Amazon in $7B Talks to Acquire AI Chip Startup Decart

Amazon is in talks to acquire the AI Chip Startup Decart, whose software lets AI models run across different chip architectures, for about $7 billion, according to the Wall Street Journal. The reported negotiations, which surfaced on October 11, would mark one of the largest AI infrastructure deals of the year and signal how urgently the biggest cloud providers want to loosen their dependence on a single chip supplier.

People familiar with the matter told the Journal that a deal could be signed soon unless the talks collapse or another bidder emerges. Neither Amazon nor Decart responded to requests for comment, according to reports citing the story. Reuters said it could not independently verify the report.

Why Amazon wants Decart's chip-flexibility software

Decart was founded in San Francisco in 2023 by three Israeli engineers — brothers Dean and Orian Leitersdorf and Moshe Shalev. Its core technology tackles a practical bottleneck in the AI boom: most AI models are built for specific chips and require extensive performance tuning to run anywhere else. Decart's software removes much of that burden, letting developers move models between chip types without rebuilding them.

That capability has become strategically important because surging AI demand has made Nvidia graphics processing units both scarce and expensive, according to the Seoul Economic Daily, which republished an AFP/Yonhap account of the Journal's reporting. Relying on one type of chip leaves companies exposed to supply shortages and price increases, so the ability to switch chips flexibly is key to securing computing power reliably and holding down costs. The chip-flexibility angle is part of a broader wave of AI-driven hardware reshuffling, as surging model demand collides with limited GPU supply.

Amazon has been developing its own answer to this problem. The company is building Trainium3, its custom AI chip designed to compete with Nvidia GPUs. According to the Journal's account, AWS officials met Decart's founders, came away impressed, and gave the AI Chip Startup Decart early access to Trainium3 in 2025. Trainium, which Amazon introduced in 2020, is now used through AWS by customers including OpenAI, Anthropic and Uber, and in April Amazon said revenue commitments tied to the chips had surpassed $225 billion.

Decart would also bring deep-pocketed backers' attention: according to market research firm PitchBook, the startup has raised more than $400 million from investors including Sequoia Capital, Benchmark and Radical Ventures, and was valued at about $4 billion as of May. A later funding round brought in Nvidia as a new investor, which suggests the chip giant itself sees Decart's portability layer as worth owning a piece of.

Amazon would not be the first suitor. Anthropic reportedly explored acquiring Decart for around $6 billion in August but ultimately decided against it, according to The Jo AI, which cites the Journal's reporting. For the AI Chip Startup Decart, a $7 billion exit would represent a significant premium over both its last valuation and the figure Anthropic walked away from.

World models, Trainium3, and a $7 billion question

Beyond chip portability, the AI Chip Startup Decart is developing what it calls world models — AI systems that learn the rules of the physical world, such as physics and spatial structure, to generate and simulate three-dimensional environments. According to the Seoul Economic Daily's account, these models are used to train and test robots and self-driving systems in simulation before deployment in the real world, as well as in gaming and video production. The AI Chip Startup Decart's Lucy model can edit live video in real time, while Oasis builds simulation environments for training and testing robotics and autonomous driving systems.

For the AI Chip Startup Decart, that combination — infrastructure software that runs models on any chip, plus world models for robotics — would give Amazon two assets in one acquisition. The company has been pushing to sell Trainium chips directly into other companies' data centers, not just as part of its cloud service, and Google and Microsoft are accelerating their own custom silicon for the same reason. Owning Decart's portability layer could make Trainium an easier sell to customers worried about being locked into a single vendor's hardware.

Big-money AI deals are becoming routine — Salesforce's $3.6B Fin deal closed just this week — but the reported $7 billion tag for the AI Chip Startup Decart would rank among Amazon's largest acquisitions ever, exceeding the $6.5 billion purchase of the MGM film and television studio in 2022. Yet the structure of the reporting carries the usual caveats: talks are still in progress, another bidder could appear, and neither company has confirmed anything publicly. Deals at this scale and stage still fall apart regularly.

Still, the direction of travel is clear. As AI models get bigger and compute stays expensive, the most valuable companies in the stack may not be the ones that own the scarcest chips — but the ones, like the AI Chip Startup Decart, that make it cheap and easy to use everyone else's. The AI Chip Startup Decart sits exactly in that layer, and $7 billion is Amazon's estimate of what that optionality is worth. For more context on the original reporting, see the Seoul Economic Daily's write-up and The Jo AI's coverage.