Amazon Pushes to Sell In-House Developed AI Chips Externally... Challenging Nvidia's Dominance

Amazon is pushing to sell its in-house developed artificial intelligence (AI) chips—which it had previously only used internally for its cloud service (AWS)—to external companies, launching a full-scale challenge against Nvidia, which currently dominates the AI chip market.

Peter DeSantis, Amazon's head of AI, revealed in an interview with Bloomberg that Amazon has begun discussions with potential customers to supply its custom AI accelerator chip, 'Trainium,' to external data centers. Amazon CEO Andy Jassy also strongly hinted at the possibility of selling chips externally in a shareholder letter last April, stating that if the chip business were operated as an independent business, it could generate approximately $50 billion in annual revenue.

Currently, Amazon's custom silicon business (including Trainium, Graviton, and Nitro) has already surpassed an annual revenue run rate of $20 billion as of the first quarter of this year and is recording triple-digit growth. Major companies such as OpenAI, Anthropic, and Uber are already using Trainium through AWS, and as of last April, Trainium chips had achieved over $225 billion in revenue commitments. DeSantis added that the third-generation Trainium chips, which began shipping this year, are virtually sold out, and demand for the fourth-generation chips debuting next year is also very high.

One of the main reasons Amazon is considering the external sale of its chips is the demand for 'Sovereign AI' in overseas markets and outstanding cost efficiency. The company is highlighting to external customers that it can perform the same tasks cheaper than equivalent GPUs, and it is actively targeting the surging demand for infrastructure that can process and control data domestically to comply with local laws and security requirements, particularly in places like Europe. Google has also jumped into this competition by offering its custom chip, the Tensor Processing Unit (TPU), for some external customers to deploy in their own data centers.

In addition to Trainium, Amazon is also accelerating the spread of its general-purpose processor, the 'Graviton' chip. Meta has recently started adopting Graviton, and it has been driving growth within Amazon as well, being deployed more than any other processor over the past three years. On the day this news was reported, Amazon's stock price showed an upward trend of 2 to 3 percent, drawing a positive reaction from the market.

Experts analyze that while Amazon's move could put long-term pressure on Nvidia's immense pricing power, it is not a major threat in the short term. The global demand for AI computing is growing explosively to the point of far exceeding supply, meaning the two companies can coexist by dividing the market. In fact, while selling its own AI chips, Amazon remains one of Nvidia's largest customers, revealing plans to deploy an additional 1 million or more Nvidia GPUs starting in 2026.