AMD is taking the fight to NVIDIA Corp. to loosen its ironclad grip on the multibillion-dollar artificial intelligence (AI) hardware market.
Culminating a decade-long corporate comeback, AMD said Monday that Microsoft Corp. will deploy its highly anticipated Helios rack-scale AI system.
Microsoft will integrate Helios into its Azure cloud platform to power frontier model inference, support internal AI services, and provide compute resources to external enterprise customers. Additionally, the tech giant will introduce two new computing instances powered by AMD’s sixth-generation EPYC Venice central processing units (CPUs) to handle agentic AI, data pipelines, and semiconductor design.
While financial terms and exact compute capacities remain undisclosed, AMD confirmed that shipments to Microsoft and other primary customers are scheduled to begin later this year.
“We are expanding the Azure infrastructure portfolio with AMD Helios to give customers the performance, scale, and choice they need to build and run the next generation of AI applications,” Microsoft CEO Satya Nadella said in a statement.
The rollout comes at a critical juncture for Microsoft. Despite launching seven in-house AI models in June, the company has seen mixed results from its 365 Copilot and GitHub Copilot platforms, leaving it as the worst-performing Magnificent Seven tech stock so far this year.
By adopting Helios, Microsoft joins an expanding roster of industry leaders turning to AMD to alleviate severe hardware shortages.
Meta Platforms Inc. previously committed to deploying up to six gigawatts of AMD GPUs over time, starting with an initial one-gigawatt deployment on Helios racks this year. OpenAI, Oracle Corp., Cohere, Elon Musk’s SpaceXAI, and India’s Tata Consultancy Services have also secured commitments for the system. According to AMD, eight of the world’s top ten AI companies now run workloads on its Instinct graphics processing units (GPUs).
Helios represents AMD’s first fully integrated, rack-scale platform, combining its proprietary Instinct MI455X GPUs, Venice CPUs, Pensando networking chips, and ROCm software. AMD data center head Forrest Norrod noted that the system focuses heavily on providing the “lowest cost per token” for enterprise buyers. AMD CEO Lisa Su previously asserted that Helios offers significant memory bandwidth and capability advantages over NVIDIA’s competing systems during inference tasks.
The system is aimed squarely at NVIDIA’s popular Grace Blackwell and Vera Rubin architectures. NVIDIA currently commands upward of 95% of the data center GPU market, compared to AMD’s 4.5%. However, the massive scale of Helios could disrupt this monopoly. Industry analysts at the Futurum Group estimate the market share for AMD could realistically surge to 20% or 25%, translating into hundreds of billions of dollars in future revenue. Futurum estimates the massive, 7,000-pound Helios rack will retail between $5 million and $5.5 million.
The product launch follows a powerhouse financial period for AMD, which reported a 57% year-over-year increase in data center revenue to $5.78 billion in the first quarter. AMD projects that Helios will serve as the primary growth driver to secure tens of billions of dollars in AI-related revenue from 2027 onward.
