The deal that did not happen
Anthropic has decided against acquiring Decart, the Israeli startup whose software makes the chips used to train and run AI models work more efficiently. Bloomberg reported that the two companies discussed a transaction worth roughly $6bn and that Anthropic performed due diligence on Decart before walking away. No agreement was signed. Representatives for both companies declined to comment.
Bloomberg said the companies could still pursue other ways of working together. That is the standard formulation after talks end, and it is worth reading as exactly that until something is announced.
What Anthropic was looking at
Decart was founded in late 2023 by Dean Leitersdorf and Moshe Shalev, who met during service in the Israeli military’s Unit 8200. It is headquartered in Tel Aviv with offices in New York and San Francisco, and The Times of Israel reported in August that it had 106 employees, 89 of them in Israel.
The company has two lines of work. One is a software platform that raises the utilisation of GPUs during training and inference, which is the part that would matter most to a lab spending heavily on compute. The other is a set of generative video models that transform live or recorded footage from a text prompt.
Decart had raised more than $450m before the talks, including a $300m round led by Radical Ventures in May 2026 that valued it at $4bn, according to the same report. The $6bn figure discussed with Anthropic would have been a premium on that valuation four months later.
Why a compute-efficiency company
The logic of the deal was never obscure. Anthropic has spent 2026 signing compute contracts at a scale that dwarfs the price of Decart. Aivio News reported last week that the company has tied up 14.8 gigawatts of capacity in eleven months. Against commitments of that size, technology that raises the throughput of hardware you are already paying for is bought once and saves continuously.
What is not known is why the talks ended. Bloomberg’s account says only that Anthropic performed due diligence and decided not to proceed, and neither company has said anything on the record. Any explanation beyond that would be invention.
The context Anthropic is in
Anthropic rarely buys companies. It has been directing its money at computing capacity instead, while press reporting has repeatedly placed it on a path towards a public listing. A $6bn acquisition would have been a conspicuous departure from that pattern, which may be part of why it did not happen — though that is a reading of the circumstances, not something either party has confirmed.
For Decart, the outcome leaves an independent company with a fresh valuation, a large recent round and, on the evidence of the last month, at least one frontier lab that considered it worth six billion dollars.
What to watch
Whether the two companies announce a commercial arrangement instead of an acquisition, and whether another buyer emerges. Decart’s efficiency work is relevant to every lab and cloud provider buying accelerators at scale, and a valuation established in a failed sale process tends not to stay private for long.
