In a market where NVIDIA virtually dominates the AI acceleration chip market, Amazon has clearly carved out a niche for itself. Amazon CEO Andy JassydisplayAWS's self-developed AI computing chip, Trainium, has already reached a revenue scale of several billion US dollars, indicating that its strategy to reduce the cost of AI computing has gained market support.
Trainium2 has been adopted by 10 enterprises, with cost-effectiveness being the key factor.
Andy Jassy pointed out that the Trainium2 chip business is currently experiencing strong growth momentum. According to official AWS data, over 100 million Trainium2 chips have been produced, and more than 10 enterprises are using the chip to provide computing power through the Amazon Bedrock service platform.
Andy Jassy stated frankly that the key to Trainium's ability to stand out from the massive cloud customer base lies in its superior price-performance ratio compared to other GPU options. This means that Amazon is attracting cost-sensitive enterprise customers by offering its own computing solutions that are cheaper than NVIDIA's but perform similarly or even better under specific workloads.
Anthropic provided 50 computing units, with "Project Rainier" being a key driving force.
In an interview, AWS CEO Matt Garman confirmed that Anthropic, a key strategic partner of Amazon, made significant contributions.
Matt Garman revealed that the collaboration between the two parties"Project Rainier"In its efforts, Anthropic utilizes over 50 Trainium 2 chips for training and building the next generation of Claude models. This explains why Amazon was willing to make a huge investment in Anthropic in exchange for the latter's commitment to making AWS its primary model training partner.
Trainium3 boasts 4 times the performance; the next generation of products will "join the fray if you can't beat them."
At the AWS re:Invent conference, Amazon also officially showcased its...Trainium3 chipIt is claimed to be 4 times faster and consume less power than its predecessor.
However, facing the powerful moat built by NVIDIA with its CUDA software ecosystem, Amazon seems to have prepared a more flexible strategy. In its upcoming Trainium 4 chip design, Amazon will support interoperability with NVIDIA GPUs in the same system. Through NVIDIA's NVLink Fusion technology, Amazon will break down the clear boundaries between its self-developed chips and the GPU camp, providing customers with more flexible hybrid architecture options.
This suggests that Amazon may no longer adopt a head-on "replacement" strategy in the future, but instead provide a hybrid architecture that allows customers to simultaneously leverage the versatility of NVIDIA and the cost advantages of Trainium chips on the AWS cloud platform.
Judging from its current revenue scale, Amazon's self-developed chip business has clearly taken a more stable step forward.
