Amazon hurriedly launches the latest AI chip, Trainium3, directly challenging Nvidia and Google

Amazon Web Services (AWS) has recently accelerated the deployment of its next-generation AI accelerator chip, Trainium3, to select data centers, officially opening it for customer use starting from 12/2. This chip is seen as Amazon’s key weapon to catch up with NVIDIA (NVIDIA) and Google in the AI computing space, with the company highlighting its high performance and cost-effectiveness. However, despite being a leader in cloud services, AWS is still less attractive to AI model developers compared to Microsoft and Google. The urgent priority is to deliver in-house hardware that can truly secure enterprise customers and strengthen ecosystem stickiness.

Trainium3 Launches, AWS Accelerates AI Chip Deployment

AWS Vice President Dave Brown stated that the latest generation AI accelerator chip, Trainium3, has completed deployment in some data centers and is now officially in use, available for enterprise customers immediately. This chip was launched only a year after the previous generation, which is quite rare for the industry. Brown also added that AWS will begin ramping up its distribution at a very fast pace starting early next year.

Amazon emphasizes that the Trainium series delivers high performance and cost-effectiveness, enabling large-scale AI training at lower costs. AWS engineers joked that when they first powered on Trainium3 in August, all they hoped for was that it wouldn’t smoke or catch fire, underscoring the tight development and launch timeline of this chip.

Challenging NVIDIA Is Not Easy: Software Ecosystem Remains the Biggest Weakness

Although AWS continues to stress the price-performance advantage of Trainium, NVIDIA still enjoys a huge advantage in the AI developer ecosystem. An automation construction company said that while their infrastructure is on AWS, they still mainly use NVIDIA GPUs when training AI models for autonomous excavators, citing the need for a powerful and user-friendly solution—which is NVIDIA.

Industry observers generally believe that due to the lack of sufficient software toolchains, Trainium still presents a high adoption barrier for many development teams, posing challenges for AWS in capturing the high-end AI training market.

Anthropic Becomes the Largest User, AWS Hopes to Drive Demand Through It

Currently, the largest number of Trainium users in the market comes from AI startup Anthropic. AWS announced earlier this year that it had deployed more than 500,000 Trainium chips for Anthropic, aiming to increase that number to 1 million by the end of the year.

AWS CEO Matt Garman said the relationship between the two companies is very good, and Anthropic’s demand for model training is enormous, so they purchase computing power from AWS, Google, and other cloud providers. Anthropic previously established a partnership with Google, allowing it to use Google TPUs worth tens of billions of dollars. Observers point out that aside from Anthropic, AWS currently has few publicly disclosed major Trainium customers, making it difficult for the market to fully assess the chip’s real-world performance.

AWS Launches Nova 2, Forge at Conference, Stresses Real-World Performance Is Key

At the annual re:Invent conference, Amazon simultaneously launched an updated language model series, Nova 2. The multimodal version, Nova Omni, can handle text, images, audio, and video inputs, with outputs in text or image formats.

Rohit Prasad, in charge of AGI development, stated that although Nova has not ranked at the top in standard benchmarks in the past, the truly important metric is real-world usage, and he believes the new version is competitive enough.

AWS also launched an updated tool, Nova Forge, enabling enterprises to customize their own data even before model training is complete. Reddit has already used this tool to build a model that determines whether posts violate platform rules. CTO Chris Slowe commented that while many companies want to use the most powerful model directly, the real value is in “making the model your domain expert.”

(OpenAI and AWS Reach a $38 Billion Cloud Partnership to Break Away from Microsoft and Pave the Way for IPO)

This article, Amazon Launches Latest AI Chip Trainium3, Directly Challenges NVIDIA and Google, first appeared on Chain News ABMedia.