Amazon in Talks to Buy Decart for About $7 Billion

Amazon is in talks to acquire Decart, an Israeli AI startup that builds real-time video and world models, in a deal valued at about $7 billion, The Wall Street Journal reported on Sunday, 11 October, citing people familiar with the matter. According to Reuters’ account of the report, the Journal said an agreement could come soon, provided the talks do not fall apart and no other suitor emerges. Amazon declined to comment outside regular business hours, and Decart did not respond to requests for comment. Nothing has been signed.

The price would be a big step up. The Journal put Decart’s valuation at nearly $4 billion in May, citing PitchBook data, when the company raised $300 million in a round led by Radical Ventures with Nvidia joining as a new investor. BitsMinds covered that round, which took Decart’s total funding past $450 million less than three years after its founding in 2023.

The second buyer in two months

Amazon is not the first company to try. In August, Bloomberg reported that Anthropic was in talks to buy Decart for about $6 billion, a deal BitsMinds wrote up at the time as the biggest acquisition Anthropic had ever considered. On 7 September, Bloomberg reported that Anthropic had dropped the deal after completing due diligence, as The Deep Dive summarised; neither company explained what changed, and a person familiar with the talks said the two could still work together in other ways. If the Journal’s figure holds, Amazon would be paying roughly $1 billion more than the price Anthropic walked away from.

What Decart makes

Decart’s website describes three product lines. Lucy is a real-time video editing model that changes what a camera sees while it is still rolling: swapping a person for a character, adding or removing objects, restyling a background. The company pitches it for virtual try-on in shopping, product placement in advertising, live streaming and gaming, and lists its latest version, Lucy 2.5, as running live at 30 frames per second.

Oasis, now in its third version, is a world model that generates interactive, physically plausible environments for training and testing robots, autonomous vehicles, factory systems and drones. The third line, the Decart Optimization Stack, is the least visible and arguably the most strategic: hardware-level software that helps chip makers and AI teams run models faster and at higher utilisation across Nvidia GPUs, AWS Trainium and Google TPUs.

Already on Amazon’s chips

The two companies already have a working relationship. Decart appears on the AWS Trainium customers page, where co-founder and chief executive Dean Leitersdorf credits Amazon’s in-house chip with up to four times the frame throughput and twice the cost efficiency of top GPUs on Decart’s real-time video models, with latency cut from 40 milliseconds to 10. The same entry says the capabilities are expected to reach AWS customers through Bedrock, Amazon’s hosted model service.

That history is the clearest reading of why Amazon would pay this much, though neither company has said so. Decart is one of the few model makers that has publicly shown its workloads running better on Trainium than on Nvidia hardware, and its optimisation stack is the kind of software that makes a non-Nvidia chip easier to sell. Lucy’s try-on and ad-insertion use cases map neatly onto Amazon’s retail and advertising businesses, and Oasis is built for the robotics simulation that warehouse automation depends on. Nvidia, an investor since May, would see a portability layer it helped fund end up inside its biggest cloud customer’s chip effort.

For now the deal exists only in the accounts of people familiar with the talks, and Decart has already been one signature away from a sale once this year. Whether Amazon gets further than Anthropic did should become clear quickly: the Journal’s sources expect an answer soon.