Amazon Plans to Sell 144-Chip Trainium Systems as Nvidia Faces New Data Center Challenge
Updated
Updated · The Motley Fool · Jul 6
Amazon Plans to Sell 144-Chip Trainium Systems as Nvidia Faces New Data Center Challenge
3 articles · Updated · The Motley Fool · Jul 6
Summary
Amazon is preparing to sell its Trainium AI chips to external customers, extending a strategy it began inside AWS four years ago to cut reliance on Nvidia.
Trainium3 alone does not match Nvidia's top Blackwell GPUs, but Amazon says 144-chip UltraServers can deliver comparable rack-scale performance at lower cost.
That pitch could appeal to companies and regions such as Europe that want AI infrastructure without depending on Nvidia or storing data with U.S. hyperscalers.
Nvidia still looks protected in the near term because CUDA locks in developers, most AI software is optimized for its GPUs, and demand for its data center chips still exceeds supply.
The bigger risk is longer term: Amazon is joining Microsoft and Google in turning from major Nvidia customers into direct AI chip competitors.
With tech giants now selling their own chips, is Nvidia's near-monopoly on AI hardware facing a serious, long-term threat?
As the US and China build separate AI ecosystems, how will this tech divide reshape the global semiconductor supply chain?
Cloud Giants vs. Nvidia: The Race for AI Chip Supremacy in a $50 Billion Market
Overview
The AI compute market is rapidly changing as Amazon, Microsoft, and Google challenge Nvidia’s dominance by developing and, in some cases, selling their own custom AI accelerator chips. These cloud giants want more control over pricing, supply, and innovation, aiming to deliver cost-effective and high-performance solutions for the growing AI sector. With the XPU market reaching $31 billion in 2025 and custom silicon design expected to double by 2028, their efforts signal a major shift. This transformation is driven by the need to optimize costs and performance, reshaping the future of AI infrastructure.