Amazon is exploring the possibility of selling its custom AI chips to third-party companies, marking a significant expansion beyond its traditional cloud services business. CEO Andy Jassy revealed in an annual shareholder letter that the company’s in-house silicon unit, responsible for chips like Trainium used in Amazon Web Services (AWS), could generate annualized revenue exceeding $20 billion. Jassy indicated strong demand for these chips, suggesting that Amazon could potentially sell entire racks of its AI hardware to external customers in the future.
This development positions Amazon as a more direct competitor to established chipmakers like Nvidia, as the company seeks to leverage its growing semiconductor capabilities to capture a larger share of the AI hardware market. The move aligns with Amazon’s broader strategy of heavy investment in AI infrastructure, including a planned $200 billion capital expenditure budget for 2026, much of which will be spent on expanding AWS data centers and associated technologies.
Amazon’s decision to potentially offer its AI chips commercially underscores the escalating competition in the AI chip sector and highlights the strategic importance of custom silicon in powering large-scale AI workloads. By commercializing its chips, Amazon could diversify revenue streams and solidify its role not just as a leading cloud provider but also as a key player in AI hardware innovation. This strategy also echoes trends of other AI firms, like Anthropic, investigating custom silicon options to better control their AI compute needs.
