Amazon negotiates $7.5 billion acquisition of AI startup Decart
The deal aims to help Amazon reduce its reliance on Nvidia by acquiring technology that allows AI models to run on various types of chips.
The short version
- Amazon is negotiating an acquisition price of approximately $7.5 billion for Decart.
- Decart specializes in 'world models' that learn physical laws to generate real-time video and virtual environments.
- The company's software can help run AI models on different chips without the heavy adjustments usually required by Nvidia's CUDA platform.
- Decart was founded in 2023 by three Israeli engineers, including graduates of the military intelligence unit 8200.
- The startup's valuation has grown from $500 million in late 2024 to a potential $7.5 billion in less than a year.
Why it matters
This move signals a major shift in the AI infrastructure war, where hardware providers are trying to escape the 'walled garden' of Nvidia's software. For users, this could eventually lead to more diverse and cheaper hardware options for running complex AI models.
The $7.5 billion deal
Amazon is currently in the final stages of negotiations to purchase Decart, an AI startup founded by Israeli engineers. As reported by The Chosun Ilbo on MSN, the deal is valued at approximately $7.5 billion, which is roughly 10 trillion Korean won. This acquisition comes at a time of growing skepticism regarding AI investments and a cooling U.S. stock market.
The valuation of Decart has seen an extraordinary trajectory since its founding in 2023. In October 2024, the company secured a seed investment of $21 million, and by December, it reached a valuation of over $500 million. By May of this year, that valuation jumped to $4 billion, and the current negotiations would nearly double that figure again within six months.
While the $7.5 billion figure is the current target, it is notably higher than previous estimates. The Wall Street Journal reported an estimated acquisition price of $7 billion in October. Because neither Amazon nor Decart has officially confirmed the final price, there is still a possibility that the negotiations could fall apart before they are finalized.
World models and real-time tech
Decart is known for developing 'world models,' which are AI systems designed to understand the physical laws and spatial structures of our reality. These models allow the AI to generate videos or virtual environments that behave in ways that mimic the real world. Two of their primary products are 'Lucy' and 'Oasis.'
Lucy is a tool capable of editing videos in real time, responding to user inputs in as little as 30 milliseconds. Oasis, which was unveiled in 2024, creates virtual worlds that users can interact with. One of the most notable features of Oasis is its ability to generate screens in real time without the need for a traditional game engine.
The startup was established by three Israeli engineers, including Moshe Shalev and brothers Dean and Orian Laitersdorf. The founders are graduates of Unit 8200, a military intelligence unit in Israel. This background in high-level intelligence likely informs their approach to complex spatial computing and rapid technological development.
Breaking the Nvidia monopoly
Industry analysts suggest that Amazon is less interested in the video generation technology and more interested in Decart's chip optimization. Currently, Nvidia maintains a dominant position because of its proprietary software platform called CUDA. Running AI models built for CUDA on different types of hardware usually requires significant and difficult adjustments for developers.
Decart offers a solution called the Decart Optimization Stack (DOS). This software allows developers to use various types of AI chips interchangeably, significantly reducing the burden of moving models away from Nvidia's hardware. By acquiring this technology, Amazon can better support its own self-developed AI chips, such as the Trainium3 released last year.
The acquisition is a strategic move to reduce Amazon's dependence on Nvidia's graphics processing units (GPUs). By creating a way for models to run smoothly on different chips, Amazon can build a more flexible infrastructure. This creates a unique situation where a startup that received investment from Nvidia could end up being used by Amazon to counter Nvidia's market position.
The path forward for Amazon
If the deal is completed, it will provide Amazon with a critical piece of the puzzle for its independent AI hardware strategy. The company is looking to secure computational resources in an increasingly competitive market. Integrating Decart's optimization stack could make Amazon's Trainium chips more attractive to developers who are currently locked into Nvidia's ecosystem.
The success of this acquisition depends on how well Amazon can integrate the DOS software into its existing cloud services. If successful, it could lower the barrier to entry for companies wanting to run large-scale AI models without paying the 'Nvidia tax.' It also positions Amazon as a primary alternative for enterprise-level AI infrastructure.
For the broader tech industry, this move highlights the importance of software interoperability. As more companies try to build their own silicon, the ability to move models between different chips becomes a vital commodity. Decart's technology serves as a bridge that could allow for a more diverse and competitive hardware landscape in the coming years.
What to watch
- Whether the final acquisition price will exceed the $7.5 billion negotiation target.
- How quickly Amazon can integrate the Decart Optimization Stack into its cloud services.
- The reaction from Nvidia to the acquisition of a startup they previously invested in.
Try it
Look up 'CUDA' and 'Trainium3' to see the difference between Nvidia's proprietary software and Amazon's custom AI chip.
Sources and links
Read the original at The Chosun Ilbo on MSN
First reported by The Chosun Ilbo on MSN. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.