What brings Anthropic to acquire the Israeli company Decart
There is quite a bit of synergy between the two companies that makes the potential acquisition deal a logical step: while Anthropic holds the largest market share in the field of code generators, Decart's models, which are capable of making decisions, planning tasks, analyzing the environment, and understanding language, could help Anthropic enter worlds where it is less dominant. And this is not the only reason why the deal will benefit Anthropic.
The Israeli startup Decart is on its way to signing an acquisition deal by the "Claude" manufacturer Anthropic, in a historic deal that may bring the AI giant to Israel for the first time, as reported this morning (Thursday) by the Bloomberg website. Until now, Anthropic has operated in the local market mainly through Israeli sales representatives stationed in Ireland - and now it seems that the company founded by Dean Leitersdorf and Moshe Shalev in Tel Aviv, which had already conducted similar negotiations with Nvidia - is on its way to a huge exit.
If the deal materializes, it will make the Israeli entrepreneurs millionaires and generate a handsome return for some of the company's early investors: Zeev Ventures of serial investor Oren Zeev, private Israeli-American investor Tilly Kalisky, and the funds Sequoia and Benchmark. Some prominent figures such as Anthropic scientist Andrej Karpathy, Prime Minister's advisor Michael Eisenberg (via Aleph fund), and the commander of Unit 8200 during the October 7 massacre Yossi Sariel also hold shares in the company.
The founders realized they needed to sell
What brings Anthropic to pay five or six billion dollars to a company founded only three years ago? The most important component is perhaps the realization that permeated among the founders and investors that there is a need to sell the company: at Decart, they did indeed develop graphical language models for the worlds of gaming and 3D, but these struggled to generate stable revenues for the company and led the company to make frequent product changes. Most of its revenues came through another, more basic product that deals with improving the processing efficiency of language models on graphical processors, an important feature for the needs of the inference stage, meaning running language models after they have already been trained - a product it sold to AI labs. All the while, Decart was forced to spend tens of millions of dollars on renting computing resources, graphical processors, and server farm space.
And yet, between the two there is quite a bit of synergy that makes such a deal a logical step: Anthropic brings to the table not only language and inference models - mainly in the fields of code writing and running AI agents that perform tasks in organizations - it also has a huge market that reaches every self-respecting organization and it holds the largest market share in the field of code generators.
Decart's language models are aimed at the visual world - in what it calls "world models", the company knows how to simulate 3D objects, environments, and backgrounds and calculate the relationships between them in various situations. The first use Decart made in the field is in assisting game development - but it left the market over time. However, its models were found to be particularly effective for the world of robotics - in what is referred to by Nvidia as "physical AI", a growing market that includes robots in a manufacturing plant, autonomous cars, autonomous weapons means such as vehicles, naval vessels or unmanned aerial vehicles, drones, and even humanoid robots that perform tasks such as manufacturing products in a factory, arranging boxes in a warehouse, or household cleaning work.
Decart's models, which are capable of making decisions, planning tasks, analyzing the environment, and understanding language, could help Anthropic enter also worlds where it is less dominant today and compete with Elon Musk's xAI which is integrated into the space and aviation worlds or Google's DeepMind.
Another reason why Anthropic is willing to go far and embrace the Israeli company is its optimization engine - the original technology developed by the company that allows it to improve the efficiency of training and running models not only on Nvidia chips but also on a series of competing chips such as Google's TPU processors and Trainium chips from Amazon - developed by the Israeli company Annapurna. In this way, Anthropic can appeal to organizations that operate on a variety of processor types and perform for them what they are looking for most of all: the ability to lower AI computing costs, the price per token, and the latency of complex processing like that of video, sound, and 3D objects.
