Amazon is in talks to acquire Decart, a San Francisco artificial intelligence startup, in a deal that could value the company at around seven billion dollars, the Wall Street Journal reported on Sunday, citing people familiar with the matter. Reuters, which carried the report, said a deal could be reached soon if the talks do not collapse and no rival bidder emerges.

The Decart deal would put a major AI startup acquisition on the table. Decart builds software that lets developers shift AI workloads between chips from different vendors, plus a pair of "world models" — systems trained to understand and simulate the physical world. One of them, Lucy, edits live video in real time. The other, Oasis, generates interactive 3D environments for training and testing robots and self-driving systems.

The deal on the table

According to the Journal's reporting, Amazon and Decart have been negotiating a purchase price that would nearly double the startup's spring valuation. The talks are private, and the people briefed on them cautioned that they could still fall apart.

Neither company has confirmed the discussions. Amazon declined to comment outside regular business hours, while Decart did not respond to requests for comment, Reuters reported.

Why Decart drew Amazon's interest

Decart was founded in 2023 by three Israeli engineers. Its core product is software that removes the expensive, time-consuming work of tuning an AI model to run on a specific processor, so the same workload can move between chips from different suppliers. As demand for computing power outruns supply, enterprises have grown eager to avoid dependence on a single GPU maker, and portability has become a genuine commercial edge.

On top of that infrastructure play, Decart has pushed into world models, the corner of AI research concerned with systems that learn the rules of physical space — geometry, physics, cause and effect — and use them to simulate environments. Lucy processes and edits video as it streams, a capability the company says is backed by Nvidia. Oasis builds the kind of simulated worlds that let engineers train a robot or an autonomous vehicle thousands of times before it ever moves in the real world. The same technology has drawn interest from gaming and video production.

Investors have already priced in the momentum. Decart was valued at nearly four billion dollars this spring, according to PitchBook data cited by the Journal, and it raised three hundred million dollars in a funding round led by Radical Ventures that brought Nvidia in as a new investor.

A second look after Anthropic walked away

Amazon is not the first major AI company to take a hard look at Decart. Bloomberg News reported last month that Anthropic explored acquiring the startup in August at a price near six billion dollars, then decided against the deal after conducting due diligence.

The higher figure now attached to the Decart deal tops the price Anthropic reportedly considered. The logic for Amazon is concrete. Its cloud arm, AWS, sells customers on running AI workloads on a mix of hardware, including its own Trainium and Inferentia chips alongside Nvidia's. Software built specifically to make that mixing painless would strengthen the pitch.

World models fit Amazon's business too. The company uses robots across its warehouses, and simulation is a standard way such systems get trained. Amazon has also been expanding its investments in AI-generated media. Owning the models that generate simulated environments — and the software that runs them on any chip — would give Amazon another layer of control over the AI stack.

What comes next

For now, the talks remain talks. Large technology acquisitions have faced tougher antitrust scrutiny in recent years, and a purchase of this size would be expected to draw regulatory attention. The Journal's sources stressed that the negotiations could still break down.

The discussions also point to where the industry's attention has moved: past chatbots and into AI that handles the physical world — video edited live, environments simulated, robots trained before they move. If the Decart deal closes, the software that decides which chips tomorrow's AI runs on would increasingly sit inside its cloud.