Anthropic has canceled the planned acquisition of the Israeli AI startup Decart after the company negotiated for months and conducted a due diligence review. The deal would have valued Decart at $6 billion – one and a half times the amount from the last funding round in May. Both companies remain silent about the reasons.
Months of review end without agreement
Initial reports about the talks emerged on August 13, 2026, including in Fortune, already citing the target price of $6 billion. Bloomberg reported on the end of the talks. Such a deal would have been, to the best of current knowledge, Anthropic’s largest acquisition. After weeks of review, however, the company decided against pursuing the acquisition further – a final agreement was never reached. It remains unclear whether the price or findings from the due diligence were decisive. Both sides declined to comment to Bloomberg. It is also uncertain whether Anthropic and Decart will instead agree to a loose commercial cooperation, as a person familiar with the matter suggested to Bloomberg. Before the Anthropic talks, Nvidia is said to have already made a higher purchase offer for Decart. The founders Dean Leitersdorf and Moshe Shalev, along with the other shareholders, nevertheless opted for Anthropic’s lower stock offer – in the hope of benefiting from a future Anthropic IPO.
Investors now seek a new buyer
Nvidia itself, with around $300 million, is among Decart’s investors, alongside Amazon, Sequoia, Benchmark, Radical Ventures and individual investors like OpenAI co-founder Andrej Karpathy. According to Decart, the company has raised more than $450 million in total. In the funding round in May, the company was valued at around $4 billion – significantly below the now-reported purchase amount. Investors and founders are reportedly now looking for other buyers for the company. Leitersdorf and Shalev, both veterans of the Israeli military unit 8200, founded Decart in September 2023. The company develops an optimization layer that aims to use chips more efficiently during the training and operation of AI models, as well as so-called world models that learn the behavior of physical objects from text and video material. The in-house model Lucy generates high-resolution videos from this in real time, for example for virtual fitting in online retail or during livestreams. The chip efficiency is likely to have been particularly interesting for Anthropic: the company is simultaneously buying computing power on a large scale for its language model Claude. For this, it recently built its own compute team with the recruitment of Monzo founder Tom Blomfield.
A purchase would have given Anthropic direct access to technology that could lower its own training and operating costs – instead, it remains open for now whether a looser cooperation follows. It will be crucial whether Decart finds a new buyer before the race for cheaper AI computing power closes the window of opportunity.
