Amazon Nears a $7.5 Billion Deal for Decart

The term sheet has been through the lawyers. What is left is a signature, possibly before the weekend closes.

Amazon has cleared most of the legal obstacles to buying Decart, the Israeli artificial-intelligence startup, according to a report on November 18 in Calcalist, the Israeli business daily. The consideration starts above $7 billion and rises toward $7.5 billion if performance targets are met. It is a mix of cash, stock and payments deferred into the future.

The structure tells the story. The three founders have committed to stay for five years, and most of the money is not payable until the fifth year. That is not a normal acquisition of a product line. It is a purchase of people whose work the buyer cannot yet see, secured by a schedule that makes leaving expensive.

Decart was founded in 2023 and built its reputation on models that generate video in real time, including interactive worlds that respond as the user moves through them. The company has published research on running those models on hardware that consumer devices can afford, which is the part that matters to a cloud provider selling to game studios, advertisers and robotics firms.

The road to this point was not straight. Anthropic opened talks with Decart, completed due diligence in early September and walked away from the table. Nvidia was approached about reviving the discussions and declined. Amazon then took the deal over and moved quickly, according to the Calcalist account, which cited people familiar with the negotiations.

The price is large for a company that has not shipped a mass-market product. It is larger than what Amazon paid for MGM in 2022, previously its biggest purchase, and larger than most venture-backed AI deals of the past two years. Amazon declined to comment.

Amazon’s position in the model business is unusual among the large cloud providers. It sells access to other companies’ models through Bedrock, its own Nova family of models, and its own chips, Trainium, to customers who want to avoid paying Nvidia margins. What it has not had is a research team working on generative media at the frontier, the kind of group that produces both products and recruiting credibility.

That gap is expensive to close by hiring. A senior researcher at a frontier lab can command several million dollars a year in compensation, and a team assembled one hire at a time takes years to cohere. Buying a company with twenty or thirty people who already work together collapses that timeline, which is why the effective price per engineer at the top of the market has run into the tens of millions.

Microsoft, Google and Meta have all signed arrangements of this shape since 2024, paying for teams and licenses rather than for equity, and antitrust officials in Washington and Brussels have begun to ask whether some of those arrangements amount to acquisitions that were never notified. A clean purchase at a disclosed price is, by comparison, the easier file to close.

The earnout does the rest of the work. Founders who must remain five years to collect most of their money are, in practice, employees with an unusually large retention bonus. The buyer gets continuity; the seller gets a valuation that reflects a future the market has not yet priced.

The risk sits in the fifth year as well. Deferred consideration is only as good as the targets attached to it, and targets set in 2026 for a technology moving this fast are guesses. If the models are overtaken, the buyer can decline to pay the balance and the founders can leave, having collected the cash portion and the stock. Both sides know this when they sign.

Regulators will look at the deal, though probably not for long. Amazon is not a dominant supplier in generative video, and the target’s revenue is small. The more interesting question is jurisdictional: Israel’s innovation authority has historically reviewed exits of companies that received government grants, and the Committee on Foreign Investment in the United States has taken an interest in transactions involving dual-use technology. Neither has been reported as an obstacle so far.

The Israeli angle matters to Amazon beyond the technology. The country’s AI sector has produced a steady supply of infrastructure and security startups, and Amazon’s cloud unit operates regions there. Buying a team in Tel Aviv is also a way to deepen relationships with the local venture community, which has been the source of several of the company’s earlier acquisitions.

For Decart’s investors, the outcome would be a fast and generous return. For Amazon, it is a bet that the next phase of AI competition is about generating video and simulated environments at a cost that makes them ordinary, rather than a text interface that everyone already has.

What the deal does not settle is whether Amazon can keep the team after the fifth year, or whether the founders will do what founders usually do once the schedule is complete and the earnout is paid. The answer will not be visible in the announcement.

Related Posts

  • November 20, 2026
  • 20 views
Sandisk Completes Eight-for-One Split as NAND Outlook Darkens

Eight for one. Sandisk’s stock split took effect on Nov. 19, and the shares began trading on the adjusted basis on Nov. 20, a change that divides the price by…

  • November 20, 2026
  • 16 views
A Lawmaker’s Privacy Suit Puts Model Design on Trial

For three years British courts have sidestepped the question of whether the company that builds an AI model answers for what its users produce with it. A written defence filed…