Back to list

AI Business36氪

Anthropic Dropped Out After Due Diligence, Why Is Amazon Still Chasing Decart?

36氪1 min read
Read original
Anthropic Dropped Out After Due Diligence, Why Is Amazon Still Chasing Decart?

AI Summary

Amazon is reportedly negotiating to acquire Israeli AI startup Decart for about $7 billion. Anthropic had approached Decart at a valuation of about $6 billion but dropped the deal after due diligence. Decart's technology stack spans real-time video generation, world models, and chip optimization that lets AI workloads run across different accelerators.

AI Insights

Amazon's pursuit of Decart signals that the AI competition is shifting from chip performance to software-level efficiency, where compilers, optimization stacks and migration tools determine whether custom silicon like Trainium can challenge Nvidia's CUDA ecosystem. Anthropic's withdrawal after due diligence suggests that valuations rising faster than product maturity carry real risk, and buyers will scrutinize technical depth and integration costs more closely. The broader trend is that AWS is vertically integrating across chips, model platforms and data tools, while the industry's real bottleneck is increasingly energy and infrastructure rather than compute alone.

Analysis written by AI from the original article. Check it against the source.

Translated text

Amazon is reportedly negotiating to acquire Israeli AI startup Decart for a potential deal value of about $7 billion. Anthropic had also approached Decart at a valuation of about $6 billion but dropped its pursuit after due diligence. Nvidia and Nebius were also reported as potential buyers. Founded in 2023 by three engineers, Decart raised $300 million in May 2026 at a valuation near $4 billion, with investors including Radical Ventures and Nvidia. Decart is not just a video app but a technology stack spanning chips, models, and applications. It focuses on three areas: real-time video generation, world models that aim to understand and simulate real-world environments, actions, and changes, and chip optimization that lets the same AI task run efficiently on different hardware. Decart calls this technology the Decart Optimization Stack (DOS), designed to run AI workloads on Nvidia GPUs, Amazon Trainium, and Google TPUs. For Amazon, the appeal is clear: Trainium's chip performance alone cannot displace Nvidia, since customers depend on CUDA, developer tools, model adaptation, and established engineering habits. Decart can lower the migration cost for customers and help Amazon's in-house chips enter real training and inference workloads. Amazon is also expanding across the AI stack, including the Bedrock multi-model platform and the acquisition of the DuckLabs team behind DuckDB. Separately, Amazon submitted a preliminary plan for 36 data center buildings at a former coal plant site in Homer City, Pennsylvania, including a 4.5-gigawatt gas generation facility. Local residents oppose the project over water pollution, noise, greenhouse gas emissions, and electricity costs.

Related News

3