Amazon Nears $7 Billion Decart Deal to Fix Why AI Developers Pick Nvidia Over Its Trainium
Amazon is negotiating to acquire Israeli AI startup Decart for approximately $7 billion. Decart's software, the Decart Optimization Stack, optimizes AI model performance on various hardware, including Amazon's Trainium chips. The deal aims to boost Amazon's AI chip competitiveness against Nvidia. Amazon confirmed running Decart's Lucy2 model on Trainium3 achieved 80% Model FLOPS Utilization.
How this was made
The 30-second read
Why it matters
A $7B acquisition would be Amazon’s largest pure AI software deal, aiming to make Trainium economically and operationally easier to use by automating hardware-specific inference optimization.
Market read
Traders may reprice AWS AI infrastructure competitiveness if the acquisition is credible and progresses quickly.
What to watch
Deal completion risk, potential customer migration inertia, and whether DOS performance holds up in third-party benchmarks could materially change the realized benefit.
Background
Amazon has invested heavily in Trainium as custom AI silicon, but adoption has been constrained by developers’ CUDA familiarity and the engineering effort to port inference workloads efficiently.
Ticker impact
WSJ and SiliconAngle report Amazon is in talks to acquire Decart for about $7B to improve Trainium inference optimization versus Nvidia’s CUDA ecosystem.
Likely upward bias as traders price in improved AWS AI economics and a credible software unlock for Trainium competitiveness.
The article frames the acquisition as a direct solution to the CUDA lock-in problem and cites specific Trainium performance claims tied to Decart’s optimization stack.
Market effects
Raises the probability that cloud AI chip competition will hinge more on inference optimization software layers than raw hardware specs.
Limited direct regional impact; deal is Israel-based but the market read-through is global cloud/AI infrastructure.
Could intensify competitive pressure on Nvidia’s CUDA moat and on other accelerator ecosystems if Amazon’s software layer reduces switching costs.
Counterpoint
Even with strong claims, integration and validation of DOS across AWS customer workloads may take longer than the market expects, limiting near-term Trainium adoption impact.
Key entities
- companyAmazon
US cloud and AI infrastructure provider negotiating to acquire Decart to improve Trainium inference optimization.
- companyDecart
Israeli AI startup offering the Decart Optimization Stack (DOS) for faster inference across Nvidia GPUs, Google TPUs, and Amazon Trainium.
- companyNvidia
Dominant AI hardware and CUDA software ecosystem that the article says developers default to.
- productAWS Trainium
Amazon’s custom AI chips whose adoption is described as hindered by CUDA ecosystem lock-in.
