← Back to feed

Amazon is in talks to acquire Decart for approximately $7 billion, according to reporting first published October 11, 2026. Amazon and Decart declined to comment; no deal has been announced and talks could still collapse or attract a competing bid.

The price is 75% above Decart’s May 2026 valuation implied by a $300 million round led by Radical Ventures. Decart has about 100 employees and has not disclosed revenue. An outside estimate puts annual revenue in the low tens of millions.

Two Products, One Real Prize

Decart’s public-facing product is real-time generative video. Oasis generates interactive worlds; Lucy edits live video as it plays. Neither business justifies $7 billion on its own.

The second product does. The Decart Optimization Stack is software that makes AI models run faster and cheaper across Nvidia GPUs, Google TPUs, and Amazon’s Trainium chips without requiring the model to be rebuilt for each chip. Amazon became a Decart strategic customer in May 2026. Decart’s CEO has stated publicly that Trainium3 runs Decart models at half the cost of equivalent Nvidia GPU workloads — a claim based on the company’s own benchmarks.

The CUDA Problem

Amazon’s chip strategy is the context that makes $7 billion legible. Amazon’s chip revenue — Trainium and Inferentia together — passed a $25 billion annualized run rate in the second quarter of 2026. AWS spent $54.2 billion on capital expenditure in that same quarter, the bulk of it on infrastructure built around AI workloads.

Trainium is competitive on paper. The barrier is software. Nvidia’s CUDA ecosystem represents two decades of library development, tooling, and developer muscle memory. Migrating AI workloads to any non-Nvidia chip requires developer time and risk. The Decart Optimization Stack, if it performs as claimed, removes that migration cost — models run on Trainium without a rewrite.

If Decart’s software can shift even a fraction of AWS’s AI workloads from Nvidia to Trainium, the margin savings at scale dwarf the acquisition price. Amazon’s $54.2 billion quarterly capex gives the scale of what is at stake.

Anthropic Said No

Anthropic reportedly evaluated a Decart acquisition in August at approximately $6 billion and walked away after due diligence in September. A sophisticated buyer passed at a lower price than Amazon is now discussing, for reasons that have not been disclosed publicly.

That detail matters. Anthropic has deep interest in inference efficiency — it is simultaneously Decart’s type of customer and a company that understands inference optimization at a technical level. An informed buyer deciding the technology was not worth $6 billion is a material data point against the deal’s upside case.

Stakes

For Amazon the bet is about Trainium adoption velocity more than it is about any video model. Decart’s optimization stack is essentially a CUDA compatibility layer that competes with Nvidia’s moat using Nvidia’s own customers. If it works at scale, it accelerates the timeline on which AWS can route AI inference away from Nvidia silicon.

For the market, a $7 billion exit for a 100-person company with tens of millions in revenue underscores how much enterprise buyers are willing to pay for credible infrastructure software in AI. The price is not a valuation of what Decart has built. It is a valuation of what Amazon loses if a competitor acquires the tool instead.

Decart was founded in 2023.