What is reported, and what is not

Amazon is in talks to acquire Decart in a deal valued at around $7bn, the Wall Street Journal reported on Sunday, citing people familiar with the matter. Reuters, carrying the report, said a deal could happen soon if the talks do not fall apart or another buyer appears.

That is the whole of it. No agreement has been signed, no terms have been confirmed by either party, and neither company commented. Amazon declined to comment outside regular business hours; Decart did not respond to requests for comment. Everything below is reporting about a negotiation, not about a transaction.

A computer graphics card lying on a desk
Decart builds software for moving AI workloads between different makers' chips. Illustrative photograph. Trần Chính · pexels · Pexels License

What Decart does

The company is best known for two generative systems. Lucy, which Nvidia has backed, edits live video in real time. Oasis generates simulated environments used to train and test robotics and self-driving systems. Decart has also built software that lets developers move AI workloads between different makers’ chips, which is the part that explains why a cloud provider would want it: switching costs between accelerator vendors are the thing every buyer of compute currently complains about.

Decart had early access to AWS’s Trainium3 chip, and its chief executive used Trainium3 to demonstrate Lucy, so the commercial relationship predates the reported talks.

The price is the story

PitchBook data cited by the Journal valued Decart at close to $4bn in May, when it raised $300m in a round led by Radical Ventures with Nvidia joining as a new investor. A $7bn price would be roughly double that, five months later.

A pen resting on a blank notepad on a desk
A signed agreement and a price would confirm the deal. Illustrative photograph of a notepad and pen. Lukas Blazek · pexels · Pexels License

The comparison point that matters more is Anthropic. Bloomberg reported that Anthropic explored buying Decart in August at a figure around $6bn, and decided last month not to pursue it after due diligence. One buyer looked closely and declined; another is reported to be offering a billion dollars more.

That does not mean either party is wrong. Anthropic and Amazon want different things from the same asset: a model developer buying chip-portability software is buying flexibility, while a cloud provider buying it is buying a reason for customers to stay. But the sequence is worth holding on to if the deal is announced, because the diligence that produced two different answers is not public.

What would confirm it

A signed agreement, a price, and a statement from either company. Until then the checkable facts are the ones above: a reported negotiation, a reported number, an earlier process that ended without a deal, and an existing commercial relationship through Trainium.