Amazon is reportedly in talks to acquire Decart, an AI startup, for roughly $7 billion, according to a Wall Street Journal report published Oct. 11. The San Francisco-based company develops software that enables developers to shift AI workloads between different chip types. The potential transaction would give the Seattle tech giant access to tools designed to help businesses optimize computing across various processor architectures.
Decart, founded in 2023 by Israeli engineers Dean Leitersdorf, Orian Leitersdorf, and Moshe Shalev, secured $300 million in May at a valuation approaching $4 billion, with investments from Sequoia Capital, Benchmark, Radical Ventures, and Nvidia. The startup's Optimization Stack works with Nvidia GPUs, Google TPUs, and Amazon Trainium chips, supporting customers in refining AI training and inference operations across different hardware platforms. Beyond chip optimization, Decart builds Oasis, an interactive world model aimed at physical AI uses like robotics, and Lucy, which alters live video for applications including gaming, ads, and virtual try-on. The two companies already collaborate through an AWS partnership established in May that includes a commercial agreement and joint market initiative, and Decart is a member of the Amazon Partner Network. The proposed purchase follows Amazon's April agreement to buy satellite operator Globalstar, a separate deal expected to close in 2027.
The $7 billion price tag carries particular weight given that AI lab Anthropic, which counts Amazon as a primary investor and cloud supporter, abandoned its own $6 billion pursuit of Decart last month after completing due diligence, according to TechRepublic. The reported discussions don't establish whether AWS would roll out new migration services or adjust customer pricing. For Decart customers and partners, a separate concern centers on whether an Amazon acquisition would alter support for competing hardware or access to optimization capabilities.
The report explains that for AWS partners and managed service providers counseling customers on AI infrastructure, Decart's technology could widen the range of tools for optimizing workloads, though the business case hinges on model type, performance needs, migration expenses, and available computing capacity. Partners evaluating alternative chips should weigh training or inference costs together with engineering time, software licensing, data transfer, and ongoing support, since a lower compute rate alone wouldn't prove a migration delivers savings. Support for multiple chip architectures doesn't guarantee every model can transfer without changes, so a pilot should test model quality, latency, throughput, and integration with existing customer tools before partners recommend moving to production. More processor options could help customers reduce reliance on a single supplier, but chip portability doesn't ensure automatic switching between cloud providers or guarantee spare capacity during demand spikes.
The report recommends that partners watch for confirmed changes to Decart's support for Nvidia and Google hardware, commercial terms, and AWS integrations if an acquisition goes forward, since those details would clarify whether the technology broadens customer choice or introduces new dependencies. For service providers, workload assessment, performance testing, and ongoing cost management represent potential areas to investigate. The immediate guidance is to assess existing capabilities and ground customer proposals in measured results while monitoring for a confirmed deal and any subsequent product or partner announcements. For businesses navigating cloud infrastructure decisions, the acquisition signals a shift toward greater flexibility in chip selection, though the real test will be whether cost savings justify the engineering effort required to move between platforms and whether any single vendor's ecosystem ultimately constrains rather than expands strategic options.