Michael Burry Is Shorting AI. Here's What He Gets Right - And What He Misses
The man who predicted 2008 now has 80% of his portfolio in puts on Nvidia and Palantir. His AI bubble thesis is worth taking seriously. It's also not the full picture.
What Burry gets right about the AI bubble - and what builders should actually care about.
Michael Burry doesn't do interviews. He posts cryptic charts on X, closes his hedge fund, launches a Substack, and puts 80% of his portfolio into short positions. If you know who he is, you take notice. If you're building AI products, you should read his thesis carefully - and then read this counterpoint.
What Burry is actually saying
In his Cassandra Unchained newsletter (launched November 2025, 20,000+ paid subscribers in its first day), Burry laid out three core claims about the AI market:
Microsoft, Google, and Amazon are spending $200B+ on AI data centres. The revenue that justifies this capex hasn't arrived yet. Burry draws a direct parallel to the fiber-optic overbuild of 1998–2001.
Palantir, he argues, is charging premium valuations for what is fundamentally consulting work dressed up as AI. The model is a services business, not a software business - and services businesses don't get 100x revenue multiples.
Any competition from AMD, custom silicon from hyperscalers (Google TPU, AWS Trainium), or from open models that need less compute to be useful would dent Nvidia's moat. At its current multiple, the market is pricing in no meaningful competition for 5+ years.
What he gets right
The infrastructure argument is hard to dismiss. Data centre capex has grown faster than enterprise AI software revenue in every quarter since 2023. The ROI on AI infrastructure is not yet visible in most company P&Ls. That's not fraud - it's a time lag. But it's a time lag that can be very expensive if interest rates move or if a major customer pauses their AI spending.
The consulting-as-AI observation about Palantir is also directionally correct. Their AIP platform charges significant professional services fees on top of software licensing. That's a recurring revenue model - but it's not the SaaS multiple investors are pricing in.
What he misses
Burry's framework is built on equity valuations - which is his job. But the question of whether AI is a bubble is separate from the question of whether AI is useful. The Dotcom bust didn't mean the internet was a bad idea. It meant companies like Pets.com were incorrectly valued relative to companies like Amazon.
The evidence for real AI adoption is strong at the product layer, even if the infrastructure layer is overbuilt. Cursor went from zero to $3B ARR in two years with genuine enterprise contracts at 64% of Fortune 500 companies. That's not hype - that's software being used. GitHub Copilot, Claude API, OpenAI API - all of these have real, paying enterprise customers doing real work.
The better framing: there may be a bubble in AI infrastructure stocks. There is not a bubble in AI adoption at the product level. If you're a business deploying AI to automate bookings, document processing, or customer support - that ROI is real and it's happening now regardless of what Nvidia's stock does.
DestinPQ's take
Burry might be right about Nvidia's multiple. He's not right that AI is a bad idea. The businesses we work with - clinics, restaurants, franchise chains - are seeing measurable ROI from AI agents. The product-layer adoption is real. The infrastructure stock valuations might not be. These are different questions.
Frequently asked questions
Should I pause my AI investment because of Burry's prediction?
No - unless your "AI investment" is buying Nvidia stock. Building AI into your business operations is a different decision from speculating on AI infrastructure stocks. The ROI calculation for deploying an AI agent to automate customer bookings doesn't change based on what NVDA trades at.
Has Burry been right about previous tech calls?
His 2008 housing call was correct and early. His 2022 "everything bubble" prediction didn't play out cleanly - markets did correct but recovered faster than his position allowed. Being right in direction and right in timing are different things. Burry can be directionally right about AI infrastructure and still lose money on the specific trade.
Building AI for your business - not speculating on it
Bubble or not, AI that reduces your operational costs pays for itself. We ship production agents in 10 days.
Builds AI-powered products for real businesses. Writes about practical AI adoption, model costs, and what actually works in production.
All posts →