Amazon is in talks to acquire Decart in a transaction that could be worth around $7 billion, the Wall Street Journal reported on October 11, citing people familiar with the matter. This is a reported negotiation, not a signed agreement or completed acquisition. The price could change, and talks do not guarantee a deal.
The interest in the company extends beyond the videos it generates. Decart also develops software that helps developers switch between different AI chips, the newspaper said. That connection between a model and the infrastructure running it helps explain the story: making generated video interactive requires more than a convincing picture. It also requires quick responses and sustainable operating costs.
An existing relationship with Amazon
San Francisco-based Decart, founded by Israelis, already has a relationship with Amazon. In its official December 2025 announcement about Trainium3, Amazon identified the company as a user of its technology. It said Decart achieved 4 times faster inference for real-time video generation at half the cost of GPUs. These are supplier-reported results for a particular workload, not a promise for every customer or model.
Inference is the stage at which a trained model produces a new output. A service generating a single image can sometimes tolerate a wait. When someone changes a live scene or moves a character through an AI-generated world, the delay itself changes the experience. Software and hardware optimization therefore matter alongside output quality. Success in one benchmark does not establish how a whole service performs under load.
The existing commercial relationship provides context for the reported talks, but does not establish Amazon's objectives in a possible acquisition. The Trainium3 announcement describes technical cooperation; the new report describes acquisition discussions. Neither establishes that all Decart products would become exclusive to AWS, that prices would fall or that customers would automatically gain a new service. No acquisition announcement has set out such terms.
From prerecorded video to responsive video
Decart's Lucy is designed to transform video as it is created or streamed. The company presents character replacement, object addition and removal, background and style changes, and visual effects. Its intended users include creators, online retailers, advertisers and game developers. These are the company's capability descriptions, not evidence that every transformation works equally well across all footage, lighting conditions or source material.
The practical change is that editing can become part of the interaction. Replacing a background during a livestream, for example, differs from uploading a file and waiting for an edited result. The live case requires checks on motion continuity, input-to-output delay and the system's response to changing requests. An attractive frames-per-second figure does not answer those questions by itself, or reveal the total cost of running a service.
Oasis illustrates another aspect. In the research project Decart and Etched published in 2024, user actions influenced model-generated frames instead of relying on a conventional game engine. The researchers also described limitations, including long-term memory, object consistency and precise control. The demonstration explains a technical direction; it does not establish that a generated world is a reliable simulation for every industrial application.
Why the software between chips matters
The report's chip-portability detail adds a different dimension from buying an application. An AI accelerator choice affects model execution, memory use and delivery of results. Software that simplifies switching can broaden deployment options. Its existence does not, however, make hardware migration instantaneous, adaptation-free or equivalent in quality, availability and cost.
For a development team, the relevant test is its own workload. A short marketing clip, a continuous video stream and an interactive environment are different tasks. Comparing them requires consistent resolution, required quality and user numbers, while including integration time, data transfer and failure handling. This is a practical inference from the technology, not another Decart performance claim or an Amazon commitment.
Choosing between a model, a service and separate infrastructure matters too. A team may like a visual demonstration yet find the service unsuitable for its workflow or budget. Conversely, infrastructure improvements can matter even when the interface looks similar. An acquisition report in this field is therefore worth assessing through what developers can build, alongside the headline valuation and the company's most impressive video.
What changes now for developers and customers
For now, the reported talks are not a product update. They do not themselves change a contract, licence, price or regional availability. A business evaluating Decart should still base its decision on the actual offering: terms of use, treatment of uploaded content, integrations and support. The talks provide no new confirmation of Hebrew support or any particular product's availability in Israel. A possible ownership change does not resolve those questions.
The Israeli connection is a company founded by Israelis operating in global AI infrastructure. The report did not announce plans for jobs in Israel or changes to local operations. For Israeli readers, the immediate distinction is between two existing facts, technology and an AWS relationship, and a possibility still under discussion, a change of ownership. Keeping them separate prevents business talks from becoming an accomplished fact.
An announcement of an agreement, if one is signed, would change the picture by explaining what is being acquired and on what terms. Until then, around $7 billion is a possible valuation in a news report. The technical story is clearer: a responsive real-time experience depends on the model, software and hardware together. The business question is whether Amazon will buy that combination or continue working with it through a partnership.