, Amazon, Google, Meta) combined represent 70–80% of NVIDIA’s data center revenue. If their profitability is inflated— if depreciation costs are understated— if reported returns on AI investments are rosier than reality—

—then the demand curve for NVIDIA’s chips could weaken, potentially sharply.

This is why the debate has exploded.

II. Understanding the Accounting Question: Depreciation, Useful Life, and AI Hardware

To evaluate the legitimacy of Burry’s criticism, one must understand a central accounting principle: depreciation.

What Is Depreciation?

Depreciation is the method companies use to expense long-term assets—such as servers, data center hardware, and GPUs—over their useful life.

If an AI server costs $250,000 and is depreciated over 5 years:

• Annual expense = $50,000

• Earnings look smooth and stable

But if the asset is actually obsolete after 2 years:

• Correct annual expense should be = $125,000

• That means reported profits were artificially higher in years 1 and 2.

This is the core of Burry’s suspicion: AI hardware loses economic value much faster than companies claim.

The Rapid Obsolescence Issue

NVIDIA’s release cadence has accelerated:

• A100 (2020)

• H100 (2022)

• H200 (2024)

• B100 (2025)

• B200 (2026 expected)

Every generation is ~2× faster, and often far more power-efficient.

Hyperscalers may be depreciating GPUs over 5 years, but the economic usefulness may be 2–3 years at best. That gap can meaningfully distort profits.

Is This New?

No. High-speed obsolescence has existed in:

• smartphones

• CPUs

• networking equipment

• telecom hardware

• semiconductor fabs

But AI capex—unlike smartphones—is measured in tens of billions, which magnifies discrepancies.

III. NVIDIA Responds: What Did They Actually Say?

NVIDIA rarely comments on investor debates, but the intensity of Burry’s claims—and the global press coverage—prompted a public response.

Their key points:

1. Hyperscalers set their own depreciation schedules

NVIDIA emphasized that they do not dictate or influence how hyperscalers depreciate hardware. These schedules are determined by:

• internal accounting teams

• auditors

• industry standards

• technical assessments

This point distances NVIDIA from any claim of manipulating earnings through client accounting practices.

2. Replacement cycles differ from depreciation cycles

NVIDIA explained that companies often replace hardware faster than they depreciate it, especially when the performance gains justify upgrades.

This is not financial manipulation. It is operational necessity.

For example:

• Data centers may deploy new GPUs every 18–36 months

• But depreciate old hardware over 4–5 years

• The old hardware remains in secondary workloads or is resold

This isn’t unique to AI—Amazon does this with servers; Apple does it with retail hardware; telecom firms do it with networking gear.

3. Total cost of ownership (TCO) justifies faster upgrades

NVIDIA argues that newer chips’ energy efficiency and performance gains reduce operating costs enough that upgrading early makes financial sense—even if the accounting depreciation is slower.

4. Demand is driven by ROI, not accounting

The company insists that AI workloads generate real economic returns for hyperscalers and their customers. Thus, accounting choices do not meaningfully inflate the business case for AI adoption.

IV. Does Burry Have a Point? A Balanced Evaluation

The truth lies somewhere between alarmism and dismissal.

Let’s examine the validity of each side.

A. Where Burry Is Right

1. AI hardware obsolescence is real and accelerating

This is undeniable. NVIDIA’s pace of innovation compresses product cycles, meaning that:

• older GPUs lose competitiveness faster

• data centers face pressure to upgrade

• useful life is often shorter than accounting life

This can distort margin reporting for hyperscalers.

2. Depreciation can materially influence reported earnings

If a hyperscaler underestimates depreciation expense by billions:

• operating margins look inflated

• ROIC looks higher

• AI investments look more profitable than they are

This could create overconfidence in AI’s near-term economic returns.

3. Capex cycles can reverse

Even if AI is the biggest technology wave in decades:

• spending could normalize

• growth could flatten

• demand could become cyclical

NVIDIA’s current growth rate (50–200% YoY in some quarters) is not perpetual.

4. Investors tend to overlook accounting subtleties during boom cycles

Just as they did:

• in telecom (1998–2001)

• in cloud infrastructure (2010–2015)

• in shale oil (2014–2018)

Burry’s skepticism is consistent with past cycles that showed massive front-loaded investment followed by years of digestion.

B. Where Burry Overreaches

1. Depreciation does not determine cash flow

Even if depreciation is understated:

• cash outlays already occurred

• GPU purchases are paid upfront

• demand is a function of real workloads, not accounting treatment

AI adoption is driven by:

• LLM training

• inference demand

• enterprise AI adoption

• cloud AI services

• personalized models

• agentic systems

• advanced robotics

Depreciation policies do not change this underlying need.

2. Companies routinely depreciate assets beyond their replacement cycle

This is standard practice across:

• airlines (planes replaced before end of depreciation)

• data centers (servers often replaced early)

• hospitality (furnishings upgraded early)

It is not a red flag unless:

• resale value is zero

• assets truly become worthless

• upgrades yield no incremental economic return

AI workloads do not fit these scenarios.

3. Secondary markets for GPUs are growing

NVIDIA GPUs hold significant value even after being replaced. Enterprises, startups, universities, and research labs buy older chips.

This mitigates depreciation concerns.

4. The AI capex cycle is not solely hype-driven

Real revenue is being generated:

• Microsoft AI services

• Meta AI-driven engagement

• Amazon AWS Trainium/Inferentia

• Google Gemini Enterprise

• Adobe, Snowflake, ServiceNow AI upgrades

• AI in drug discovery, robotics, design, simulation

Unlike past bubbles, AI has clear economic use cases.

5. Hyperscalers’ internal returns justify continued investment

AWS, Azure, and Google Cloud are already monetizing:

• AI acceleration services

• AI training clusters

• AI inference APIs

• enterprise AI subscriptions

Some analysts estimate 20–40% internal IRR on incremental AI investments—far higher than typical tech capex.

Accounting adjustments barely influence those ROI calculations.

V. The Bigger Picture: What’s Actually at Stake?

This debate ultimately reflects two very different worldviews:

Burry’s Worldview: Cycles Mean-Revert

• Systems get overbuilt

• Accounting masks true economics

• Hype inflates expectations

• Capex peaks and collapses

• Leaders face aggressive valuation compression

• Investors underestimate downside tail risk

In this model, AI will follow the pattern of:

• telecom fiber

• shale oil

• 3D printing

• blockchain hardware

• dot-com infrastructure

The narrative: Overinvestment always ends in correction.

NVIDIA’s Worldview: AI Is the New Industrial Revolution

• Productivity gains justify continual upgrades

• Each generation unlocks entirely new markets

• GPU demand scales with model size & usage

• AI becomes the operating layer of global computing

• Capex is not cyclical but secular

• Depreciation debates are accounting noise compared to exponential real demand

This is a “compute as electricity” model: You don’t stop upgrading transformers in an electrified economy.

VI. The Investment Lens: How Should Investors Interpret This?

1. The risk Burry highlights is real—but slow-moving

Even if depreciation is overstated:

• the unwind takes years

• hyperscalers have massive balance sheets

• AI workloads are accelerating

• secondary GPU markets will absorb hardware

• cash flows remain strong

This is not a crash catalyst—it's a margin-adjustment debate.

2. NVIDIA’s growth could normalize, but normalization ≠ collapse

Today’s 40–80% YoY data center growth is unsustainable long-term, but a transition to:

• 15–25% steady growth

• expanding TAM

• dominance in inference

• expansion into automotive, robotics, simulation

• high software margins

is still extremely bullish compared to almost any other megacap.

3. The debate is more relevant for hyperscalers than NVIDIA

If anyone faces accounting pressure, it’s:

• Microsoft Azure

• Google Cloud

• AWS

Not the GPU supplier.

4. For NVIDIA investors, the real question is valuation, not depreciation

NVIDIA trades at a premium because:

• revenue is exploding

• margins are expanding

• demand outstrips supply

• the AI narrative is strong

• long-term adoption is undeniable

The bear case is not about depreciation—it’s about whether the market is pricing perfection.

5. Market psychology matters

Burry’s warnings resonate because:

• NVIDIA is the most crowded trade on Earth

• AI expectations are sky-high

• investors fear missing hidden risks

His criticism might not be a catalyst, but it is a sentiment dampener.

VII. Final Verdict: Does the Big Short Have a Case Against NVIDIA?

Bottom Line:

Burry is directionally insightful—but operationally overstated.

Here’s the final assessment:

✔️ Legitimate Concerns

• GPU useful life may indeed be shorter than depreciation schedules

• Hyperscalers may be overstating AI profitability

• AI capex will eventually face normalization

• Obsolescence risk is real

✘ Overstated Implications

• This does not materially undermine NVIDIA’s revenue

• Demand is driven by workloads, not accounting

• Cash flow is unaffected by depreciation

• Secondary markets support hardware value

• AI is not a cyclical fad—it's a technological foundation shift

NVIDIA’s risk is not accounting—it’s valuation and market psychology.

If sentiment shifts, if capex normalizes, or if growth slows meaningfully, the stock could experience compression. But not because hyperscalers depreciate GPUs too slowly.

VIII. Key Takeaways for Investors

1. Burry raised an intellectually valid accounting issue, but not a thesis-breaking one.

2. Depreciation differences do not change NVIDIA’s cash inflows or demand drivers.

3. AI hardware cycles are fast—but so are the adoption curves that justify upgrades.

4. Hyperscaler profitability may be overstated, but the AI business case remains strong.

5. NVIDIA’s long-term story remains intact—but the stock’s valuation leaves no room for error.

6. The real risk for NVIDIA investors is future growth expectations, not accounting adjustments.

7. Expect more volatility as the “AI bubble vs. AI revolution” debate continues.