
The talks would extend Amazon’s push against Nvidia by offering its in-house AI accelerators beyond AWS, after Trainium’s 2020 launch and more than $22.5 billion in reported revenue commitments.
Amazon is in talks to sell its in-house Trainium artificial intelligence chips to data centers operated by other companies, expanding a strategy that has so far been centered on Amazon Web Services and broadening its challenge to Nvidia in AI semiconductors. Trainium, launched in 2020, is an AI accelerator used to train and run AI models. Amazon said in April that it had secured more than $22.5 billion in revenue commitments for the chip. AWS customers cited in the source include OpenAI, Anthropic, and Uber, underscoring demand as Amazon considers turning an internal cloud component into a broader infrastructure product.