Amazon pushes cheaper AI with custom chips as demand remains strong

Amazon CEO Andy Jassy said that the artificial intelligence infrastructure remains too expensive for the general use of the company, but this is exactly what Amazon’s web services aim to change by developing its own tokens and reducing the costs of inference.

Speaking on CNBC Squawk boxJassy has rejected speculation that recent advances artificial intelligence The effectiveness of the model, like those introduced by Deepseek, would reduce the need for infrastructure construction.

“We have a very high request,” he said. “I don’t see us alleviating our construction centers right now.”

Jassy pointed out that WES knows no drop in AI infrastructure demand, despite macroeconomic uncertainty and the imminent threat of prices. He explained that if more effective models are welcome, the challenges of AI infrastructure are deeper and more persistent.

“If you build border models like us, you work on many of the same problems,” said Jassy. “The more we can make the cost of AI, the more customers will use it.”

The drop in costs unlocks the expense of greater customers

Jassy has established parallels between quarters of today’s AI and the first days of AWS. Costs below per unit of calculation, he said, often lead to an increase in customer spending – no less – as companies find new ways to innovate.

“This allows them to save money in what they are building, but they don’t spend less,” he said. “It triggers them to make more innovation.”

According to Jassy, ​​there are two main levers to reduce IA costs: the chips themselves and the cost of inference, the prediction process using formed models. Although training dominates spending today, inference will become the large -scale cost center.

AWS responded by building its own personalized AI chips, which, according to Jassy, ​​offer 30% to 40% of better price performance than current GPU instances. He also noted that reduction in inference costs implies progress in hardware and software techniques.

Jassy clearly supervised the mission: “If you sat in meetings with the AWS team at the moment, they have the impression that it is their responsibility and their mission to make the cost of AI significantly less than today.”

This thrust to reduce AI costs is a potential catalyst for certain segments of the cryptography sector. In particular, the developers have long explored the use of AI, but have often struck a wall due to the infrastructure and cost constraints. With a more affordable technology to come in the years, blockchain-The native IA applications, from the analysis of channels to decentralized autonomous agents, could become more feasible on a large scale.