Amazon is in talks to acquire Decart, an AI startup whose software makes models run efficiently across different companies’ chips, in a deal that could be valued at around $7 billion. That is according to a Wall Street Journal report on Sunday, reported by Reuters, citing people familiar with the matter. The talks could produce a deal soon, the Journal said, but they could also fall apart, or another suitor could emerge. Neither Amazon nor Decart responded to Reuters’ requests for comment, and Reuters said it could not immediately verify the report.
That caution matters, because Decart has been here before. This is the third time in a few months that a giant of the AI industry has circled the company, and the previous two approaches ended without a sale.
What Amazon would be buying
Decart, founded in 2023 by brothers Dean and Orian Leitersdorf and Moshe Shalev, sells two things. The first is the Decart Optimization Stack, or DOS, a platform designed to make AI models run faster and switch more easily between chips from different manufacturers. In its own announcement of DOS 2.0, the company says the stack handles large language model, agentic, video and world-model workloads across Nvidia GPUs, Google TPUs and Amazon’s Trainium chips. The performance figures in that announcement — more than 1,600 tokens per second for agentic inference, and full-HD video inference at up to 100 frames per second — are the company’s own claims, not independent tests.
The second product line is Decart’s own models. Lucy can edit live video in real time, according to the company’s website. Oasis generates simulated environments that can be used to train and test robotics and autonomous-driving systems. Both models run on the same optimization stack Decart licenses to customers, which is part of the pitch: the software’s value is easiest to show on demanding, real-time workloads of its own.
A crowded auction
The reported $7 billion price would be a steep step up from Decart’s last public valuation. The company raised $300 million in May at a valuation of nearly $4 billion, in a round led by Radical Ventures that brought Nvidia in as a new investor, and it has raised roughly $450 million in total, according to CTech.
Since then, buyers have come and gone. CTech reports that Nvidia initially pursued an acquisition before Anthropic emerged as the preferred bidder in August, in a deal Bloomberg put at potentially $6 billion. Last month, Anthropic walked away after completing due diligence. The reasons for that decision were not publicly established, and this article does not guess at them. The pattern will be familiar to readers of this site: as with the reported Nvidia talks over Reflection AI, an early-stage approach at a headline price is a long way from a signed acquisition.
Why Amazon, specifically
Amazon is not a cold caller here. Amazon Web Services is already a strategic customer of Decart’s, and the Journal reported that AWS gave the startup early access to Trainium3, Amazon’s third-generation AI chip, in 2025, after being impressed by its founders.
That relationship points to the strategic logic, which is analysis rather than anything the companies have announced. Amazon builds Trainium as its in-house alternative to Nvidia’s accelerators, but chips alone do not move customers: software that lets a model run well on several vendors’ hardware lowers the risk of trying non-Nvidia silicon in the first place. A tool that squeezes more work out of expensive compute, and that has already been tested inside Amazon’s own ecosystem, fits that problem directly. Decart’s cross-chip ability is valuable precisely because the industry’s largest buyers are trying to reduce their dependence on a single dominant supplier.
At roughly $7 billion, a completed deal would also rank among Amazon’s largest acquisitions in recent years — larger, on CTech’s comparison, than the roughly $6.5 billion it paid for the MGM studio in 2022.
What is not known
Almost everything that would make this a done deal. There is no announced agreement, no confirmed price, and no comment from either company. The last suitor to get this far examined the business closely and chose not to proceed. Until Amazon or Decart says otherwise, this is a negotiation reported by one newspaper’s sources — a serious one, involving a customer that already knows the technology, but a negotiation all the same.
