Amazon는 Nvidia 칩과 Amazon 자체 칩 모두에서 AI 구동 속도를 높이는 기술을 보유한 설립 3년 차 AI 스타트업을 70억 달러에 인수하려 하고 있다.

Amazon is reportedly closing in on a $7 billion deal for a startup with barely 100 employees and no disclosed revenue, and the real prize has nothing to do with the flashy video models everyone is talking about.

Amazon (NASDAQ:AMZN) is reported to be in talks to pay around $7 billion for Decart, a company with about 100 employees and no disclosed revenue. The talks were reported on October 11, 2026, and no deal had been announced as of October 12, 2026. Amazon and Decart both declined to comment, and the deal could still fall apart or draw another bidder.

Amazon shares closed at $259.53 on October 12, a session move of -1.11%, so the market did not treat the reported talks as an obvious win. The price only makes sense if Decart is valued as a way to move AI developers off Nvidia (NASDAQ:NVDA) software, which is worth far more to Amazon than its video models.

**Two Businesses Inside One Small Startup**

Decart's visible business is real-time generative video. Its Oasis model generates interactive worlds, and Lucy edits live video as it plays.

The second business is the Decart Optimization Stack, software that makes AI models run faster and cheaper on Nvidia GPUs, Alphabet (NASDAQ:GOOGL) TPUs, and Amazon's Trainium chips without rebuilding the model for each chip. That is what Amazon would most likely pay for if the talks lead to a deal.

Nvidia's CUDA software, the programming layer most AI models are written and tuned for, is the main reason developers stay on Nvidia hardware. Amazon has built Trainium for years, but getting customers to port their models has proven harder than building the chip.

Amazon became a Decart strategic customer in May 2026. Chief executive Dean Leitersdorf has said Trainium3 runs Decart's models at half the cost of GPUs, a claim the company made about its own results.

**A Small Price for a $2.80 Trillion Company**

Against Amazon's $2.80 trillion market value, the reported price is small. Amazon spent $54.2 billion on capital expenditures (data centers and equipment) in the second quarter of 2026 alone.

Its chips business passed a $25 billion annualized revenue run rate that quarter. If Decart's software could shift even a small share of AI workloads onto Trainium, the saved Nvidia margin could cover the reported price.

**A Rival Looked at Decart and Walked Away**

Anthropic reportedly considered a Decart deal at about $6 billion in August and walked away after due diligence in September. A sophisticated buyer passed at a lower price than the one now being discussed, for reasons that went unreported.

Decart has not disclosed revenue, and the only outside estimate puts it at several tens of millions of dollars. That would make the reported price a very high multiple of Decart's sales.

The reported figure is about 75% higher than the May 2026 round, which raised $300 million led by Radical Ventures. An earlier round valued Decart at $3.1 billion.

Nvidia is also a Decart investor, which complicates the idea that a deal would be a move against Nvidia. Nothing reported explains how the price splits between video and optimization, as there has been no announcement, no disclosed terms, and no regulatory filing.

**What the Decart Talks Mean for Amazon's Investment Case**

The Decart talks change little about Amazon's investment case. That case rests on AWS revenue of $42.2 billion, up 36.7% year over year in the second quarter, plus a backlog of $496 billion.

The stock trades at a forward P/E of 24x. The average analyst target of $331.53 stands well above the October 12 close, which suggests analysts expect AWS growth to carry the shares regardless of Decart.

If the talks produce a deal, Decart would give Trainium useful software support. If they collapse, Amazon would still have its existing customer relationship with Decart, so the downside for shareholders looks limited.

Watch whether Amazon announces a signed agreement before its next earnings report and whether it discloses Trainium adoption figures on that call. No deal and no new Trainium customers would weaken the case that Trainium can compete with Nvidia.

Contact [email protected] for any questions or corrections.