Companies · · 3 min read
Editor: Bohdan S.
Amazon nears deal worth up to $7.5 billion for AI startup Decart
Amazon is close to buying Israeli AI firm Decart, the FT reported, a move that would give its cloud arm software built to help its in-house chips take on Nvidia.
The short answer
The Financial Times (FT) reported on Saturday that Amazon is close to a $7.5 billion purchase of AI startup Decart, a company whose software makes it easier to run AI models on chips other than Nvidia's. Tech news site Calcalist separately reported the deal could top $7 billion in cash, stock and deferred payments. Neither company has announced anything.
What happened
The Financial Times (FT) reported that Amazon is nearing an agreement worth $7.5 billion to acquire Decart, framing the deal as part of the rivalry between Amazon Web Services (AWS) and Nvidia. Earlier in the week, Calcalist said the deal had cleared most of the legal issues that slowed talks and could be signed within days. According to Calcalist, Amazon would pay more than $7 billion through cash, stock and deferred payments, with performance-linked amounts lifting the total close to $7.5 billion. The founders would commit to staying for five years, with a large slice of the money paid only in year five. Neither company has confirmed a deal.
What it means
Decart, founded in Israel, sells software called DOS that shortens the work of tuning an AI model for a particular chip. SiliconANGLE reported that the job normally takes months for every processor a model needs to run on, and Decart says its platform cuts that to a few weeks. Cheaper tuning makes it easier for companies to shift AI workloads from one type of chip to another. The firm also builds two world models, Lucy and Oasis, that edit live video or generate interactive 3D environments.
That chip-switching ability is the strategic prize. Amazon designs its own AI processors, called Trainium, as an alternative to Nvidia's graphics processing units, which still dominate the market for AI hardware. Calcalist noted that the ability to move between chips is gaining importance as companies scramble for computing capacity, so tools that make switching cheaper could make Trainium easier to adopt. Decart is already a customer: when Amazon launched Trainium3 servers in December 2025, it said Decart was generating real-time video frames four times faster at half the cost of GPUs.
The price marks a sharp jump. Decart raised $300 million in May at a valuation of nearly $4 billion, in a round led by Radical Ventures that included Nvidia, Adobe Ventures and Toyota Ventures. Calcalist said earlier talks valued the company at $5 billion to $6 billion, and that Anthropic had pulled out of a proposed acquisition in early September after largely finishing its due diligence, the detailed checks buyers run before a deal. Decart then tried to restart talks with Nvidia, which declined, before Amazon stepped in.
For Amazon, the deal would rank among its larger purchases. Calcalist noted that it paid about $6.5 billion for the MGM film studio in 2022 and agreed this year to buy satellite company Globalstar in a deal valued at around $11 billion. Decart is still small, though: Calcalist said estimates put its revenue in the low tens of millions of dollars, mostly from projects for semiconductor companies.
The bull case
Owning Decart could make Trainium a more practical choice for AI developers by lowering the cost of moving models off Nvidia hardware, which might draw more computing demand to AWS. The deferred, target-linked payments and five-year founder commitments tie much of the price to Decart's team staying and delivering, which reduces the risk that Amazon overpays for talent that later leaves.
The bear case
The reported price sits far above the valuation Decart carried in May, for a company with revenue estimated in the low tens of millions. Anthropic walked away after doing its checks, and Nvidia declined to revive talks. Until a deal is signed and announced, it could still fall through, and software alone may not be enough to loosen Nvidia's hold on AI developers.
Why it matters
For investors
The deal is small next to Amazon's overall size, so its main market relevance is as a signal of how hard cloud giants are working to reduce their dependence on Nvidia chips.
The wider context
Big technology companies are spending heavily to control more of the AI supply chain, and competition among chip suppliers could eventually shape what businesses and consumers pay for AI services.
Market impact
| Asset (ticker) | Possible move | Timeframe | Why |
|---|---|---|---|
| Amazon (AMZN) | ↑ Up | Long term | Decart's chip-switching software could make Amazon's Trainium processors easier for AI developers to adopt. |
The FinPrism newsroom's view of how these figures could move, as of . It is general news commentary, not a recommendation to buy, sell or hold anything.
Frequently asked questions
Is Amazon buying Decart?
Not officially yet. The Financial Times reported on Saturday that Amazon is close to a deal worth $7.5 billion, and Calcalist reported earlier in the week that signing could come within days. Neither Amazon nor Decart has announced an agreement, so the terms could still change or the talks could collapse.
What does Decart do?
Decart makes software called DOS that speeds up the work of optimizing AI models for different chips, making it easier to move workloads between hardware. It also develops world models named Lucy and Oasis, which change live video in real time or create interactive 3D environments. It raised $300 million in May at a valuation of almost $4 billion.
Why does Amazon want Decart?
Amazon builds its own AI chips, called Trainium, to compete with Nvidia. Decart's software reduces the time and cost of tuning AI models for a new chip, which could make Trainium a more practical alternative to Nvidia's dominant processors. The FT linked the reported deal to Amazon's rivalry with Nvidia.
Sources: Financial Times, Calcalist (CTech), SiliconANGLE, Amazon
How this was made: written with AI assistance from the sources listed above, following our editorial standards.