Amazon is reportedly in talks to acquire artificial intelligence startup Decart AI for around $7 billion, according to The Wall Street Journal.
The potential price remains uncertain. CTech subsequently reported a possible $5.5 billion valuation. For AWS customers, the significance lies in whether Decart’s software could improve AI workload efficiency and make it easier to move models between different chips.
Decart AI develops software to make AI models more efficient to train and run, alongside real-time generative AI models. Its Lucy tools transform live video, while Oasis generates interactive environments that respond to user inputs.
Anthropic was reportedly considering a roughly $6 billion acquisition of Decart in August. In September, the company reportedly ended those talks after conducting due diligence. The purchase would have represented a major expansion of Anthropic’s acquisition activity.
The Journal reported the acquisition talks, citing people familiar with the matter, while CTech’s subsequent report put the potential valuation below $7 billion.
Decart’s chip optimization software could help Amazon improve the efficiency of workloads running on its Trainium chips and simplify moving models between different hardware platforms. These are potential benefits, rather than confirmed reasons for Amazon’s interest.
For IT teams, easier movement between chips could expand their hardware options, but the value would depend on application compatibility, migration effort, and performance on their own workloads. Lower operating costs for Amazon would not automatically translate into lower AWS prices.
Major AI acquisition agreements announced this year include SpaceX’s $60 billion agreement to buy Cursor and Nvidia’s reported $12.9 billion agreement to acquire Hugging Face. Stripe also announced an agreement to acquire OpenRouter, with Axios reporting a price of more than $8 billion.
Amazon looking to strengthen its AI data center capabilities
AWS held 28% of worldwide cloud infrastructure services revenue in the second quarter of 2026, compared with Microsoft’s 20% and Google’s 15%, according to Synergy Research Group. These figures cover the broader cloud infrastructure market rather than AI workloads alone.
At the same time, AI-specific workloads require considerably more computing power than many traditional cloud applications, forcing Amazon to rapidly expand its data center infrastructure to meet demand. Amazon is reportedly discussing an approximately $8 billion sale-leaseback arrangement involving Nvidia chips, allowing it to raise financing while retaining access to the hardware. On its July earnings call, Amazon forecast approximately $220 billion in companywide cash capital expenditure for 2026, up from its earlier estimate of about $200 billion.
Amazon’s proprietary-model strategy is also evolving. The company is reportedly considering consolidating parts of its Nova lineup around a single frontier model. Amazon has said it continues to support existing Nova customers while investing in next-generation frontier research.
For AWS buyers, the Decart talks are a development to monitor rather than a reason to change deployment plans. Any future offering would need to demonstrate workload performance, compatibility with existing tools, and savings after migration and support costs.
Teams evaluating Trainium should also ask whether their models can move back to GPUs without extensive code changes, and whether existing monitoring and deployment tools would still work. Easier access to alternative chips is useful only if the operational effort does not outweigh the savings, making portability and ongoing support important parts of any cost comparison.
The next questions are whether Amazon reaches an agreement and how Decart’s technology would become available to customers. Those details would determine whether the acquisition brings practical benefits to enterprise AI deployments.
Read more: Amazon’s reported plans to consolidate its Nova AI models highlight another potential shift for businesses building AI applications on AWS.