Decart's founders traded a higher Nvidia bid for Anthropic paper that never printed

Take the numbers the reporting does supply and the price looks less like a multiple than an option on scarcity: several tens of millions of dollars of annual revenue at a three-year-old firm of roughly 100 people ^[4], bought at $6bn to $7bn ^[1], is somewhere between 86 and 233 times revenue depending on where "several tens" actually sits ^[19], and $60m to $70m per employee ^[20]. Neither figure tells you whether the technology works.

The currency does more work. Dean Leitersdorf and Moshe Shalev ^[3] were willing to take less than Nvidia was reportedly prepared to pay, with much of the consideration in Anthropic shares ^[11][10], and Anthropic is still private, has not published a prospectus, and expects to list within two months ^[13]. So the headline range was denominated in an instrument with no public mark, quoted against a rival bid from a company that is already an investor in Decart ^[10]. The discount reflected the founders' own view of Anthropic's IPO, expressed in equity they will now not receive.

The scalability reading needs care, though. Calcalist's Ctech raises the possibility that Decart's efficiency gains weakened at very large scale, and that the chip-agnostic promise (Nvidia GPUs plus silicon from Google and Amazon) did not hold up ^[8][7][6], but it raises them as questions, and reports no explanation from Anthropic of what it found ^[23]. The same piece offers two other candidates: that Anthropic, having seen the internals during diligence, concluded it could build the equivalent, and that Qatar's investment in Anthropic made buying an Israeli company awkward ^[9].

These three readings point at different assets. If the technology failed at scale, the finding generalises and any buyer reproduces it, which is the version that hurts Decart most. If it was fit, the same article notes that Anthropic sells models and services while Nvidia could fold chip-efficiency work into a performance offering and keep it away from competitors ^[16][17], and the asset is undamaged. If it was alignment, note that Leitersdorf is 27, finished a computer science doctorate at 24, and has said he wants to build something on the scale of Google or Meta ^[18], which is exactly why he would take paper over cash, and also exactly what a buyer might worry about when pricing retention.

By walking away, Anthropic avoids spending pre-IPO equity on a nine-figure-revenue-free acquisition weeks before a listing. Calcalist offers the contrast that Musk waited until SpaceX's IPO closed before buying Cursor for $60bn ^[14], a price between 8.6 and 10 times the Decart range ^[22]; and the one bidder said to be willing to pay more is currently digesting Hugging Face at $12.9bn ^[12], roughly 1.8 to 2.2 times what Decart was being valued at ^[21].

On the evidence available, what's established is the structure of the deal; why it died is not. The falsifier is cheap and public: if Decart signs with another buyer at or above $6bn, then the collapse was about currency and strategic fit and the raised-diligence-bar story loses its only support; if the next bid arrives at half, Calcalist's questions about the chips were the right ones, and every AI target selling efficiency claims should expect its benchmarks reproduced at scale before signing rather than after.