Tech giants' AI wrestling is in full swing. According to Bloomberg, Amazon is quietly advancing an extremely ambitious plan aimed at challenging Nvidia's current monopoly position in the field of AI chips.

According to reports, Amazon is currently stepping up the development of a new AI chip: Trainium 2, at its engineering lab in Austin, Texas. Compared to the previous generation, it has four times higher performance, three times more memory capacity, and significant energy efficiency and cost advantages.

Through these optimizations, Amazon hopes to reduce the procurement cost of AI chips and improve overall efficiency in data processing.

However, to truly challenge Nvidia's leadership in the AI hardware market, Amazon still faces huge challenges.

Trainium 2 performance has been greatly improved, and testing and delivery is scheduled to be completed before the end of this year

Currently, Amazon's core chip design engineer Rami Sinno is leading the team to accelerate the development of the second-generation self-developed AI chip, Trainium2.

In an interview with Bloomberg, Sino said their goal is to deploy these chips in data centers as soon as possible, and they plan to complete testing and delivery by the end of this year.

This chip is Amazon's third-generation product in the field of AI hardware and aims to provide a more efficient and cost competitive solution for machine learning model training.

Amazon's chip business is headed by James Hamilton, a former pioneer in the field of cloud computing.

Hamilton's team proposed the idea of developing their own chips as early as 2013. Amazon's first AI chip, Inferentia, was launched in 2019 and focuses on inference tasks, while the Trainium series mainly targets the need to train machine learning models.

Currently, Amazon's Trainium 2 chips have begun to be deployed in data centers, and it is expected that they will soon be fully promoted in multiple core data centers, including Ohio. Amazon's goal is to string Trainium2 into clusters of up to 100,000 chips.

Amazon said that compared to the previous generation, Trainium2 is four times better in performance, three times more memory capacity, and has significant advantages in terms of energy efficiency and cost.

Through these optimizations, Amazon hopes to reduce the procurement cost of AI chips and improve overall efficiency in data processing. Analysts believe that if Amazon's Trainium2 can take on more AI work within the company, as well as occasional projects from major AWS customers, then it is likely to be viewed as a success.

Amazon's AI hardware journey has a long way to go

Analysts believe that if it wants to truly challenge Nvidia's leading position in the AI hardware market, Amazon still faces huge challenges.

First, designing reliable AI chips is an extremely complex task, especially when balancing performance, energy efficiency, and cost.

Second, software tool support is also critical. Although Amazon's Trainium series chips have made some progress at the hardware level, they are still insufficient compared to Nvidia's mature software tools (such as CUDA). Analysts believe that currently, the Neural SDK software tools provided by Amazon are still in their early stages and cannot match Nvidia's solutions.

To overcome this technological divide, Amazon is actively cooperating with large customers and partners to promote the application of its AI chips. Well-known companies, such as data analysis company Databricks and AI startup Anthropic, have begun testing Amazon's Trainium chips and have achieved initial results in some projects.

Tom Brown, Anthropic's chief computing officer, said:

“We were impressed by the price-performance ratio of Amazon's Trainium chips. We've been steadily expanding its application across an increasingly broad range of workloads.”