Michael Burry, the famous American investor and hedge fund manager who predicted the 2008 subprime mortgage crisis, has now issued a warning on the AI industry.
vves
Reacting to the details in Anthropic’s S-1 filing, he wrote that the AI industry is “catastrophically vulnerable” and remains openly bearish. So, what is making Burry go against the market’s overwhelmingly positive view of AI, which has driven Anthropic’s reported valuation expectations to upwards of $2 trillion? And are the financial risks emerging in Anthropic’s business model a warning sign for the wider AI industry? Let’s deep dive.
Michael Burry’s Three Red Flags in Anthropic’s Business Model
Infrastructure Commitments
As per Burry’s estimate, Anthropic has committed around $537 billion towards computing and equipment capacity across its technology partners. However, the actual commitments are much higher. This includes $125 billion to Broadcom, $114.5 billion to Google, and $110 billion to Amazon.
The capital expenditures needed just to train, operate, and maintain foundational AI models are much higher than the actual cash generated by the business. Based on his estimate, the obligations are more than 27 times trailing 12-month revenue of $19.7 billion as of June 30, 2026. Burry views it as unsustainable for growth. It burdens the company with massive infrastructure debt and makes it highly sensitive to any macroeconomic tightening or revenue misses.
The Efficiency Paradox
Burry’s another big claim is that Anthropic may not be able to sustain its current growth momentum. In his view, the very factor that is driving the growth today could hurt future growth: AI efficiency.
Enterprise customers and developers are becoming more familiar with AI tools. They are learning to get the same results with fewer tokens by optimising their prompts and workflows. The models themselves are also becoming more efficient, reducing the computing power needed to get the task done.
While this is positive for customers, Burry believes this could work against Anthropic’s growth. If users can get the same output at a lower cost, their spending on AI services will also decline even as usage increases. In order to make up for the fall in revenue per unit of usage, Anthropic will have to keep on expanding its customer base and create more workloads.
Circular Flow of Money
Burry’s biggest concern is the AI ecosystem’s interconnected financial relationships. The same companies that have invested billions into Anthropic also supply the computing infrastructure and sell the Claude AI models via their cloud marketplaces. At the same time, they also compete with Anthropic through their own AI models.
For example, while Amazon has pledged up to $25 billion in investment, Anthropic has committed to spend at least $110 billion on Amazon Web Services (AWS) over the next decade, regardless of actual usage. Similarly, it has committed hundreds of billions to Google Cloud.
This circular flow of money exposes the fragile reality of Anthropic’s business model. It absorbs the immediate financial impact while the cloud providers capitalize on their GPU purchases as long-term assets. This means Anthropic has a balance sheet weighed down by vendor lock-in and non-cancellable infrastructure debt, and just can’t walk away.
Burry sees this interdependence as a potential financial vulnerability. If demand for enterprise AI weakens or credit access tightens, Anthropic will struggle to meet its infrastructure commitments. If so, it would impact cloud providers, chipmakers, and lenders that have based their growth expectations on sustained AI.
Why Investors Shouldn’t Ignore Burry’s Warning
There is a historical precedent for Burry’s concerns about Anthropic’s massive infrastructure commitments. The telecom boom of the late 1990s offers a glimpse into what happens when companies build infrastructure based on expectations of future demand rather than the cash flows they can generate today.
The 2000 Telecom Bust: A Warning from History
After the passage of the Telecommunications Act of 1996 in the US, the telecom industry became highly convinced that internet traffic would double every few months. Companies and investors started pouring billions of dollars into laying fibre optic cables across land and under oceans. Much of this expansion was funded through borrowed money.
But two things went wrong. Internet demand failed to grow at the expected rate, while technological improvements allowed each fiber to carry far more data, so capacity grew much faster than demand. This resulted in a crash in bandwidth prices. Amid all this, telecom companies found themselves saddled with debt and infrastructure they could not monetise profitably.
Global Crossing, which was founded in 1997, spent billions laying undersea cable and reached a peak valuation of $40 billion in 2000, but suddenly found its revenue could no longer support debt. Eventually, the company filed for bankruptcy in 2002. Another company, WorldCom, met the same fate.
The Difference Between Technology Adoption and Financial Returns
Even after the bankruptcy and dot-com bust, fiber optics remain critical to today’s modern internet economy. But the companies that invested heavily in building them faced very different outcomes, depending on their debt levels and financial strength and their ability to generate returns from the infrastructure.
The same distinction is true of the AI industry today. Increased demand for AI doesn’t mean all infrastructure investments will generate adequate returns. The pace of adoption, utilisation of computing capacity, infrastructure costs, and the ability to monetise AI services will all affect the profitability of companies such as Anthropic.
The Margin Challenge for AI Companies
AI companies lack the pricing power of traditional software companies. A SaaS company can create a product once and then sell it to many more customers at relatively low marginal costs. Gross margins on this model are often 80% or greater. Generative AI has a different cost structure. Every prompt and response requires computing resources, and training and improving models requires more investment in infrastructure.
This makes growth more capital-intensive. Aswath Damodaran’s analysis of Anthropic’s potential IPO assumes Anthropic will need another $1 of capital for every $3 of incremental revenue. Even this would translate to significant annual reinvestment requirements.
For Anthropic, this means that more customers and usage don’t necessarily translate to more profitability. A large portion of its revenue has to be spent on buying computing power from partners such as Amazon, Google, and Microsoft.
For investors, the key question is whether Anthropic can create a technology advantage that enables it to charge premium prices, use computing power more efficiently, and retain a greater share of the value it creates. The company’s long-term growth will depend on how well it converts that growth into returns for shareholders.