Amazon Pushes to Sell AI Chips Externally, Challenging Nvidia's Dominance

Amazon is in negotiations to sell its self-developed artificial intelligence (AI) chips to external corporate data centers, Bloomberg reported on June 19.

The strategy aims to tap into the demand from companies looking to reduce their reliance on Nvidia.

Peter DeSantis, Amazon's senior vice president of utility computing, stated in an interview with Bloomberg that the company has begun discussions with potential customers who wish to purchase its Trainium chips directly, rather than through Amazon Web Services (AWS).

The names of the potential clients were not disclosed.

Amazon's AI accelerator, "Trainium," has been adopted by companies such as OpenAI, Anthropic, and Uber Technologies through AWS since its launch in 2020.

According to Amazon, the value of revenue commitments generated by these chips exceeded $225 billion (approximately 342 trillion won) as of April this year.

DeSantis noted that the third generation of Trainium, which began shipping earlier this year, is already "mostly sold out," and there is strong interest in the fourth generation, scheduled for release next year.

Amazon CEO Andy Jassy also hinted in a shareholder letter last April that there is significant potential for selling its chip racks to third parties.

Alphabet, the parent company of Google, also announced last April that it would directly supply its proprietary AI semiconductor, the Tensor Processing Unit (TPU), to select customers.

Amazon is reportedly pursuing these external sales, particularly to target the "sovereign cloud" demand in the European market, where there is a requirement for local control of computing resources.

Addressing concerns that external sales might cannibalize its AWS business, DeSantis dismissed the idea, stating, "There is so much under-consumption in the AI space."

DeSantis also offered a outlook on the commercialization of quantum computers.

In an interview with CNBC, he predicted that "the first commercially useful, small-scale quantum computers will emerge within the next 5 to 7 years," and that performance would improve rapidly each year, similar to Moore's Law in the semiconductor industry.

He explained that quantum computers are not merely faster than traditional computers, but have distinct advantages in solving problems that are difficult for existing computers to handle, such as in chemistry and materials science.

Amazon unveiled "Ocelot," a quantum computer chip specialized for error correction, last year.

Microsoft has set a goal to introduce a commercial quantum computer by 2029.

Following this news, Amazon's stock closed at $244.39 on June 18, up 2.90% ($6.89) from the previous trading day.

(Photo: Yonhap News)
※ Please note: This article was translated by AI and may contain errors.