The Amazon Decart deal could give AWS new software for switching between AI chips while adding real-time video and world-model technology to Amazon’s AI portfolio.
The Amazon Decart deal could become one of Amazon’s largest artificial intelligence acquisitions as the company looks for more control over the software and infrastructure behind advanced AI.
Amazon is in advanced discussions to acquire Decart for around $7 billion, according to people familiar with the talks.
The Wall Street Journal reported that a deal could come together soon, although negotiations are not final and another bidder could still emerge.
Decart was founded in 2023 and has built both AI infrastructure software and models designed to generate or edit visual worlds in real time.
Amazon Decart deal would buy more than an AI model
The most important part of the Amazon Decart deal may be Decart’s infrastructure software.
The startup has developed the Decart Optimization Stack, or DOS, which helps AI developers run workloads across different types of chips.
According to Decart’s official product information, DOS works across Nvidia hardware, AWS Trainium chips and Google’s Tensor Processing Units.
That flexibility matters because AI developers increasingly want to avoid being locked into one chip supplier.
Nvidia remains the dominant provider of high-end AI accelerators, but Amazon, Google and other cloud companies are spending billions to develop their own alternatives.
Amazon’s Trainium chips are central to that strategy.
If the Amazon Decart deal closes, Amazon would gain software designed to make AI workloads run efficiently across those competing architectures.
That could help AWS make Trainium easier for more developers to use.
Decart also builds world models
Decart is also working on a more experimental category of AI.
Its Oasis models generate interactive environments that can respond to what a user or autonomous system does.
These are often called world models.
Instead of producing a static image or short video, a world model attempts to simulate an environment that continues changing as actions take place inside it.
Decart says Oasis 3 can create real-time environments for autonomous vehicles, robots and other physical AI systems.
The Amazon Decart deal would therefore bring Amazon technology that sits somewhere between generative video, simulation and robotics.
The startup also develops Lucy, a model that edits live video in real time.
Lucy 2.5 can transform video while it is being streamed, rather than requiring users to wait for a finished render.
Those products could fit several parts of Amazon’s business, including AWS, advertising, gaming and robotics.
Amazon is building alternatives to Nvidia
The Amazon Decart deal also fits Amazon’s wider effort to reduce dependence on Nvidia.
Amazon has developed Trainium for training and running AI models at scale.
Decart received early access to Amazon’s Trainium3 chips and has worked on optimising its models for the hardware.
That relationship already gave Amazon a close look at Decart’s technology before the acquisition talks.
AI infrastructure has become one of the most expensive parts of the current technology cycle.
The AI Decode recently examined that problem in its report on AI spending and the 9% GDP revenue test, where the scale of planned data-center investment raises questions about how much revenue AI providers eventually need to generate.
More efficient software can improve that calculation.
If models can run on a wider range of chips or use existing hardware more efficiently, cloud companies may be able to lower part of the cost of serving AI.
Decart’s valuation has risen quickly
The price under discussion also shows how quickly Decart’s value has increased.
The startup was valued at nearly $4 billion in May after raising $300 million.
Its investors include Sequoia Capital, Benchmark, Nvidia and Toyota Ventures.
A roughly $7 billion Amazon Decart deal would therefore represent a substantial premium to that recent private valuation.
Amazon would be paying for a company that combines infrastructure software with newer world-model technology.
Those are two areas investors currently value highly.
The AI Decode recently covered Gemini 4 Argon, which showed how major AI developers are increasingly competing on efficiency, long-running workloads and infrastructure cost alongside benchmark performance.
Decart attacks the same problem from a different direction by trying to make hardware itself easier to use efficiently.
The deal is still not guaranteed
The Amazon Decart deal remains a negotiation.
Amazon and Decart have not announced a final agreement, and the Journal reported that other potential buyers may still be interested.
A $7 billion purchase would also require Amazon to show that Decart can create value beyond its current startup valuation.
World models remain an early technology.
Training autonomous systems inside synthetic environments could reduce the cost and danger of real-world testing, but generated environments still have to be accurate enough to teach useful behaviour.
The same applies to real-time AI video.
Low latency is useful, but commercial adoption will depend on quality, cost and reliability.
For Amazon, the infrastructure side may offer the more immediate payoff.
The Amazon Decart deal could strengthen Trainium, give AWS new optimisation software and add a team already working across several competing chip platforms.
If the talks result in an acquisition, the next question will be whether Amazon keeps Decart operating across Nvidia, Google and AWS hardware or pulls the technology more tightly into its own cloud ecosystem.
