The AI "arms race" accelerates again! Amazon launches a new chip array and large language model

According to Zhitong Finance APP, Amazon (AMZN.US) is expanding its artificial intelligence product lineup, launching a powerful new chip array and large language model, claiming it can compete with major rivals.

The Seattle-based company is assembling hundreds of thousands of Trainium2 semiconductors into clusters, making it easier for its partner Anthropic to train large language models required for generative AI and other machine learning tasks. Amazon stated that the new array will increase the startup's current processing capacity fivefold.

At its annual re:Invent conference, Amazon announced that its cloud services division, AWS, began offering the latest chips to customers on Tuesday.

Additionally, Amazon CEO Andy Jassy introduced a new AI model called Nova, capable of generating text, images, and videos, representing Amazon's latest effort to compete with OpenAI and other large language model builders powering chatbots and other generative AI tools.

AWS is the largest seller of rented computing power, operating many servers that other companies rent to train AI applications. AWS also offers models built by other companies, including Anthropic's Claude and Meta (META.US)'s Llama. However, the company has yet to produce a large language model widely regarded as competitive with OpenAI's state-of-the-art GPT models.

Amazon has released previous generations of a product called Titan over the past two years, but its capabilities were limited. Some parts of the Nova model are now available, with another part set to be released next year, including a "multi-modal to multi-modal" version that can accept text, speech, images, and video as input and generate responses in each mode.

Jassy stated that Amazon will continue to develop its own models while providing models made by other companies. He said, "We will provide you with the broadest and best features you can find anywhere."

Last month, Amazon announced it would invest an additional $4 billion in Anthropic. As part of the deal, Anthropic stated it would use Amazon's cloud and chips to develop its advanced models.

Gadi Hutt from Amazon's Annapurna Labs chip manufacturing division, who works with customers, stated in an interview that a new chip cluster called Project Rainier will contain "far more than" 100,000 chips. Amazon expects this cluster to become the world's largest dedicated AI hardware.

Amazon hopes that the company's third-generation AI semiconductor chips can prove competitive with NVIDIA (NVDA.US) products, providing AWS customers with alternatives when developing generative AI products. For most companies, NVIDIA's GPUs are currently the default hardware for such tasks, but they are expensive and often in short supply.

Amazon stated that starting early next year, it will offer customers computing power supported by NVIDIA's new Blackwell chips