Michael Burry Warns on Nvidia: Bull vs. Bear Case for the AI Chip Boom

The AI trade is entering a more demanding phase. For the past three years, the bullish thesis was relatively simple: hyperscalers needed more compute, Nvidia dominated AI accelerators, and larger models required ever-bigger clusters. That logic still has strong support, but investors now face a harder question: can AI revenue and productivity gains keep pace with the extraordinary capital spending required to build the infrastructure?

Michael Burry’s criticism of Nvidia’s new $500 billion AI financing initiative fits into this broader debate. He has called the plan a “sign of desperation,” questioning whether AI demand is becoming too dependent on financing. Nvidia, meanwhile, argues that partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR can unlock more than $500 billion of third-party capital for AI infrastructure.

The real bull-versus-bear debate is broader than financing. It comes down to hyperscaler spending, AI monetization, chip replacement cycles, competition and returns on capital.

The Bull Case: AI Demand Is Still Expanding

The strongest bullish argument is straightforward: the spending cycle has yet to break.

Nvidia’s fiscal first-quarter 2027 revenue reached $81.6 billion, up 85% year over year, while Data Center revenue rose 92% to $75.2 billion. The company guided to roughly $91 billion of revenue for the following quarter, even without assuming China Data Center compute revenue.

Its largest customers are also accelerating investment. Alphabet raised its 2026 capex outlook to $195–205 billion. Meta expects $130–145 billion. Microsoft spent $41 billion in its latest quarter, with around two-thirds directed toward shorter-lived assets such as CPUs and GPUs.

More importantly, AI infrastructure is beginning to generate real revenue. Google Cloud revenue surged 82% in the second quarter to $24.8 billion, while backlog reached $514 billion. Microsoft’s cloud revenue surpassed $214 billion for fiscal 2026, with demand increasingly broadening beyond frontier-model companies.

That matters because the next stage of AI growth may come from inference, agents, coding, advertising, search, industrial AI and enterprise deployment rather than frontier training alone. If AI becomes embedded in everyday production workloads, semiconductor demand could remain structurally high for years.

Rubin provides another catalyst. Nvidia says the platform is already in production, with systems arriving in the second half of 2026. AWS, Google Cloud, Microsoft, Oracle and CoreWeave are among the expected early adopters.

The opportunity also extends beyond Nvidia. Larger AI clusters require more HBM memory, advanced packaging, networking, optical connectivity, power infrastructure and cooling. Even if Nvidia eventually captures a smaller share of each incremental AI dollar, the overall AI infrastructure market can continue expanding.

The Bear Case: Can Monetization Catch Up?

Bears can look at the same capex numbers and reach a very different conclusion.

Alphabet spent $44.9 billion on capex in the second quarter and generated negative $5.9 billion of free cash flow. Meta spent $31.1 billion and produced only $784 million of quarterly free cash flow. Microsoft has also acknowledged that AI infrastructure investment is pressuring cloud margins.

The key question is therefore simple:

How much incremental AI revenue will each additional dollar of capex generate?

The boom can continue while hyperscalers believe compute remains scarce. The market becomes more vulnerable once capacity catches up and companies begin demanding stronger returns from each new data center.

Technological depreciation creates another risk. Nvidia is moving rapidly through Hopper, Blackwell, Blackwell Ultra and Rubin. Older GPUs can still generate revenue through inference, fine-tuning and lower-cost cloud workloads, but their economic value may fall quickly as new architectures improve performance per watt and cost per token.

This is where Burry’s financing criticism becomes relevant. Nvidia’s new Wall Street partnerships aim to unlock more than $500 billion of third-party funding. His bearish interpretation is that increasingly sophisticated financing may be needed to sustain a spending cycle whose long-term returns remain uncertain.

The real danger would emerge if AI demand weakened at the same time that older GPU rental rates and residual values fell sharply. Highly leveraged AI cloud operators would then face weaker cash flow and weaker collateral simultaneously.

Competition adds another pressure point. Google is expanding TPU deployment beyond purely internal use, while AMD and custom accelerators backed by companies such as Broadcom are targeting workloads where Nvidia’s premium economics may become harder to justify.

Nvidia can still grow in that environment, but market share, pricing power and semiconductor margins could gradually normalize.

What Decides the Next AI Trade?

The next phase of the AI chip trade comes down to four variables:

AI revenue growth, hyperscaler capex, GPU utilization and semiconductor margins.

The bullish path remains intact if AI applications keep spreading, inference expands, hyperscalers remain capacity-constrained, older GPUs find profitable lower-tier uses and Rubin launches another upgrade cycle.

The bearish path begins when capex grows materially faster than monetization, cloud margins compress, older GPU economics deteriorate and hyperscalers finally slow infrastructure spending.

The central question is larger than Nvidia’s $500 billion financing plan:

Can AI generate enough economic value to justify the infrastructure already being built — and the even larger wave of investment still coming?

If the answer remains yes, the AI semiconductor cycle still has room to run. If that relationship begins to break, investors may need to reassess the entire chain from Nvidia and memory to networking, foundries and AI cloud operators.