IT Home On October 24, Amazon CEO Andy Jassy said that the cost of AI inference is still too high. For AI to be widely used in enterprise software, the cost needs to drop to about one-tenth of current levels, and Amazon’s self-developed chips are expected to help reduce costs.
Jasi revealed in an interview released by venture capital firm a16z on the 23rd local time that “in my opinion, for AI to be used as widely as possible, the inference cost is still one order of magnitude higher.”
Amazon plans to invest $220 billion this year (IT Home note: at current exchange rates, this is approximately 1.47 trillion yuan) in capital expenditure, yet it still cannot build enough infrastructure to meet AI demands. Jasi revealed that the main computing power supply quota for 2026 is basically filled, most of 2027 has been booked by customers, and a significant portion of 2028 has also been sold out. In his view, with such strong market demand, even if Amazon invests more funds, it would not be a waste.
Jia Xi compared AI requirements to a “balancer”: one end is the AI laboratory, where the computing power consumed is “unprecedented”; the other end is enterprise customers, who mainly use AI for specific tasks such as customer service and fraud detection.
What is still not fully developed is the intermediate market, namely the various software that businesses use daily. These software applications have not been widely integrated with AI yet, but Jasi believes that the biggest AI demand in the future may come from this area.
The main factor restricting enterprises from adopting AI is the inference cost. AI models consume computing power every time they answer a question or complete a task. If the benefits of introducing AI are not sufficient to offset the higher operating costs, enterprises have little incentive to reengineer their existing software.
Jasi compares Amazon’s experience from many years ago. At that time, Intel dominated the cloud server processor market. Amazon wanted to achieve higher computing performance at the same cost, but struggled to convince Intel to prioritize this need. As a result, they decided to develop their own Graviton processors. In his view, “the same story is happening again in the AI field.”
The AI chip market is currently dominated by NVIDIA. Amazon, on the other hand, has launched its own Trainium chip, hoping to offer a lower-cost option for customers. Jasi did not directly mention NVIDIA in the interview; instead, he emphasized that Amazon remains an important partner of the dominant supplier in the market.
According to the introduction, Trainium’s cost-performance ratio is about 30% to 40% higher than that of similar processors. Customers can get more computing power for the same amount of money. “Our chip business plays an important role in reducing inference costs.”
Amazon previously revealed that Trainium has received future revenue commitments of over $225 billion (about 1.5 trillion yuan at current exchange rates), including agreements with Anthropic and OpenAI.