Nvidia's Biggest Threat? AMD's New Chips Are Built For AI War

Benzinga
2025.09.03 14:47
portai
I'm PortAI, I can summarize articles.

Nvidia's dominance in AI chips is being challenged by AMD's new MI350 series GPUs, which offer higher performance and cost efficiency. AMD aims to capture market share in inference workloads, with plans for the MI400 series and Helios rack solution by late 2026. The competition is intensifying as Nvidia and other companies innovate to maintain their positions, reflecting a trend towards diversified supply chains in the AI chip market.

Nvidia Corp's NVDA dominance in AI chips is facing its fiercest challenge yet as Advanced Micro Devices Inc’s AMD new GPUs step squarely into the AI arms race. At this year's Hot Chips 2025, AMD unveiled details of its MI350 series while reiterating a 2026 launch for the MI400 lineup—advancements aimed at shaking up Nvidia's grip on AI infrastructure.

Track AMD stock here.

With explosive AI adoption driving unprecedented demand for compute, memory, and networking, AMD is positioning itself to capture significant market share in inference workloads where performance and cost efficiency are key differentiators.

Read Also: Broadcom Vs. Nvidia: The AI Chip War You’re Not Watching

AMD Ups The Ante

AMD's MI355X GPU, a flagship in the MI350 series, boasts higher power consumption and clock speeds than its MI350X sibling, delivering a 9% bump in performance. Designed for liquid-cooled data centers, MI355X racks can scale up to 128 GPUs, double that of air-cooled MI350X racks. AMD is also betting on ease of integration, touting compatibility with Nvidia-trained AI models, a move that could lure hyperscalers seeking flexible alternatives to Nvidia's dominant ecosystem. The company confirmed that its MI400 series and Helios rack solution are on track for a late-2026 rollout, followed by the MI500 in 2027, signaling long-term ambitions to stay competitive.

Nvidia's Rivals Close The Gap

Nvidia isn't standing still. Its MGX architecture-powered NVL72 design offers customers such as Meta unprecedented customization, while its Spectrum-X ethernet solutions target scaling challenges across massive AI clusters.

Google's Ironwood TPU, co-designed with Broadcom, is also narrowing the performance gap with Nvidia's B200 GPUs, pushing hyperscaler investment into custom silicon. These developments underscore a rapidly evolving AI chip race, with Nvidia's dominance increasingly under pressure from AMD and specialized XPU solutions.

Networking and optics are emerging as critical battlegrounds too, with companies like Broadcom and Ayar Labs innovating on scale and efficiency. AMD's momentum highlights a clear trend: hyperscalers are seeking diversified supply chains and competitive alternatives, reshaping the AI chip landscape.

  • Google's No Longer Just Search – And Microsoft's Not Sleeping Well

Photo: Shutterstock

Source: Benzinga The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.