On October 28, at a conference in Las Vegas, Amazon announced the birth of Trainium2, a new generation of artificial intelligence chip for its cloud computing service. The main function of this chip is to train AI systems. Compared with the previous generation product, the new generation of Trainium2 is 4 times faster and twice as energy efficient.
The move is the latest move by Amazon Web Services (AWS) in its battle with Microsoft for dominance in the AI market. Because not long ago, Microsoft had just released its AI chip Maia.
Of course, if Amazon's move is just to compete with Microsoft, then our situation will be a bit small. As a top company in the world, I think this is more of a redemption or breakthrough that Amazon must complete for itself, because only in this way can it compete with some of the top companies that have been working in the field of AI and cloud computing for many years. For example, Google has been providing Tensor Processing Unit (TPU) to its cloud computing customers since 2018. After several years of accumulation, the data behind this is huge.
Although it is full of difficulties and surrounded by powerful enemies, there is also good news for Amazon. At the conference in Las Vegas that day, in addition to announcing Trainium 2, AWS also released its fourth-generation custom central processing unit chip Graviton 4, which has a performance improvement of 30% compared to the previous generation.
With the gradual maturity of large-scale language model technologies such as chatgpt, cloud computing companies have become more competitive, and customized chips that can provide sufficient computing power are becoming more and more valuable. AWS plans to start offering these new training chips next year. This reflects the rapid progress of AI technology and the fierce competition within the cloud computing field.
Finally, a little easter egg. AWS and Microsoft both use Arm Ltd's technology in their chips, part of an ongoing trend of chips made by Intel and Advanced Micro Devices in cloud computing. (This shows that no top company will cede such a large market to others.) Oracle is using chips from startup Ampere Computing for its cloud services.
