MUMBAI: Amazon may soon be putting more than packages on the move. Its AI chips could be heading out the door too. Amazon.com is exploring the possibility of selling its homegrown artificial intelligence chips directly to companies for deployment in their own data centres, signalling a significant expansion of its semiconductor ambitions beyond the confines of Amazon Web Services (AWS).
According to a Bloomberg report, the tech giant has begun discussions with potential customers about using its custom-built Trainium processors outside the AWS ecosystem as demand for AI computing infrastructure continues to soar.
Peter DeSantis, Amazon’s Senior Vice President overseeing AI and semiconductor initiatives, confirmed that conversations are underway, though he did not disclose the names of prospective customers. The move reflects Amazon’s efforts to broaden the reach of its AI hardware as enterprises increasingly seek alternatives to traditional chip suppliers.
At the centre of the strategy is Trainium, Amazon’s family of AI accelerators launched in 2020 to train and run large language models and other AI workloads. While the chips were initially designed to strengthen AWS’s cloud offering, they have already attracted heavyweight users including OpenAI, Anthropic and Uber Technologies.
The stakes are far from small. Amazon said earlier this year that the Trainium platform has generated more than $225 billion in revenue commitments, underscoring the growing appetite for specialised AI infrastructure.
Demand for the current generation of Trainium chips has reportedly been strong enough to create supply constraints, while Amazon is already developing a fourth-generation version that is generating early customer interest.
The latest development builds on comments made by Amazon Chief Executive Officer Andy Jassy in his April shareholder letter, where he suggested it was “quite possible” that the company would eventually offer entire racks of its AI chips to third parties.
If realised, the strategy would mark a notable shift in Amazon’s role within the AI ecosystem. Rather than simply providing computing power through AWS, the company could become a direct supplier of AI hardware to enterprises running their own data centres.
The move comes as technology giants race to loosen Nvidia’s grip on the booming AI chip market. Nvidia’s graphics processing units remain the dominant choice for training and deploying advanced AI models, but escalating demand and supply constraints have encouraged competitors to develop alternative solutions.
Amazon is not alone in that pursuit. Alphabet has also begun offering its in-house tensor processing units (TPUs) to a select group of customers for use in their own facilities, expanding competition in a market increasingly defined by the search for AI-ready infrastructure.
The broader AI boom is reshaping the economics of cloud computing and semiconductor development. As enterprises invest heavily in generative AI, demand has surged not only for cloud services but also for specialised processors capable of handling increasingly complex workloads.
At the same time, the rise of sovereign AI initiatives particularly in Europe has created fresh demand for infrastructure that allows organisations and governments to keep computing and data processing within local jurisdictions.
Against that backdrop, Amazon’s chip ambitions appear to be growing well beyond the cloud. What began as an effort to power its own AI services is steadily evolving into a broader play for a larger share of the global AI infrastructure market.
As the battle for AI supremacy intensifies, Amazon seems increasingly keen to ensure that Trainium is not just powering its cloud, but potentially powering everyone else’s as well.
