logo

Michael Burry AI bubble: His Next Big Short

Michael Burry AI bubble: His Next Big Short
72

Michael Burry AI bubble warnings should not be read as a simple anti-technology rant. The stronger version of his argument is more specific. AI may be real, useful, and widely adopted, while many AI stocks still price in returns that have not been proven.

That distinction matters. A bubble does not require fake technology. The dot-com bubble was built around the internet, which changed the world. The loss came from paying prices that assumed every network, portal, and fiber project would earn durable cash flows.

Burry's latest warning sits in that same frame. A readable syndicated version of CNBC's inside-look analysis, plus Fortune and Investopedia coverage, points to bearish exposure tied to Nvidia and Palantir. It also points to a public argument that today's market feels like the late stage of the 1999-2000 bubble. The key question is not whether AI works. It is whether capital spending, margins, chip lifespans, and stock valuations all work at the same time.

For broader technology context, Wired tracks the infrastructure and product cycle behind the AI boom. Burry's case is narrower: whether those advances can justify current market economics.

Risk disclaimer: this article is analysis, not investment advice. It maps the thesis, the countercase, and the signals that would move the probability of a Michael Burry AI bubble call higher or lower.

Last reviewed: August 27, 2026.

What Burry Is Actually Betting Against

Burry is best known for identifying stress in the mortgage market before the 2008 financial crisis. That history gives every new warning more attention than a normal short thesis would receive. However, it also creates a risk for readers. The market turns a complex trade into a headline about another "Big Short."

The reported trade was more precise. According to Fortune, Scion Asset Management disclosed put options tied to Nvidia and Palantir in a quarterly filing. Put options can gain value when a stock falls. They can also be used as hedges, risk controls, or defined-risk views on valuation.

That means the filing does not prove Burry is making an all-in crash call. It also does not show the exact premium paid, strike prices, expiry dates, or whether the position still exists. A 13F filing is delayed by up to 45 days after quarter-end. It is a snapshot, not a live trading screen. In this case, the key disclosure came from Q3 2025 reporting, so the position could have changed before readers saw it.

Still, the disclosed targets are telling. Nvidia is the profit center of the AI infrastructure boom. Its chips sit at the center of data-center spending by Microsoft, Amazon, Alphabet, Meta, Oracle, and other cloud buyers. Palantir is a high-multiple software proxy for the belief that AI will lift enterprise productivity and create durable revenue.

Together, they represent two different layers of the AI trade. Nvidia reflects the hardware scarcity story. Palantir reflects the software monetization story. A bearish view on both says the market may be too confident about the entire AI value chain.

The Michael Burry AI Bubble Thesis: Useful Technology, Weak Economics

The Michael Burry AI bubble thesis is not that AI has no value. The sharper claim is that the economics may be weaker than the market assumes. A technology can be revolutionary and still produce poor returns for investors who pay too much.

There are four moving parts.

First, AI demand must become large enough to justify the infrastructure buildout. Companies are spending huge sums on data centers, networking, power, and GPUs. That spending assumes future users will pay enough for AI services to cover the cost of capital.

Second, AI products must generate attractive margins. Many consumer and enterprise AI tools are expensive to run. Inference costs, model training, cloud capacity, and customer support can eat into revenue. If users love the product but resist high prices, margins may stay thin.

Third, chips and data-center assets must keep their value for long enough. If GPU upgrade cycles are short, depreciation becomes a bigger drag. A company may report strong demand today, then face lower resale values and faster replacement needs tomorrow.

Fourth, equity valuations must leave room for error. When stocks price in near-perfect growth, small disappointments can matter. A modest cut to capex plans or margin assumptions can produce a large change in valuation.

This is why the thesis is about mispriced certainty. The market may be assuming that AI demand, AI margins, GPU useful life, and cloud returns are all already settled. They are not.

Why Nvidia Is Central to the Debate

Nvidia sits at the center because its revenue converts AI enthusiasm into reported numbers. When hyperscalers build data centers, Nvidia often sees the order first. That makes the company look like the cleanest winner in the AI cycle.

The bull case is strong. Nvidia has dominant chips, a deep software ecosystem, and a lead in high-performance computing. Demand has been supported by real customers with real budgets. Its platform also benefits from switching costs, developer tools, and supply constraints.

However, the bear case does not need Nvidia to be a weak company. It only needs the stock to price in more durable growth than customers can support. That is a much lower bar.

The most important issue is the spending cycle. Hyperscalers can overbuild even when demand is real. Telecom companies did something similar during the internet boom. Fiber was useful, but too much capital chased the same demand curve. Capacity eventually became cheap. Equity holders paid the price.

AI infrastructure could follow a different path. Utilization may stay high. Enterprise demand may broaden. Software may make each chip more productive over time. Nvidia also argues that useful life can be longer than skeptics assume, because systems can serve different workloads as the market matures.

The dispute is therefore measurable. Watch capex guidance from major cloud companies. Watch whether AI revenue becomes visible and high margin. Watch depreciation schedules, leasing terms, and resale values for older GPUs. These data points matter more than speeches about the future.

Palantir Shows the Software Side of the Bet

Palantir adds a different angle. It is not the main supplier of GPUs. It is a software company whose valuation depends on AI adoption becoming a durable enterprise budget line.

That makes it a useful test case. If AI tools create clear productivity gains, companies should keep paying for them. Contract sizes should rise. Gross margins should remain healthy. Revenue growth should become less dependent on hype and more tied to repeatable operating value.

The risk is that AI software becomes crowded. Many firms are selling workflow automation, analytics, coding support, and agent tools. If buyers see these tools as useful but interchangeable, pricing power may fade. High valuation multiples become harder to defend.

Palantir bulls argue that the company is not a generic AI wrapper. They point to government contracts, operating discipline, enterprise deployment experience, and a product stack built around complex workflows. Those are serious points. They also explain why a simple "AI is a bubble" label is too blunt.

Still, the market question remains. How much future success is already priced in? A great company can still be a risky stock when expectations are high. That is the core tension behind Burry's software-side bet.

The Dot-Com Parallel: What Fits and What Does Not

Investopedia reported that Burry compared today's market feeling to the last months of the 1999-2000 bubble. That comparison is powerful because it is familiar. It is also easy to misuse.

The useful parallel is capital intensity. In the late 1990s, investors funded networks, portals, hardware, and telecom capacity around a real technological shift. The internet was not fake. The problem was that many businesses could not earn returns that matched the capital deployed.

AI has a similar pressure point. Training frontier models and serving users at scale require expensive infrastructure. The larger the models become, the more capital the ecosystem needs. If revenue arrives later than expected, or margins prove weaker, valuations can reset fast.

Another parallel is narrative compression. In a bubble, many different companies get grouped under one story. The internet story lifted weak portals beside strong platforms. The AI story can lift chipmakers, software vendors, power providers, cloud platforms, and data-center owners at the same time.

However, there are clear differences. Today's leading AI companies are not all pre-revenue startups. Many are profitable giants with strong balance sheets. Cloud platforms already serve paying customers. Nvidia has reported major revenue, not just promised it.

That is why the right question is not "Is this exactly 2000?" It is "Which parts of 2000 are repeating?" The answer may be overinvestment, crowded funding, and valuation risk, rather than a collapse in the usefulness of the technology.

The Bull Case Burry Has to Beat

Any serious analysis must give the other side its best version. AI bulls do not need perfection. They need enough real demand to justify a large share of the buildout.

Their first argument is adoption. AI tools are spreading across coding, customer service, search, design, legal work, finance, security, and data analysis. Even partial productivity gains can support large markets if the tools become normal business infrastructure.

Their second argument is platform control. The largest cloud companies can absorb early capex because they own distribution. They can bundle AI into existing software, cloud contracts, and enterprise relationships. That gives them more pricing options than standalone startups.

Their third argument is that infrastructure creates option value. Building capacity before demand is fully visible can look wasteful in the short run. It can also secure strategic control if demand later accelerates.

Their fourth argument is that accounting can understate long-term usefulness. A GPU does not become useless just because a newer model appears. Older chips can shift to smaller models, inference, research, or lower-cost workloads.

These arguments are not weak. They explain why shorting a powerful theme can be painful even when valuation risk is real. Timing is part of the trade. A bubble can keep expanding while skeptics are early.

That is the hardest problem for any Burry-style thesis. Being directionally right is not enough. The catalyst must arrive before the position becomes too costly to hold.

The Evidence Table

The cleanest way to track this debate is to separate claims from proof. That makes the thesis testable instead of personality-driven.

Use five checks.

  • AI capex: compare data-center spending with disclosed AI revenue.
  • Nvidia demand: track backlog, gross margins, customer concentration, and GPU resale values.
  • AI software margins: watch net retention, gross margins, contract growth, and churn.
  • Dot-com risk: compare earnings quality, free cash flow, debt issuance, and capex discipline.
  • 13F conviction: treat filings as delayed clues, then wait for future filings and realized catalysts.

This table also shows why the debate is not binary. Some AI stocks may be overvalued while others remain strong. Some infrastructure may earn high returns while other capacity becomes excess. The market will not settle this with one headline.

Michael Burry AI bubble Signals to Watch Next

Prediction-market thinking is useful here because it forces clear conditions. Instead of asking whether AI is a bubble, ask what events would change the odds.

Start with hyperscaler capex. If Microsoft, Amazon, Alphabet, Meta, or Oracle slow data-center spending, the bearish case gains weight. If they keep spending while showing clear AI revenue, the bull case strengthens.

Next, watch AI revenue quality. Revenue tied to durable contracts matters more than demos, pilots, or bundled features. Gross margin matters too. High growth with weak margins is a different story from high growth with pricing power.

Then watch depreciation. Shorter useful lives make AI infrastructure more expensive than headline revenue suggests. Longer useful lives support the case that current spending can pay off across several product cycles.

Also watch customer concentration. If a small group of buyers drives most demand, a pause by one major customer can ripple through suppliers. Broader demand lowers that risk.

Finally, watch credit markets and power constraints. Data centers need financing, grid access, and energy supply. If funding costs rise or power bottlenecks delay projects, revenue assumptions may slip.

For prediction markets, the best contracts would not ask vague questions like "Is AI a bubble?" Better markets would focus on observable milestones:

  • Will a major hyperscaler cut AI capex guidance before year-end?
  • Will Nvidia gross margin fall below a defined threshold in a specific quarter?
  • Will a major AI lab announce a material price increase for enterprise users?
  • Will a large cloud provider disclose AI revenue above a stated level?
  • Will GPU resale prices fall by a defined percentage over six months?

Those markets would turn a narrative fight into measurable probabilities. That is the right way to read Burry's warning.

What Investors Should Take From the Warning

The Michael Burry AI bubble thesis is best understood as a risk framework. It is not a clean forecast that AI will fail. It is a warning that the market may be treating several uncertain assumptions as facts.

That makes the thesis useful even for readers who disagree with it. Bulls should be able to explain why AI revenue will cover capex, why margins will hold, why GPU lives are long enough, and why valuations leave room for mistakes. Bears should be able to explain what catalyst forces repricing and when it might arrive.

The strongest conclusion is measured. AI can transform business and still disappoint equity investors at the wrong price. Nvidia can remain a great company and still face valuation risk. Palantir can keep growing and still trade with little margin for error. The dot-com lesson is not that technology hype is always false. It is that cash flows decide who survives the hype.

That is the inside value of Burry's next Big Short. It pushes the Michael Burry AI bubble debate away from "AI good" versus "AI fake." The real debate is about demand, margins, asset lives, capex discipline, and timing.

For Macro Markets readers, that is also the natural prediction-market frame. Do not anchor on a famous investor's reputation alone. Track the events that would confirm or falsify the Michael Burry AI bubble thesis, then update probabilities as the data arrives.

Author
Dale Hanson
Dale HansonCommunity Manager. X @InHansonWeTrust
Categories