Call Nvidia (NVDA.US)! AMD (AMD.US) Releases Next-Generation AI Chips and Server Products Claiming Better Performance Than Competitors

The Zhitong Finance App learned that on Thursday, AMD (AMD.US) released a series of new products for data centers, including next-generation server processors, AI accelerators and server architectures, and claimed that the performance of many products will surpass rival Nvidia (NVDA.US) to further compete for the rapidly growing artificial intelligence (AI) computing power market.

At the new product launch in San Francisco, AMD unveiled a new generation of EPYC server processors, the Mi400 series AI accelerator, and a new Helios server rack design. The company said that the new product is an important upgrade to the existing product line and has an advantage over competitors in terms of performance.

AMD is showing increasing confidence and hopes to close the gap with Nvidia in the field of AI infrastructure. In recent years, Nvidia has led the construction of global AI data centers with GPU products, generating hundreds of billions of dollars in revenue. AMD believes that as companies' demand for AI computing power continues to grow, both sides are speeding up the launch of next-generation chips with better performance.

However, the capital market reacted relatively cautiously to this press conference. AMD stock fell about 3.5% during the intraday session on Thursday. Prior to the press conference, the company's stock price had more than doubled this year, indicating that the market had previously placed high expectations for new products.

AMD predicts that by 2030, the global AI accelerator market will increase to about 1.4 trillion US dollars; the traditional data center CPU market will reach 220 billion US dollars. Taken together, the company expects its serviceable market (TAM) to reach about $2 trillion by 2030.

At the press conference, executives from a number of technology companies appeared to support AMD's new products, including two leading AI model developers, OpenAI and Anthropic.

Among them, AMD highlighted the new Helios server rack. The solution uses an open design and can be manufactured and sold by server manufacturers themselves. A single rack can be equipped with up to 72 MI400 series AI accelerators. Among them, the MI455 model achieved significant improvements compared to the previous generation in terms of memory capacity, inference speed, and operating costs.

Sachin Katti, head of the OpenAI web platform, said that as the demand for AI infrastructure continues to grow rapidly, the company plans to deploy Helios servers on a large scale.

He said, “If you haven't heard enough, I need more computing power, and I need to get more computing power as soon as possible.”

At the same time, AMD also unveiled a new generation of EPYC server processors using the Venice architecture. The company said the product will continue to be ahead of Nvidia's corresponding Vera server CPUs in terms of performance, and Nvidia previously stated that Vera will help the company further enhance its data center competitiveness.

Additionally, AMD also announced a partnership with AI chip company Cerebras Systems (CBRS.US). The two sides will jointly launch an AI infrastructure solution integrating Helios servers and Cerebras systems for enterprise customers that require extremely high inference speed.

According to the cooperation plan, AMD chips will be responsible for processing and parsing user requests, while Cerebras AI chips equipped with large storage capacity will be responsible for quickly generating inference results. The two sides said the plan will first be deployed to Cerebras' own data center in the fourth quarter and then opened to other AI companies.

Cerebras shares rose 4.86% on Thursday, boosted by news of the partnership.

Andrew Feldman, CEO of Cerebras, said that the two parties will launch related products earlier than similar solutions after Nvidia integrated the recent acquisition of Groq-related assets. He said that the company has been developing this joint product for several months and is confident that it will be the first to enter the market.

Feldman said, “We will be the first to enter the market, and we are confident about that. This is a product built for customers who value responsiveness extremely.”