{"slug": "amd-launches-helios-ai-rack-system-with-72-gpus-wins-microsoft-as-customer", "title": "AMD Launches Helios AI Rack System With 72 GPUs, Wins Microsoft as Customer", "summary": "AMD launched Helios, its first rack-scale AI system with 72 MI455X GPUs and 31TB of HBM4 memory delivering 2.9 exaflops of FP4 compute, and announced Microsoft as a customer for Azure deployment starting H2 2026. The system uses open-standard interconnects to challenge Nvidia's proprietary NVLink, with AMD shares rising 5% on the news and analysts raising price targets to $655-$700.", "body_md": "# AMD Launches Helios AI Rack System With 72 GPUs, Wins Microsoft as Customer\n\n- AMD's Helios rack system packs 72 MI455X GPUs with 31TB of HBM4 memory, delivering 2.9 exaflops of FP4 compute — roughly 50% more memory capacity than Nvidia's rival flagship\n[[1]](https://thenextweb.com/news/amd-helios-mi455x-72-gpu-rack-nvidia-rival) - Microsoft will deploy Helios on Azure for frontier AI inference and cloud services starting H2 2026, joining Meta, OpenAI, Oracle, and TCS as early adopters\n[[2]](https://finance.biggo.com/news/6551c951-c8b0-4081-9320-e080588226fb) - Rosenblatt raised its AMD price target to $655 and UBS to $700 following the announcement; AMD shares have rallied 139% year-to-date\n[[3]](https://finance.yahoo.com/technology/article/amd-stock-rises-on-price-target-bumps-microsoft-deal-170106894.html) - AMD's data center revenue is projected to reach $31.2 billion in 2026, up from $16.6 billion in 2025, with Q1 data center revenue already at $5.78 billion (+57% YoY)\n[[2]](https://finance.biggo.com/news/6551c951-c8b0-4081-9320-e080588226fb) - Helios engineering samples ship in H2 2026 with mass production targeted for Q2 2027; estimated rack pricing is $5-5.5 million versus $3.5-4 million for Nvidia's Vera Rubin\n[[4]](https://www.cnbc.com/2026/07/20/amd-helios-microsoft-ai-nvidia.html)\n\nAMD on Sunday unveiled Helios, the company's first rack-scale AI system, marking its most aggressive challenge yet to Nvidia's dominance in data center AI infrastructure. The system integrates 72 Instinct MI455X GPUs, sixth-generation EPYC \"Venice\" CPUs, and Pensando networking chips into a single double-wide server rack capable of delivering up to 2.9 exaflops of FP4 inference compute and 1.4 exaflops of FP8 training performance [1].\n\nMicrosoft has signed on to deploy Helios across its Azure cloud infrastructure starting in the second half of 2026, making it the highest-profile customer yet for AMD's integrated AI platform. Meta, OpenAI, Oracle, and Tata Consultancy Services are also early adopters, and AMD says eight of the world's top 10 AI companies now run workloads on its Instinct GPUs [2].\n\nAMD shares rose approximately 5% following the announcement, extending a 139% year-to-date rally. Rosenblatt Securities raised its price target to $655 from $490, while UBS increased its target to $700 from $670, with 82% of analysts now rating the stock a Buy [3].\n\n## The System\n\nHelios is built around 18 compute trays, each housing four MI455X GPUs paired with one EPYC Venice CPU and up to 12 networking chips. The full rack weighs nearly 7,000 pounds and contains 31 terabytes of HBM4 memory — approximately 50% more capacity than Nvidia's competing Vera Rubin NVL72 system. AMD claims 260 TB/s of scale-up memory bandwidth and 43 TB/s of scale-out networking bandwidth [1].\n\nA key architectural distinction from Nvidia: Helios uses open-standard interconnects — UALink for GPU-to-GPU communication within the rack and Ultra Ethernet Consortium specifications for rack-to-rack networking — rather than Nvidia's proprietary NVLink technology. AMD is positioning this as a strategic advantage for data center operators seeking to avoid single-vendor lock-in [4].\n\nThe system also integrates AMD's Pensando-derived Vulcano 800 AI NIC and Salina DPU, offering 800 Gbps Ethernet throughput to compete with Nvidia's ConnectX-8 and BlueField DPUs. The entire stack runs on AMD's ROCm open-source software platform [1].\n\n## The Microsoft Deal\n\nMicrosoft's commitment to Helios extends beyond GPU deployment. The company is also introducing two new Azure virtual machine series powered by AMD EPYC Venice processors and expanding Pensando DPU integration across Azure's network infrastructure, representing a full-stack AMD adoption [2].\n\nAMD CEO Lisa Su described the deployment as 'an important milestone,' noting the companies are 'extending a long-running collaboration to scale the next wave of AI infrastructure' [3].\n\n## Market Context and Competition\n\nThe Helios launch arrives as AMD's data center business is on a steep growth trajectory. The segment generated $5.78 billion in Q1 2026 revenue, up 57% year-over-year, and AMD has guided for approximately $11.2 billion in Q2 revenue, implying roughly 46% annual growth. Full-year 2026 data center revenue is projected at $31.2 billion, nearly double the $16.6 billion recorded in 2025 [2].\n\nFuturum Group analyst Daniel Newman estimated Helios rack pricing at $5 million to $5.5 million, compared with $3.5 million to $4 million for Nvidia's Vera Rubin — a premium AMD will need to justify with performance and ecosystem advantages. Newman projected AMD could reach 20% to 25% market share in the AI accelerator market, representing 'hundreds of billions of dollars in revenue' [2].\n\nCounterpoint Research analyst Neil Shah acknowledged that 'AMD's Helios chips are comparable to Nvidia's GPUs and CPUs in performance' but flagged the critical challenge: software optimization and Nvidia's entrenched CUDA ecosystem remain formidable barriers to broader adoption [2].\n\n## What's Next\n\nEngineering samples of Helios are shipping in the second half of 2026, with mass production targeted for Q2 2027. AMD plans to showcase the system at its Advancing AI conference in San Francisco, where CEO Lisa Su will deliver the keynote [[3]](https://finance.yahoo.com/technology/article/amd-stock-rises-on-price-target-bumps-microsoft-deal-170106894.html) [4].\n\nThe announcement represents a structural shift in AMD's competitive positioning — from selling individual GPUs into hyperscale data centers to offering a fully integrated, rack-scale computing platform that matches the approach Nvidia has pursued with its Grace Blackwell and Vera Rubin systems. Whether AMD's bet on open standards over Nvidia's proprietary ecosystem proves decisive will depend on how quickly the ROCm software stack matures and whether hyperscalers prioritize vendor flexibility over Nvidia's proven integration [4].\n\n## Companies mentioned\n\n## Further sources\n\n[[1] AMD Helios technical specifications and architecture details — The Next Web ↗](https://thenextweb.com/news/amd-helios-mi455x-72-gpu-rack-nvidia-rival)\n\n[[2] AMD Lands Microsoft as Flagship Customer for New Helios AI System — BigGo Finan… ↗](https://finance.biggo.com/news/6551c951-c8b0-4081-9320-e080588226fb)\n\n[[3] AMD stock rises on price target bumps, Microsoft deal — Yahoo Finance ↗](https://finance.yahoo.com/technology/article/amd-stock-rises-on-price-target-bumps-microsoft-deal-170106894.html)\n\n[[4] AMD launches Helios, its first rack AI system to rival Nvidia, adding Microsoft… ↗](https://www.cnbc.com/2026/07/20/amd-helios-microsoft-ai-nvidia.html)\n\nThe stories that matter, in one email. Free — unsubscribe anytime.", "url": "https://wpnews.pro/news/amd-launches-helios-ai-rack-system-with-72-gpus-wins-microsoft-as-customer", "canonical_source": "https://mlq.ai/news/amd-launches-helios-ai-rack-system-with-72-gpus-wins-microsoft-as-customer/", "published_at": "2026-07-21 20:43:23.889014+00:00", "updated_at": "2026-07-21 20:43:26.145139+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-infrastructure", "ai-chips", "ai-products", "ai-startups"], "entities": ["AMD", "Helios", "MI455X", "Microsoft", "Azure", "Nvidia", "Meta", "OpenAI"], "alternates": {"html": "https://wpnews.pro/news/amd-launches-helios-ai-rack-system-with-72-gpus-wins-microsoft-as-customer", "markdown": "https://wpnews.pro/news/amd-launches-helios-ai-rack-system-with-72-gpus-wins-microsoft-as-customer.md", "text": "https://wpnews.pro/news/amd-launches-helios-ai-rack-system-with-72-gpus-wins-microsoft-as-customer.txt", "jsonld": "https://wpnews.pro/news/amd-launches-helios-ai-rack-system-with-72-gpus-wins-microsoft-as-customer.jsonld"}}