Chanos asks why Nvidia sells the GPUs if renting them is this good
Jensen Huang says Nvidia's compute is "fungible, durable and highly rentable." One of Wall Street's best-known short sellers heard that and asked the obvious question: then why are you selling it? Two days later the argument has become the cleanest framing yet for the AI capex question.
Jim Chanos, who called Enron, replied to Huang's X post with four words that cut straight to the accounting: "Then why not rent them out yourself? Or simply keep raising prices?" Huang had reshared an index showing three-year-old H100 rental rates up roughly 22% in a month to around $3.28 an hour — a number that inverts the normal assumption that aging silicon steadily loses economic value. Chanos later clarified that his target isn't Nvidia selling chips; it's the neoclouds and GPU landlords buying them to rent out. His point is structural: if the returns on rented compute were as attractive as Huang's framing implies, the seller with the best cost basis would want to own the rental business rather than hand it to CoreWeave and its peers.
The gap between the two positions is really a disagreement about useful life, and it carries real money. Huang has spent the year making the case in public — an August essay titled "AI Factory Compute Is Becoming an Investable Asset Class," then this week's post — that a GPU is a productive asset with years of earnings in it, not a depreciating box. That thesis is now load-bearing: Nvidia has lined up a $500 billion-plus AI-factory financing venture with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, and any instrument that prices long GPU lives needs long GPU lives to be true. Michael Burry's counter-estimate is that hyperscalers are understating depreciation by roughly $176 billion across 2026–28 by booking five-to-six-year lives on hardware with two-to-three-year economics. Nothing here is settled, and Huang is plainly an interested party — but Chanos's question is the one the financing structures have to answer.
China's Enflame Technology has finished pricing its STAR Market IPO, raising 6.119 billion yuan at 142.18 yuan a share, and expects to post its first consolidated profit between 2026 and 2027. Founded in 2018, the company has iterated four generations of cloud AI chip across five products and counts Tencent among its anchor customers; first-half 2026 revenue grew nearly 280% year on year, per the prospectus. The listing also produced a number that says more about sentiment than fundamentals: the online retail allotment rate came in at 0.02455315%, meaning Chinese retail investors oversubscribed the tranche by several thousand times. We covered the pricing end of this in Enflame raises $908 million, and retail got 0.025% of what it wanted.
An Anthropic pretraining researcher's resignation is still ricocheting, and the reaction inside the field has been more interesting than the post itself. Jacob Coxon quit on Monday saying both OpenAI and Anthropic are "racing straight to self-improving super-intelligence and gambling with our lives," and that the people building these systems believe they could kill us all by 2030; his thread has passed five million views. What has changed since our first write-up is the argument it triggered: Coxon points to the Hugging Face breach as a "warning shot," and his proposed remedy is a temporary ban on capability improvements — a remarkably concrete ask from someone who spent three years inside the pretraining orgs. We covered the resignation itself in OpenAI and Anthropic both got worse today — one shipped, one lost a researcher.
What to watch: whether Nvidia's next earnings call answers the rent-versus-sell question with numbers rather than adjectives.
If rented GPUs are such a good business, why is the chipmaker selling them? Tell us in the comments.
