Traders sold Nvidia to buy CoreWeave on Huang's AGI declaration

Too big to grow is a claim about arithmetic, and the arithmetic is unkind: $96 billion of revenue in a quarter, up 106% from a year earlier, with a guide to $108 billion, means the current quarter has to add about $12 billion sequentially ^[9][10], which is roughly a quarter of the $46.6 billion the entire company sold in the year-ago quarter ^[11][12]. A model release, however good, does not visibly change that number inside the ninety days a trader cares about, and Gil Luria's description of Nvidia as a source of funds, the thing you sell in order to buy CoreWeave, is the cleanest available account of a 2% decline on a day when the names levered to OpenAI rose ^[3][8].

The rotation reading has belief in OpenAI's model accruing to whoever sells OpenAI capacity, which is why Luria points at SoftBank, Oracle and CoreWeave, and why he adds that Astra restores OpenAI to state of the art after about a year away ^[6][7]. Against that, CoreWeave has spent months under suspicion about its debt, and a 15% day in a heavily doubted name is what a squeeze looks like as much as what a re-rating looks like ^[3]. Basil Halperin's objection goes further, and it is the one worth sitting with: equities are a poor instrument for this question at all, since the two leading labs are private, competition between OpenAI and Anthropic could leave the industry earning Uber and Lyft money rather than Google and Apple money, and a genuinely dangerous system would send capital into Treasuries and gold rather than into the stocks ^[21].

Which is where the definitional slippage starts to matter more than the tape. Huang told Lex Fridman in March that AGI meant the ability to create a billion-dollar company, and made no mention of that benchmark on Sunday ^[14]; Luria's version of the same test, a billion-dollar business built by one person with a model, or an agent hired by an agent alone, remains a hypothetical ^[15]; Gary Marcus asked whether the man selling the chips had an incentive to call the race ^[16]; and the one checkable figure in the claim, the training fleet, moved from 300,000 Grace Blackwell chips to 100,000 when the first post was deleted, a factor of three ^[2][23].

So, a view, held with the confidence one session can carry, which is not much: Tuesday was a funding rotation inside a trade that is already owned, not the market putting a number on artificial general intelligence, because a 2% move in the largest name and a 0.3% move the last time Huang said this ^[4] are consistent with nearly any belief you care to hold, and Fortune's own two explanations, that markets disbelieve him or that AGI is already priced, cannot be told apart by a single day's prints ^[5]. The counter-thesis is Luria's, and it is respectable: the market did respond, just in the equities where an incremental dollar of OpenAI compute demand actually moves next year's earnings ^[6]. The measure that would settle it, on Halperin's logic, is real interest rates, since faster growth raises rates and higher rates shrink the present value of future profits, which is how a real AGI could arrive alongside a falling Nvidia ^[22]. Halperin, who by the well-educated-adult standard says we absolutely have not got there ^[19], spent three hours last weekend babysitting GPT-5.6 while it moved his iTunes playlists to Spotify ^[20]. That is not a rate move.