Seven billion dollars is a lot to spend on a company most people have never heard of. But consumer recognition may be the least interesting thing about Decart.

Amazon is in talks to acquire the AI startup for around $7 billion, according to The Wall Street Journal. The report describes negotiations, rather than a completed acquisition.

The potential attraction becomes clearer when you look beneath Decart’s video demonstrations. Alongside its visual models, the company builds software designed to extract more performance from the processors running them. For a cloud business, that could mean serving more customers with the computing capacity it already owns.


Amazon Already Knows What It Is Buying Into

Decart’s relationship with Amazon predates the acquisition report.

In its May funding announcement, Decart identified Amazon as a strategic customer and described a commercial agreement and joint effort to bring its technology to AWS customers. Its Lucy 2 model was already running on Trainium3, Amazon’s custom AI chip.

That matters because an acquisition would build on a working technical relationship. Amazon has had an opportunity to see how Decart’s software performs inside its own infrastructure.

Nafea Bshara, a vice president at Amazon’s Annapurna Labs, said in the announcement that Lucy 2 exceeded 80% Model FLOPS Utilization—a measure of how effectively a model uses the chip’s theoretical computing capacity.

As he put it, “that’s real money saved and real time back.” The result describes a particular deployment, rather than a guaranteed improvement for every AI application.

The Software Behind the Spectacle

Decart calls its underlying technology the Decart Optimization Stack, or DOS. It offers tools to make AI training and operation faster and more efficient across different processors, including Nvidia GPUs, Google TPUs and Amazon Trainium.

In plain English, buying powerful chips is only part of the job. The software must keep them productively occupied.

Consider a live AI service responding to thousands of people at once. Every unnecessary delay affects the experience. Every wasted unit of computing capacity affects the cost. Improving the software can change both without requiring the operator to replace all its equipment.

That is the commercial logic behind optimization: make each expensive machine do more useful work.

It connects directly to the question in Blockster’s coverage of AI’s chip-financing boom: how does enormous spending on infrastructure become a business that pays for itself? Financing secures the hardware. Efficient operation helps determine what that hardware can earn.

For Amazon, owning a company that improves this process could offer benefits across many customers. That is a plausible strategic rationale, rather than a motive Amazon has confirmed.

The Part Shoppers and Creators Could Actually See

Decart also makes technology with a more immediate visual appeal.

Its Lucy 2.5 model, introduced in July, can transform streaming video at 1080p resolution and 30 frames per second, according to the company. Users can change clothing, replace characters, remove objects or alter a scene through prompts and reference images while the video continues.


The difference is timing. A conventional editing workflow asks you to make a change and wait for the result. Lucy is designed to keep responding as the action unfolds.

For a retailer, that suggests possibilities such as virtual try-ons. For an advertiser, it suggests changing elements inside a video. For creators and game developers, it opens another way to build experiences that respond to an audience.

Decart’s separate Oasis product focuses on interactive simulated environments for applications including robotics and autonomous vehicles. The company therefore combines infrastructure software with models intended to demonstrate what that infrastructure makes possible.

Those capabilities help explain the potential fit. They do not establish a roadmap for Amazon products.

AI Acquisitions Can Fill Very Different Gaps

The comparison with other AI deals is useful because the companies being purchased do different jobs.

When OpenAI confirmed that the io team had merged with it in July 2025, the incoming group brought hardware engineering, product development and manufacturing expertise. Jony Ive and LoveFrom remained independent while taking on design and creative responsibilities.

A potential Decart purchase would point toward another need: making AI economical to operate, alongside developing new visual experiences.

Decart had already attracted substantial backing. It announced a $300 million funding round in May, led by Radical Ventures, with Nvidia among the new investors.

Nvidia’s involvement also shows how intertwined these businesses are. Amazon develops its own AI chips while continuing to expand its Nvidia infrastructure. In August, AWS and Nvidia announced plans for two million additional GPUs across AWS infrastructure in 2027–2028.

Software that works across multiple chip families fits that mixed environment.

What Would Make the Deal Worth It?

The eventual test would be measurable: lower costs for useful AI work, reliable performance under heavy demand and products customers keep paying to use.

An acquisition would also raise questions about Decart’s future availability across competing hardware platforms. Its ability to work with different processors is part of its appeal. How Amazon might preserve or change that approach remains unknown.

For now, the existing partnership offers the clearest explanation for the interest. Amazon already supplies the computing infrastructure. Decart promises to help that infrastructure work harder—and turn the resulting performance into experiences people can use.

Reporting by Lidia Yadlos

AI