Michael Burry (who in your head probably looks like Christian Bale thanks to The Big Short), the investor who famously predicted the 2008 housing crash, has launched a Substack newsletter after deregistering his hedge fund. The $379 annual subscription capitalizes on the 1.6 million followers he’s built on X, offering what he describes as his “sole focus” going forward.

The newsletter’s inaugural post takes (which he kindly enough made accessible for free as a Thanksgiving gift today) readers back to 1999, when Burry was a 27-year-old neurology resident at Stanford making $33,000 annually while carrying $150,000 in medical school debt. There he wrote his Valuestocks.net article “Buffett Revisited”. A fellow resident casually mentioned making $1.5 million on Polycom stock. Physicians crowded around terminals checking stocks while patients waited. In that environment, Burry was writing investment analysis late at night, getting paid $1 per word by MSN Money under the pen name “Value Doc.” His VSN Fund returned 68.1% in 1999, and by February 2000, the San Francisco Chronicle noted he had shorted Amazon. Fourteen days after that article appeared, the NASDAQ topped. It was a peak it wouldn’t revisit for 15 years.

Burry’s approach today is notably personal and reflective. His analysis of Apple in 1999 exemplifies his contrarian thinking. He bought it for the VSN Portfolio despite pushback, writing that great companies like Coca-Cola, American Express, and Disney had all experienced 11-year periods of negative real returns. Unlike the skeptics who simply dismissed the internet as a fad in 1999, Burry recognized the technology was transformational; he just believed the infrastructure was being overbuilt relative to near-term demand. He’s making the same argument about AI today. He believes markets are deep in bubble territory, drawing parallels between the late 1990s tech mania and today’s AI boom. His X post’s often echoed familiar warnings:

Feb 21, 2000: SF Chronicle says I’m short Amazon. Greenspan 2005: ‘bubble in home prices … does not appear likely.’ Powell ‘25: ‘AI companies actually… are profitable… it’s a different thing.’

The comparison is deliberate. Burry highlighted then-Fed Chair Alan Greenspan’s 2005 insistence that U.S. housing prices showed no signs of a bubble. This was just two years before the subprime implosion validated Burry’s famous “Big Short.” Today, he’s openly bearish on AI poster children Nvidia and Palantir, suggesting history is rhyming once again. Shortly after the newsletter was out, Nvidia circulated a seven-page memo to Wall Street analysts explicitly naming Burry in its opening, a rare move for a company of Nvidia’s stature. The memo sought to refute his claims about stock-based compensation, depreciation schedules, and what he calls “circular financing.”

Mainly, Nvidia disputed Burry’s depreciation argument. Burry contends that customers overstate GPU useful lives to justify massive capex, claiming the hardware becomes obsolete in two to three years. Nvidia counters that its A100s, released in 2020, continue running at high utilization rates with “meaningful economic value” well beyond that timeframe, justifying the standard four-to-six-year depreciation schedule. The memo also rejected suggestions of “circular financing,” noting that Nvidia’s strategic investments represent a small fraction of revenue and that AI startups raise capital predominantly from outside investors. Burry responded on Substack:

I stand by my analysis. I am not claiming Nvidia is Enron. It is clearly Cisco.

He argues Nvidia now occupies the exact position Cisco held in 1999-2000. It’s the key hardware supplier powering a massive capital investment cycle built on optimistic demand forecasts. Just as telecom companies spent tens of billions laying fiber optic cable based on projections that “internet traffic doubles every 100 days,” today’s hyperscalers are promising nearly $3 trillion in AI infrastructure spending over the next three years. The problem? In the early 2000s, less than 5% of U.S. fiber capacity was operational. Burry believes today’s AI buildout rests on similarly flawed assumptions about data center power and GPU longevity. “And once again there is a Cisco at the center of it all, with the picks and shovels for all and the expansive vision to go with it. Its name is Nvidia,” Burry wrote. The analogy might resonate with market observers who remember how that story ended. Cisco’s stock peaked above $80 in March 2000. It wouldn’t return to that level for nearly 24 years. The company survived and remained profitable, but shareholders who bought at the top experienced a generational loss. One key difference is worth noting: Cisco’s forward P/E in 2000 was around 200; Nvidia’s is under 40.

In my opinion the technical argument around depreciation matters more than it might appear. If hyperscalers must depreciate GPUs over three years instead of six, companies like Alphabet would see roughly a 10% hit to net profit. More importantly, it would signal that the economic returns on AI infrastructure spending are weaker than advertised. Alphabet, for example, is currently guiding $90 billion-plus of AI spending this year. Using 5-year straight line depreciation, you get $18 billion per year in expenses. Add $9 billion for a conservative 10% WACC. That’s $27 billion, and assuming a 70% blended margin, you need about $40 billion per year in incremental revenue directly attributable to AI to make the infrastructure spending justifiable. Therefore, the question remains: how much of Alphabet’s $60 billion annualized revenue increase is actually attributable to AI versus normal growth? Burry closes his first newsletter with characteristic understatement:

I doubted if I should ever come back. I’m back.