Amazon's $7 billion Decart talks echo a deal that changed its cloud business
Amazon's 2015 acquisition of Israeli chipmaker Annapurna Labs helped lay the foundations for its in-house silicon business. Now, the tech giant is reportedly pursuing Decart, an Israeli AI startup whose technology can help AI models run more efficiently on chips, including Amazon's own Trainium processors.
Amazon’s reported pursuit of Israeli AI startup Decart for approximately $7 billion would mark another major bet on Israeli technology, and potentially a new chapter in a strategy that began with a $350 million acquisition more than a decade ago.
In January 2015, Amazon acquired Annapurna Labs, an Israeli chip designer founded in Yokneam in 2011. The deal helped establish the foundation for Amazon Web Services’ in-house silicon business, giving the cloud giant greater control over the hardware powering its data centers. Today, Amazon’s custom chips are central to its efforts to compete with Nvidia and other technology companies in the rapidly expanding market for AI computing.
Decart offers a different approach to the same strategic challenge. Rather than designing processors, the Israeli startup develops technology that helps AI models run more efficiently on existing hardware, potentially extracting significantly more performance from expensive chips. Its software is designed to work across different chip architectures, including those developed by Nvidia, Google and Amazon. Industry estimates suggest that Decart’s technology can allow AI models to run up to eight times faster, although actual gains depend on the workload and hardware involved.
Founded in 2023 by Dean and Orian Leitersdorf and Moshe Shalev, Decart initially gained attention for its real-time video-generation technology. Its Oasis model allowed users to interact with an AI-generated virtual environment in real time, while subsequent products expanded the company’s ambitions in video generation and interactive AI. Alongside those products, Decart developed a GPU optimization business, licensing its technology to cloud providers and AI laboratories. That business generated revenue from an early stage and helped fund the substantial computing costs associated with developing its own AI models.
Amazon is already a customer, giving the potential acquisition a more concrete rationale than a purely speculative technology bet. In December 2025, CTech reported that Decart’s real-time video model ran four times faster on Amazon’s Trainium chip, following the startup’s testing of the hardware. Amazon subsequently became a strategic customer, and AWS gave Decart early access to its Trainium3 chip. The relationship gives Amazon firsthand experience with Decart’s technology and its potential to improve the economics of AI computing.
The connection echoes the logic behind Amazon’s acquisition of Annapurna Labs, although the technologies are different. Annapurna helped Amazon build its own processors, including the Graviton family for general-purpose cloud computing and Trainium and Inferentia chips designed for AI training and inference. Its Nitro technology also plays a central role in the security and management of AWS infrastructure. These products have helped Amazon reduce its dependence on third-party chip suppliers and tailor its infrastructure to its own workloads.
Decart could extend that strategy in a different direction. Even as Amazon invests heavily in developing its own silicon, software that improves the performance of chips across manufacturers could help it get more computing capacity from its infrastructure. That is increasingly important as technology companies face enormous costs associated with training and running AI models. Improving the efficiency of existing hardware can reduce the cost of delivering AI services without requiring a proportional increase in the number of chips deployed.
Amazon’s 2015 acquisition of Annapurna Labs was widely viewed as a relatively modest investment at the time. The chip business it helped establish has since become an important part of AWS’s competitive position. Estimates of the value of Amazon’s in-house silicon operations vary, and a standalone valuation above $50 billion would depend on assumptions about their revenue, profitability and future growth. Nevertheless, the strategic significance of the business is clear: Amazon has used its own chip development capabilities to differentiate its cloud services and challenge established hardware suppliers.
Decart’s trajectory has been similarly rapid. In May 2026, the startup raised $300 million at a $4 billion valuation in a round backed by Nvidia, with Amazon joining as a strategic customer.
The reported Amazon talks follow the collapse of a proposed acquisition by Anthropic, which withdrew after due diligence. Nvidia had also been involved in earlier acquisition discussions. Amazon could now emerge as a potential buyer of a company whose technology it already knows and uses, although the reported talks do not guarantee that a deal will be completed.
If the acquisition goes ahead, it would illustrate how the competition for AI dominance is expanding beyond the race to build more powerful models or manufacture faster chips. Companies are increasingly seeking control over the entire computing stack, from processors and infrastructure to the software that determines how efficiently those resources are used. For Amazon, Decart could offer another way to strengthen AWS as it competes with Nvidia, Google and other major players for a larger share of the global AI market.
