Amazon is looking to acquire the AI developer Decart for $7 billion, according to a report by the Wall Street Journal.
Citing people familiar with the deal, the publication said that a deal could close soon depending on talks continuing and Amazon remaining the sole bidder.
Decart is a San Francisco-based, Israeli AI firm focused on producing real-time AI models. Backed by $400 million in funding from firms such as Nvidia, Decart was valued at $4 billion in May.
Its flagship products include the real-time video generation and editing model Lucy 2.5 and the world model Oasis 3.0, which generates controllable simulation environments for training robots and autonomous vehicles.
Along with its models, Decart also runs the Decart Optimisation Stack (DOS), a hardware-agnostic infrastructure and software layer that optimises AI model performance on AMD and Nvidia GPUs as well as Amazon’s custom chips and Google’s tensor processing units.
Amazon’s in-house AI chips, Trainium and Inferentia, have been used by firms such as Anthropic to train its Claude AI models as well as Databricks and Uber for machine learning workloads.
But most AI models are still developed to run on Nvidia’s CUDA computing platform due to the chip giant’s dominance of the chip market and decades’ worth of complementary AI software libraries associated with the Nvidia chip ecosystem.
If it integrated DOS within its offerings, Amazon could make Trainium and Inferentia more accessible to AI developers and allow them to use its chips without having to migrate away from Nvidia chips.
The Chinese chip giant Huawei and AI firm DeepSeek have recently partnered on a similar initiative, open-sourcing a toolkit that allows AI developers to produce models that run directly on Huawei chips without any dependency on Western software.
With Amazon pushing into warehouse robotics and autonomous drone deliveries, it could look to world models such as Decart’s Oasis for next-generation robotics training.
This is an area of increasing interest for tech giants. In September, AMD announced a deal to acquire the spatial computing startup World Labs for $8.2 billion. Founded by the deep learning pioneer Fei-Fei Li, World Labs produces world models that seek to accurately simulate real-world environments.
