Advanced Micro Devices CEO Lisa Su unveiled a new artificial intelligence server, set for release in 2026, in a bold move to compete with Nvidia’s (NASDAQ:NVDA) top-tier AI offerings. The announcement came as OpenAI CEO Sam Altman revealed the ChatGPT maker would begin using AMD’s (NASDAQ:AMD) latest chips.
Speaking at the “Advancing AI” developer conference in San Jose, California, Su introduced the MI350 and MI400 series of AI chips, designed to rival Nvidia’s powerful Blackwell processors.
The upcoming MI400 chips will power a new AI server called “Helios,” scheduled for launch in 2025. This development reflects a shift in the AI chip market—from focusing on standalone processors to selling fully integrated server systems containing dozens or hundreds of interconnected chips, complete with in-house networking components.
According to AMD executives, each Helios server will house 72 MI400 chips, placing it in direct competition with Nvidia’s NVL72 server systems.
In a notable departure from Nvidia’s closed approach, AMD stated it will make much of Helios’ technical architecture—such as networking protocols—open to the broader industry, including rivals like Intel (NASDAQ:INTC). This contrasts with Nvidia’s proprietary NVLink system, which it has only recently started licensing amid growing competition.
Su emphasized the importance of collaboration in AI’s future, stating, “No single company will build the future of AI alone. It will be shaped through open industry partnerships.”
Su was joined on stage by OpenAI’s Sam Altman, who confirmed that OpenAI is working with AMD to help shape the design of the upcoming MI450 chips to better serve AI workloads. “The scale of our infrastructure buildout over the past year—and what’s ahead—has been astounding,” Altman noted.
Executives from xAI (Elon Musk’s AI venture), Meta Platforms (NASDAQ:META), and Oracle (NYSE:ORCL) also appeared at the event to highlight how their companies are deploying AMD hardware. Meanwhile, AI-focused cloud provider Crusoe revealed plans to purchase $400 million worth of AMD’s new chips.
Su reiterated AMD’s commitment to releasing new AI hardware annually, keeping pace with Nvidia’s product cadence established by its Blackwell line.
Despite the announcements, AMD’s stock closed down 2.2%. Summit Insights analyst Kinngai Chan commented that Thursday’s chip news is unlikely to immediately shift AMD’s competitive position in the AI market, where it continues to trail Nvidia.
Still, AMD has been ramping up efforts to close the gap. The company recently acquired server maker ZT Systems, positioning itself to offer complete AI systems akin to Nvidia’s rack-scale solutions.
Headquartered in Santa Clara, AMD has also made a flurry of smaller acquisitions and talent hires to strengthen its AI and chip design capabilities. Su noted the firm has made 25 strategic AI-related investments in the past year alone.
Just last week, AMD hired the full team from chip startup Untether AI. It also brought on several key personnel from the generative AI firm Lamini, including its co-founder and CEO.
One ongoing challenge for AMD is building out its AI software ecosystem. Its ROCm platform has yet to match the popularity and performance of Nvidia’s CUDA, a core part of Nvidia’s sustained leadership in AI.
Despite tighter U.S. restrictions on exporting AI chips to China, AMD remains optimistic. In its May earnings call, Su projected robust double-digit growth in the company’s AI chip segment this year.
