# Anthropic nears $7B Decart deal after founders favor it over Nvidia

> Source: <https://runtimewire.com/article/anthropic-nears-7b-decart-acquisition-nvidia>
> Published: 2026-08-16 08:01:48+00:00

Anthropic is nearing an agreement to acquire [Decart](/models/fal/decart-lucy-5b-image-to-video) at an approximately $7 billion valuation, with the Israeli AI developer's founders and largest investor, Sequoia Capital, favoring Anthropic over a reportedly higher offer from Nvidia, [Ctech reported Sunday](https://www.calcalistech.com/ctechnews/article/b1evv3aufg).

The parties are exchanging advanced agreement drafts, according to Ctech, which said most of the consideration would be paid in Anthropic shares. No agreement has been signed. Decart has hired investment bank Catalyst to advise on the process, and Ctech reported that a deal could be signed by early September.

The choice puts [Dean Leitersdorf and Moshe Shalev](https://www.calcalistech.com/ctechnews/article/6rn7vscw2), the founders named in Ctech's account, at the center of a bidding contest between the company that designs the dominant AI accelerators and one of the largest buyers of AI computing capacity. Decart's founding group also includes CTO Orian Leitersdorf, according to [Radical Ventures](https://radical.vc/portfolio/decart/), which led Decart's latest funding round.

Dean Leitersdorf completed a computer science doctorate at the Technion at 23 and later conducted postdoctoral research at the National University of Singapore. Shalev entered technology through Israel's Unit 8200 after growing up in a Haredi household, working in a butcher shop and studying accounting at night. He spent about 13 years in the intelligence unit and helped run its AI operations. Orian Leitersdorf completed his Technion doctorate at 21, according to Radical Ventures.

Decart was registered on September 7th, 2023, weeks before both Dean Leitersdorf and Shalev were called into military reserve service. The founders built Decart around a computing problem that has become increasingly expensive for frontier model developers: getting more useful work from scarce AI chips.

### A fast markup on a young company

The reported price has moved sharply during the negotiations. [Axios reported on August 13th](https://www.axios.com/2026/08/13/anthropic-decart-nvidia-ipo) that Anthropic was discussing a transaction worth about $6 billion. Ctech's new account puts the expected valuation at approximately $7 billion.

That would be $3 billion above the nearly $4 billion valuation attached to Decart's $300 million financing in May, a 75% increase in roughly three months. [Radical Ventures led that round](https://radical.vc/the-world-model-infrastructure-layer/), joined by Nvidia, Sequoia, Benchmark and Atreides, among other investors. Decart says it has raised more than $450 million in total.

Nvidia's dual role as investor and competing bidder complicates the deal. Ctech reported that Decart initially advanced talks with Nvidia, then stopped those negotiations after Anthropic entered the process. Nvidia is believed to have offered a higher valuation.

The structure makes the headline prices difficult to compare. A transaction paid primarily in Anthropic stock would leave Decart's shareholders exposed to Anthropic's value after a potential public offering. Ctech reported that Anthropic has submitted a confidential prospectus and is preparing for a possible September or October listing. Axios also tied the acquisition talks to an anticipated autumn IPO.

Google and SpaceX remain possible alternatives, according to Ctech, although the publication reported that Decart's founders and Sequoia view Anthropic as the preferred owner.

### Anthropic is buying compute economics

[Decart](https://decart.ai/) develops software and models for low-latency AI workloads. Its Decart Optimization Stack, or DOS, spans hardware-aware model design, compilers, kernel tooling and inference optimization. Decart says the system works across Nvidia GPUs, Google's TPUs and Amazon's Trainium chips.

That hardware flexibility is central to Anthropic's interest. Every improvement in inference efficiency lowers the computing required to serve Claude responses, coding tasks and agent workloads. It can also give Anthropic more freedom to shift workloads among chip suppliers and cloud partners instead of optimizing around one hardware architecture.

Radical Ventures said in May that Decart's stack delivered more than 1,600 tokens per second for agentic inference, compared with what the investor described as an industry average of about 200. Those are investor-supplied performance figures, and the results will depend on the model, hardware and workload.

Decart has used the same optimization work to build real-time video and world models. Its Lucy product transforms video in response to user input, while Oasis generates interactive environments aimed at gaming, robotics and autonomous-system simulations. Decart's [product history](https://decart.ai/about) shows a rapid expansion from GPU optimization into video generation, physical AI and commercial APIs.

Anthropic would also acquire an engineering base in Israel. Ctech reported that Decart employs about 100 people and that an acquisition would establish Anthropic's first development center in the country.

The proposed sale would end Decart's three-year run as an independent company shortly after its largest financing. Dean Leitersdorf previously told Ctech that he wanted Decart to become "the Google or Apple of AI." Anthropic offers the founders immediate access to model demand and computing capacity at a scale Decart would have needed years and substantially more capital to build alone.

That explains why Decart's shareholders may accept a lower nominal offer. Anthropic provides a direct use case for Decart's inference software, a path to deploying it across frontier models and stock tied to a planned public listing. Nvidia offered a chipmaker's distribution. Anthropic is offering Decart's technology a place inside the workload it was built to accelerate.
