Anthropic AI
Decart's technology could be particularly useful as Anthropic faces growing demand for Claude and the high computing costs associated with developing and operating advanced AI systems. Unsplash

Anthropic is in talks to acquire Nvidia-backed artificial intelligence startup Decart AI in a deal that could value the company at about $6 billion, according to a new report.

The discussions are ongoing and may not result in a transaction, Reuters detailed, confirming an earlier report by Bloomberg. The deal would be Anthropic's largest acquisition to date and would bring Decart's AI infrastructure and world-model technology into one of the industry's biggest developers of frontier models.

Decart's technology could be particularly useful as Anthropic faces growing demand for Claude and the high computing costs associated with developing and operating advanced AI systems. The startup builds software aimed at making AI training and inference more efficient across different types of chips, potentially allowing companies to get more performance from expensive computing infrastructure.

Decart's Decart Optimization Stack, or DOS, is designed to optimize AI workloads across processors made by Nvidia, Amazon and Google. The Wall Street Journal reported in June that the technology makes it easier for AI developers to move workloads between different chips, reducing the technical complexity and costs involved in switching hardware.

That capability has attracted an unusual group of customers and investors. Nvidia participated in Decart's latest funding round even though the company's software can make it easier to run AI workloads on competing processors, while Amazon has become both a customer and a provider of chips used by the startup, the Journal reported.

Decart raised $300 million in May in a funding round led by Radical Ventures, bringing its total funding to more than $450 million. The company said the investment would support three main product lines: DOS, its infrastructure platform; Lucy, a world model focused on immersive experiences; and Oasis, which is designed for physical AI applications.

The funding valued Decart at nearly $4 billion, according to the Wall Street Journal, meaning a $6 billion acquisition would represent a substantial premium just months after its latest financing.

Nvidia joined the round alongside investors including Atreides Management, Valor Equity Partners, Sequoia Capital, Adobe Ventures, Benchmark and Zeev Ventures. Decart was founded in 2023 by Dean Leitersdorf, Orian Leitersdorf and Moshe Shalev.

Beyond infrastructure, the company has been developing so-called world models that generate and manipulate interactive representations of physical environments. Its Lucy system can transform live video in real time, allowing users to change people, objects, backgrounds and visual effects while footage is being captured or streamed.

Decart says Lucy operates using its optimized inference and training infrastructure and is being developed for applications including e-commerce, advertising, live streaming, social platforms and gaming. Its Oasis model is aimed at physical AI, including simulations that could eventually be used in areas such as robotics.

If an acquisition is completed, Decart employees are expected to join Anthropic's inference and performance organization, Reuters reported. That would put the startup's infrastructure expertise directly inside the part of Anthropic responsible for efficiently running its AI models.

Anthropic's expansion is unfolding as competition among leading AI companies increasingly centers on the cost of running models as well as their capabilities. Companies are spending heavily on Nvidia chips, data centers and power while trying to make inference, the computing process used when customers interact with an AI model, faster and less expensive.

That pressure is growing as cheaper open-weight AI models gain ground. Reuters reported that U.S. companies are facing increased competition from lower-cost models developed by Chinese companies including Moonshot AI and Z.ai, prompting American developers to put greater emphasis on efficiency and more accessible models.