.jpg?w=1024)
Amazon may be preparing to sell its AI accelerator chips to other companies, seeking to capitalize on the wave of spending behind AI data centers.
The Trainium accelerator has become a popular choice on Amazon Web Services for AI companies, including OpenAI, Anthropic, and Uber, as these chips are seen as more efficient for inference, or the day-to-day running of AI services.
“We view AI infrastructure as rapidly evolving,” said Amazon’s AI chief, Peter DeSantis, at the VivaTech startup event in Paris. “And we’re constantly looking at ways to get to more customers.”
The ecommerce giant has struggled to gain a foothold in the AI market, with its recent launch of Alexa+ failing to gain much traction. At the same time, it is one of the main beneficiaries of the AI buildout as the market leader in cloud services and may see greater value in the infrastructure and hardware layers. It has also backed the primary AI developers, ensuring that they use Amazon's infrastructure.
Another competitor to Nvidia's dominance
In the company's annual shareholder letter, CEO Andy Jassy said Trainium processors were at a $20 billion revenue run rate, and that this could reach $50 billion if sold on the open market.
That would make it a strong competitor to Nvidia, but still far behind it in total revenue. Nvidia posted $81.6 billion in revenue in the first quarter of this year alone, and has reportedly increased its market share in the inference AI market over the past year despite growing competition.
Alongside the Trainium chip, Amazon is reportedly also seeing positive feedback on its Graviton processors, which have been integrated into many of its computing systems. It has not said if it would sell these separately as well.
Rivals circling AI's hardware layer
Amazon is not the only hyperscaler looking to sell its chipsets to other data centers. Google also recently announced it would sell its Tensor Processing Unit, another AI accelerator, to a select group of partners for their own use. Both Google and Amazon have been working on specialized hardware for AI workloads for years.
At the heart of both moves is a major revenue opportunity from businesses that either want to build their own data centers or smaller data center operators that want to supply their own chips. While Amazon and Google would both ultimately like these businesses to use their own data centers, many use cases require other arrangements.
Hyperscalers are among Nvidia's largest customers, buying large numbers of the company's GPUs as demand remains off the charts. At the same time, these service providers are keen not to be locked into one company, and are all actively looking to back or launch competitors to reduce Nvidia's market dominance.
Also read: Amazon’s Alexa for Shopping adds price tracking and product comparisons as the company refines its AI tools for online retail.