The competition in the artificial intelligence computing power market is entering a new stage of confrontation. Recently, Peter DeSantis, Amazon's AI chief, revealed during an interview in Paris that the company is deep in discussions about directly selling its self-developed Trainium series AI ASIC physical chips to external enterprises. This means that the computing power resources previously only provided to customers through cloud services may soon enter more companies' self-built data centers in the form of hardware products.
Trainium, Amazon's self-developed AI-specific chip, has been provided to external developers through its cloud computing platform for some time. With the explosive growth of global AI computing power demand, the demand for accessing these chips in the cloud has nearly exploded. In fact, according to information shared by Amazon's CEO Andy Jassy in April this year, the supply of Trainium chips in the cloud has already been in short supply. The Trainium3 version is almost sold out, and the new generation of Trainium4 chips has also attracted widespread attention in the industry.
Regarding this business adjustment, the outside world cannot help but speculate whether it will divert Amazon's existing AI cloud business. DeSantis gave a relatively optimistic judgment on this. He believes that the current global AI computing power demand space is extremely broad, and bringing physical chips to the market will not erode the existing cloud-based business. By selling physical chips, Amazon can reach a more diverse customer base, further solidifying its influence in the AI hardware ecosystem.
From being an initial cloud service provider to now exploring a hardware direct sales model, Amazon's strategic move reflects the pursuit of greater control over the computing power supply chain by tech giants. By transforming the Trainium chip from "rental" in the cloud to "spot" in the market, Amazon not only responds to the growing demand for computing power hardware but also tries to build a more solid market advantage by combining hardware and services in the increasingly fierce competition for AI infrastructure.
