The MI400 series of chips will be the basis of a new server called "Helios" that AMD plans to release next year.
The move comes as the competition between Nvidia and other AI chip firms has shifted away from selling individual chips to selling servers packed with scores or even hundreds of processors, woven together with networking chips from the same company.
During its keynote presentation, AMD said that many aspects of the Helios servers - such as the networking standards - would be made openly available and shared with competitors such as Intel.
The move was a direct swipe at market leader Nvidia, which uses proprietary technology called NVLink to string together its chips but has recently started to license that technology as pressure mounts from rivals.
"The future of AI is not going to be built by any one company or in a closed ecosystem. It's going to be shaped by open collaboration across the industry," Su said.
Su was joined onstage by OpenAI's Sam Altman, who said his company is using AMD's MI300X and MI450 chips.
"Our infrastructure ramp-up over the last year, and what we're looking at over the next year, have just been a crazy, crazy thing to watch," Altman said.
During her speech, executives from billionaire Elon Musk-owned xAI, Meta Platforms and Oracle took to the stage to discuss their respective uses of AMD processors. Crusoe, a cloud provider that specializes in AI, told Reuters it is planning to buy $400 million of AMD's new chips.
AMD's Su reiterated the company's product plans for the next year, which will roughly match the annual release schedule that Nvidia began with its Blackwell chips.
AMD has struggled to siphon off a portion of the quickly growing market for AI chips from the dominant Nvidia. But the company has made a concerted effort to improve its software and produce a line of chips that rival Nvidia's performance.
AMD completed the acquisition of server builder ZT Systems in March. As a result, AMD is expected to launch new complete AI systems, similar to several of the server-rack-sized products Nvidia produces.
Santa Clara, California-based AMD has made a series of small acquisitions in recent weeks and has added talent to its chip design and AI software teams. At the event, Su said the company has made 25 strategic investments in the past year that were related to the company's AI plans.
Last week, AMD hired the team from chip startup Untether AI. On Wednesday, AMD said it had hired several employees from generative AI startup Lamini, including the co-founder and CEO.
AMD's software called ROCm has struggled to gain traction against Nvidia's CUDA, which is seen by some industry insiders as a key part of protecting the company's dominance. When AMD reported earnings in May, Su said that despite increasingly aggressive curbs on AI chip exports to China, AMD still expected ..
AMD's software called ROCm has struggled to gain traction against Nvidia's CUDA, which is seen by some industry insiders as a key part of protecting the company's dominance. When AMD reported earnings in May, Su said that despite increasingly aggressive curbs on AI chip exports to China, AMD still expected strong double-digit growth from AI chips.
Source Name : Economic Times