The $500 Strike That Wasn't: Michael Burry's AI Short and the Arithmetic of Being Early

On September 29, a filing surfaced that Michael Burry had extended a short position against the AI complex — put options on Micron, Nebius, SOXX, and Palantir. The headline circulated within hours. What did not circulate was the arithmetic. One of the strikes was listed at approximately $500. Micron has never traded at $500. It did not trade at $500 in 2000, it did not trade at $500 in 2018, and it is not trading at $500 now. Either the reporting is wrong, or the position is not what the headline says it is. The ledger remembers what the headline forgets. The context here is not complicated, though the market treats it as if it were. We are in the third year of an AI capital expenditure cycle that has restructured the semiconductor supply chain. Cloud providers have committed to multi-year buildouts. Memory manufacturers have shifted capacity toward HBM. Nvidia's data center revenue has become the single most-watched earnings line in technology. Into this, Burry — the man who read the subprime mortgage market before it read itself — has placed a series of put options with expiration dates in June of next year and September 2027. I have spent the last decade auditing code and reconstructing failure timelines. My work does not involve predicting stock prices. But it does involve reading structural vulnerabilities, and there is a structural argument buried in Burry's filing that deserves more attention than the headline received. The argument, as reported, comes from Ares Management: current AI investment depends on "unproven AI revenue" supporting continued capital expenditure increases. The phrase "unproven AI revenue" is doing a great deal of work. It is not a claim that AI is useless. It is a claim that the return on invested capital has not yet been demonstrated at the scale required to justify the spend. This is a testable proposition, and the mechanism for its failure is specific. According to the reporting, corporate boards have legal latitude to reallocate capital away from AI if a single quarter's revenue misses expectations. I want to be precise about what this means. It means the capital expenditure commitments that underpin the entire AI supply chain are not contractual obligations in the way a bond coupon is. They are discretionary budgets with flexibility clauses. If that is accurate, then the AI infrastructure buildout is not a fixed commitment — it is a rolling option that boards can decline to exercise. Pics are noise; the hash is the identity. The identity of this trade is a bet on timing, not direction. Burry's history is instructive here, and I mean that in the clinical sense. He was right about subprime — direction and timing both. He was early on Tesla. He was early on ARK. He was early on the 2023 market. The pattern is consistent: he identifies structural fragility correctly and then positions for its resolution before the resolution arrives. In options terms, this is expensive. Theta decay does not care about being right. It cares about being right now. Let me reconstruct the position as reported. Micron puts expiring June of next year, strike around $500 — this number is almost certainly wrong. Nebius puts, same expiration, strike in the double digits. SOXX puts expiring September 2027, strike slightly above $400. Palantir puts, same September 2027 expiration, strike slightly above $100. Four positions, two expiration windows, four different theses. Micron is a cyclical bet on memory pricing. Nebius is a solvency bet on a cash-burning infrastructure entrant. SOXX is a sector bet. Palantir is a valuation bet. These are not the same trade wearing different tickers. The Chinese memory supply situation is the piece that most analysts are missing. Acer's CEO has noted that Chinese storage chip capacity continues to increase, and that supply shortages cannot persist indefinitely. This is a supply-side argument that operates independently of AI demand. Even if AI capex holds steady, memory pricing could compress because capacity is expanding. That would give Burry a win on Micron for reasons that have nothing to do with the AI bubble thesis. It would be a cyclical trade dressed as a structural one. Here is where the bulls have a point that the bears consistently underweight. The major technology companies funding AI capex are not leveraged buyout vehicles. They generate enormous free cash flow. Microsoft, Google, Amazon, Meta — these are not companies that need to tap credit markets to fund a data center. They can self-finance. The 2000 comparison fails on this specific point: the companies that collapsed in 2000 were often pre-revenue, debt-funded shells. The companies building AI infrastructure today are profitable, cash-generative, and investment-grade. The structure is different. The structure matters. But structure cuts both ways. The same boards that can self-finance can also self-de-fund. If AI revenue does not materialize, the reallocation decision is easier for a cash-rich company than for a leveraged one, because there is no covenant restricting the pivot. The legal flexibility that permits aggressive investment also permits aggressive withdrawal. Silence in the code speaks louder than the pitch. The volatility context is the part of this trade that tells you the most about Burry's thinking. He reportedly described VIX as being at extremely low levels, making put options relatively cheap. This is volatility arbitrage logic: buy tail protection when the market is complacent. It is a legitimate institutional strategy. It is also a strategy that is wrong most of the time, because most of the time the tail event does not occur within the option's life. The cost of being early is the premium paid. The cost of being wrong is the same. What I find most interesting is the tax angle. Burry mentioned that the shift to options was partly motivated by reducing tax liability. This suggests a longer holding period, which contradicts the "accelerating the timeline" framing in the headline. The reconciliation is this: he is positioning for a long-term structural repricing, but using leverage to amplify the return if the repricing arrives sooner than the long term. This is internally coherent. It is also internally fragile, because leverage amplifies the cost of delay. Every bug is a footprint left in haste. The footprint here is the truncated reporting. We do not know the position size relative to Burry's total portfolio. We do not know whether he is short Nvidia — the most obvious AI bubble target, and one conspicuously absent from the reported positions. We do not have the full Ares Management research. We have a CNBC report that has passed through at least one aggregation layer, and the $500 Micron strike suggests something was lost in that passage. A forensic reconstruction that begins with a data error is already compromised. What is the actual revenue trajectory of AI applications? Is it slow relative to capex, or is it simply early? The difference between those two descriptions is the difference between a bubble and a buildout. In 1995, internet revenue was slow relative to telecom capex. The capex was not wrong. It was early. The companies that went bankrupt were the ones that could not fund the gap between early and eventual. The ones that could — Amazon, for instance — survived to build the infrastructure that made the revenue eventual. The AI industry today is not homogeneous. There are Amazon-equivalents with cash reserves and diversified revenue. There are also Nebius-equivalents — infrastructure entrants dependent on external financing, burning cash, needing the capital markets to remain open. Burry is shorting both. One of those shorts is a bet on the AI cycle. The other is a bet on the credit cycle. If they diverge — if AI demand holds but financing conditions tighten — the Nebius trade wins and the Palantir trade may not. If they converge downward, everything works. If they converge upward, the theta decay is total. History is not written; it is indexed. The relevant index here is not the price of Micron or Palantir in June. It is the ratio of AI revenue to AI capital expenditure, reported quarterly, auditable. That ratio has been improving. It has not improved enough to satisfy the bears. The distance between those two statements is the entire debate. The bears are not wrong that the ratio is unproven at scale. The bulls are not wrong that it is improving. Both can be true. Only one will be profitable. My own experience with failure analysis suggests that the most reliable predictor of a system's collapse is not the severity of its flaws but the availability of exit routes. In 2022, when UST de-pegged, the failure was total because the mechanism had no circuit breaker. When Tezos had its consensus edge case, the network survived because the flaw required conditions that rarely materialized. The AI capex cycle has an exit route — board reallocation — and that is both its protection and its vulnerability. It can be scaled back gracefully. It can also be scaled back suddenly. The legal flexibility that allows graceful scaling also allows sudden scaling. The protection against collapse is the same mechanism that enables it. What the market appears to be pricing is not a graceful scale-back. It is continued acceleration. The put options Burry has purchased are, in effect, a bet against that assumption. The question is not whether AI revenue will eventually justify the infrastructure. The question is whether it will do so within the window before June of next year. If not — if the window closes with the revenue still early and the capex still committed — the options expire worthless regardless of whether the long-term thesis was correct. Precision is the only apology the chain accepts, but the chain does not trade options. The clock does. The map is not the territory; the chain is both. Burry's map of AI vulnerability may be accurate. His position on the chain — leveraged, time-limited, correctly diagnosed but possibly mistimed — is a different matter. The structural argument is sound. The structural argument was sound in 2023 as well. The market has spent two years not resolving it. The resolution may come. The options may still expire first. That is not a contradiction. It is the trade.