Amazon in talks to buy Decart for about $7 billion: what the AI startup does and why AWS wants it

On Sunday, October 11, 2026, the Wall Street Journal reported that Amazon is in talks to acquire Decart, an Israeli-founded AI startup, in a deal valued at around $7 billion. Citing people familiar with the matter, the Journal said a deal could come soon "if talks do not fall apart or another suitor emerges." Nothing has been announced. Amazon declined to comment and Decart didn't respond, according to Reuters. The talks matter for AI datacenters because of what Decart sells alongside its video models: software that gets more work out of AI chips, including Amazon's own.

This story is based on Reuters' summary of the WSJ report, Decart's and Amazon's own pages, and coverage from Calcalist, The Next Web and others, checked October 12, 2026.

The short version

• Status: reported talks, not a deal. The only source is the WSJ, citing unnamed people.
• Price: around $7 billion, against a valuation of nearly $4 billion in May 2026 (the WSJ, citing PitchBook).
• What Decart does: real-time "world models" (Lucy, Oasis) plus the Decart Optimization Stack, an inference and training layer it says runs on Nvidia GPUs, Google TPUs and AWS Trainium.
• Earlier suitor: Anthropic explored a deal of about $6 billion in August and walked away after due diligence, Bloomberg reported on September 8.

What does Decart do?

Decart's name comes up in two different contexts, and both are accurate.

A world-model lab. On its website, Decart describes itself as "a frontier AI lab building real-time world models." Its two named products are:

• Lucy, which it calls a "world editing model" that transforms live video in real time, for virtual try-on, advertising, social apps and games.
• Oasis, a "realtime interactive World Model for physical AI," aimed at robotics, autonomous vehicles and drones. Oasis first got attention as a playable game whose video is generated as you play.

An infrastructure software company. Decart also sells the Decart Optimization Stack (DOS), which it describes as an "ultra-optimized inference and training stack" that "unlocks the full performance of every major chip." The site says it is "built across NVIDIA, AWS Trainium and Google TPU." Calcalist describes the technology as helping AI models run more efficiently and "move between chips from different manufacturers." Decart's performance claims are its own and haven't been independently verified. When it raised money in May, The Next Web reported a claim of 1,600 tokens per second for agentic inference.

The two halves are connected. Decart built the stack because real-time video generation only works if inference is extremely fast and cheap. The stack is what bigger buyers appear to want.

Why would Amazon want Decart?

Neither company has said. The public record points to Trainium.

• In its December 2, 2025 Trainium3 announcement, Amazon featured Decart as a customer, saying "Decart is achieving 4x faster inference for real-time generative video at half the cost of GPUs."
• When Decart raised money in May, it said its Lucy2 model runs on Trainium3. Nafea Bshara of Amazon's Annapurna Labs, which designs Trainium, said it exceeds 80% model FLOPS utilization, according to The Next Web.
• Calcalist, citing the WSJ, says AWS gave Decart early access to Trainium3 in 2025.

The usual obstacle to custom chips like Trainium is software. Most AI code is written and tuned for Nvidia's CUDA, and moving it to other chips takes months of engineering. A team that has shown it can make demanding real-time models run well on Trainium, and that sells a stack for doing so, would help AWS make its chips easier to adopt. That's an inference drawn from the companies' public statements. Neither Amazon nor Decart has confirmed it. Microsoft is making a similar bet with its own Maia 300 accelerator, and AMD recently bought World Labs, a world-model lab.

Who else wanted Decart?

According to Bloomberg, Anthropic explored acquiring Decart for about $6 billion in August, then decided after due diligence not to proceed. The Next Web, summarizing Bloomberg's September 8 report, says Anthropic's interest was in the optimization stack, as a way to serve more demand with the compute it already has, not in world models. Calcalist reports that Nvidia also looked at buying Decart before Anthropic did. That claim comes from a single outlet. The reasons Anthropic walked away haven't been made public.

There's an Amazon connection here too. Amazon is one of Anthropic's biggest investors and cloud partners. On April 20, 2026, it said it would invest $5 billion immediately and up to $20 billion more, as our Anthropic explainer details. Anthropic also runs on Trainium at scale through Project Rainier, which our gigawatt datacenter tracker covers. If Amazon buys Decart, the know-how that Anthropic weighed buying would sit inside AWS instead.

Decart's funding and the price

Last verified: October 12, 2026

At about $7 billion, Decart would be one of Amazon's larger acquisitions.

What's confirmed, reported and uncertain

• Confirmed (company pages): Decart's products and its description of DOS across Nvidia, TPU and Trainium; Amazon's December 2025 Trainium3 statement about Decart.
• Reported: the Amazon talks and the ~$7 billion value (WSJ), the May valuation (PitchBook via WSJ), Anthropic's ~$6 billion talks and withdrawal (Bloomberg), and Nvidia's earlier interest (Calcalist).
• Uncertain: whether a deal is signed, its structure and price, regulatory review, and what Amazon would do with Lucy and Oasis.