Michael Burry Says Nvidia's Big Depreciation Defense Doesn't Add Up Nvidia put out a slide last week meant to shut down its critics once and for all. Michael Burry was not shut down. The slide argued Nvidia's chips hold their value far longer than the standard five-year depreciation schedule suggests. Fans cheered it as proof the bears were wrong. Burry, writing on his Cassandra Unchained Substack, says the comparison is rigged from the start. Nvidia's chart measures retained value using an eight-year cash flow model. It then compares that against a five-year depreciation curve. Burry calls that an apples-to-oranges trick. Stretch the timeline and almost anything looks like it holds value better. His deeper point is about why older chips like the A100 and H100 are commanding strong rental prices right now. He says that is scarcity, not durability. Memory and power shortages are propping up demand for aging hardware. Once those shortages ease and Nvidia's next architecture, Vera Rubin, ramps up, Burry expects rental prices for the old chips to fall hard. To make his case, he reaches back to 1968. Third-party computer leasing firms were undercutting IBM with generous eight-year depreciation schedules of their own. Those firms were right that mainframes would keep running for years. They were wrong about how fast rental income would collapse once IBM launched a new generation of machines. By 1970, computer leasing stocks had fallen roughly 80%. Heavily leveraged funds in that trade were wiped out. Burry argues today's AI buildout carries the same risk, just dressed up in GPUs instead of mainframes. Billions in AI infrastructure are being financed through structured debt and private credit. If rental yields compress suddenly, that risk doesn't stay with the chipmakers. It lands on the lenders and insurers who financed the buildout in the first place. Burry isn't a neutral observer here. He's built his reputation and his newer Substack business around spotting exactly this kind of mismatch years before the market catches on. That track record is why his posts move markets, even the short ones that read more like a classroom lecture than a trade call. Nvidia, for its part, has leaned hard into pitching durability as a selling point. Executives have repeatedly told investors that older chips keep earning well past their official depreciation window. The company's own data centers still use chips from several generations back, which Nvidia cites as proof of staying power. Burry's counter is simple. Proof of use isn't proof of value. A chip can keep running and still be worth far less than its book price once a newer, faster generation shows up. Vera Rubin, Nvidia's next architecture, is expected to ship within the next year. History says that's exactly when older hardware gets repriced. What It Means for YouBurry isn't predicting Nvidia's collapse tomorrow. He's flagging a mismatch between a flashy slide and the economics underneath it. When a company has to reach for an eight-year model to defend a five-year asset, that's worth noticing. Scarcity-driven pricing can look like durability for a while. History suggests it rarely stays that way once supply catches up. If you own AI infrastructure names, depreciation assumptions are one of the quieter numbers worth checking in a 10-K. They can tell you more than the headline growth rate. |
No comments:
Post a Comment