Amazon is negotiating to buy AI startup Decart in a deal valued at about $7 billion, according to Bloomberg.
The talks are ongoing, and no formal agreement has been reached. The Wall Street Journal, citing people familiar with the situation, reported the negotiations on October 11, 2026, and described them as early-stage but apparently moving toward a conclusion.
Reuters also reported on the talks. It noted, however, that it could not independently verify the details.
What Amazon would actually be buying
Decart’s core business is making AI models run better. Its main product is the Decart Optimization Stack, or DOS, a software layer that improves how AI models perform across different hardware.
The supported platforms tell the story. DOS works across NVIDIA GPUs, Google’s TPUs, and Amazon’s own Trainium chips.
Decart also builds its own models. Lucy is a real-time video and world model aimed at uses like virtual try-ons, the kind of feature that lets shoppers see clothing on themselves without leaving the couch.
Advertisement
The second model, Oasis, is an interactive world model. It generates environments that respond to user input, a category drawing growing attention from AI labs exploring simulation and gaming.
A startup with a crowded waiting room
Decart was founded on September 7, 2023. It employs more than 60 people and has raised over $450 million.
Amazon is not the first suitor. Decart has previously drawn interest from Anthropic and NVIDIA, according to the research compiled on the deal.
Anthropic reportedly weighed a $6 billion deal earlier in 2026. It walked away after due diligence and exited the process in September 2026.
So Amazon’s reported offer sits about $1 billion above where Anthropic was reportedly looking.
There is an interesting wrinkle here. Amazon has been a major backer of Anthropic, which means the startup’s reported pass did not keep Amazon itself from pursuing the same target.
Why compute is the real prize
The timing ties directly to the compute supply crunch hitting much of the AI industry. Companies building and running models are competing for limited chip capacity, and that scarcity has made efficiency software unusually valuable.
Hardware flexibility adds another layer. Being able to move AI workloads between NVIDIA, Google, and Amazon silicon gives customers options at a time when depending on a single supplier carries real risk.
There is an obvious tension, of course. A neutral optimization layer that works well on rival chips becomes a little less neutral once one of those rivals owns it.
What to watch from here
The first question is simply whether the deal closes. Talks that are reportedly progressing are not the same as a signed agreement, and Decart already has one high-profile suitor in its rearview mirror.
Due diligence is the obvious pressure point. Anthropic reportedly withdrew after that stage, so the market will be watching whether Amazon reaches a different conclusion with the same set of books.
Customers of Decart’s tools have their own stake in the outcome. Developers currently using DOS on NVIDIA or Google hardware will want to know whether cross-platform support stays a priority under Amazon ownership, or quietly tilts toward Trainium over time.
For now, the deal remains a negotiation, with a reported price of about $7 billion and no signature on the page.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.