AMD Fires Back at Nvidia with Helios AI System, Epyc CPUs

AMD CEO Lisa Su announced the Helios rack-scale AI system and sixth-generation Epyc 9006 Series CPUs at the Advancing AI conference in San Francisco on July 23, directly challenging Nvidia's dominance in AI infrastructure. Helios, powered by Instinct MI455X GPUs, delivers 15% more AI compute, 50% more memory, and 50% more scale-out bandwidth than Nvidia's Vera Rubin NVL72 system, with shipments from partners starting by the end of Q3 2026. Su projected the AI infrastructure market will reach $1.4 trillion by 2030, up from a previous estimate of $500 billion by 2028, driven by agentic AI.

AMD Fires Back at Nvidia with Helios AI System, Epyc CPUs AMD’s latest hardware innovations aim to reshape the AI infrastructure market, challenging Nvidia’s dominance and driving the next wave of AI adoption. Four months after Nvidia unveiled its Vera Rubin AI infrastructure and targeted x86 server chips with Vera, AMD has countered with its sixth-generation Epyc CPUs and the Helios rack-scale AI solution, powered by its new Instinct MI455X GPUs. At its Advancing AI conference in San Francisco today July 23 , AMD CEO Lisa Su announced that Helios is in full production, with shipments from hardware partners scheduled to begin by the end of the third quarter of 2026. Helios is a fully integrated rack-scale /servers/what-is-rack-scale-computing-and-why-is-it-relevant-again- AI system, featuring Epyc processors, Instinct MI455X GPUs, Pensando networking, and ROCm software. The Helios system /data-center-chips/amd-bolsters-ai-chips-and-software-to-further-challenge-nvidia and Epyc CPUs are designed to address the growing complexity and scale of agentic AI, which is reshaping how data centers plan and deploy their infrastructure. Industry analysts said AMD’s latest announcements represent a pivotal moment in the competitive AI infrastructure landscape, as the company directly challenges Nvidia’s market dominance with an integrated strategy encompassing both GPUs and CPUs. AMD CEO Lisa Su highlighted the rapid growth of agentic AI, stating: “Agents are scaling from millions to billions, driving unprecedented infrastructure needs.” Getty Images The Battle for Compute When compared to Nvidia’s Vera Rubin NVL72 system, /data-center-chips/gtc-2026-nvidia-unveils-vera-rubin-ai-platform-eyes-1t-by-2027 Su said Helios delivers 15% more AI compute, 50% more memory, and 50% more scale-out bandwidth. She also said Helios delivers up to 30% more tokens per dollar. “Every Helios can deliver more performance for the largest models, more capacity for longer context, and the bandwidth to scale across thousands of racks,” Su said during her keynote speech. AMD also launched its Epyc 9006 Series server CPUs, most code-named Venice, a family of four purpose-built processors for enterprise, cloud, high-performance computing, and AI workloads, including agentic AI. Each chip is optimized for different needs. AMD’s chiplet architecture allows the company to build an entire portfolio of chips, she said. “Venice is not one chip. Venice is actually an entire family of chips,” she said. AMD stated that the new CPUs offer a performance advantage over Nvidia’s Arm-based Vera. A 256-core Venice CPU offers 2.2 times the throughput of an 88-core Vera CPU /data-center-chips/nvidia-says-vera-rubin-vera-cpu-on-track-launches-dsx-os-to-run-ai-factories , while a 96-core Venice chip offers a 20% better per-core performance, AMD said. The company said its new CPUs outperform Intel and Arm’s processors as well. AMD's Market Outlook Today’s launches come as AMD works to make further inroads in both the GPU and CPU markets and overall AI infrastructure market, even as rivals target those segments amid surging AI demand. Su told the crowd at San Francisco’s Moscone Center that the driving force behind all of this is AI, particularly agentic AI. She explained that the shift to agentic AI is fueling significant growth in the AI infrastructure market. Last year, she projected the market would reach $500 billion by 2028, but now she expects it to reach $1.4 trillion by 2030, approaching the size of today’s entire semiconductor market. GPUs will continue to make up much of that market, she said. Su said agentic AI is also creating a new growth vector for server CPUs. The market, now $25 billion, is expected to reach $200 billion by 2030, she said. While GPUs do all the reasoning, organizations need a lot of CPUs to orchestrate every step around it, from calling tools to retrieving data, she said. “What we’re seeing is that the rate and pace of agentic AI adoption is much faster than any of us thought. So, we're talking about agents going from millions to billions,” Su said. “We're able to increase all of our productivity, and that requires a tremendous amount of CPU infrastructure, and every customer conversation is telling us that this build out is just beginning. AMD’s Epyc CPUs are purpose-built to handle AI, HPC, and enterprise workloads with tailored performance for diverse needs. Image: AMD Analysts Weigh In on AMD’s Strategy Industry analysts described AMD’s announcements as highly significant for the AI infrastructure market. IDC analyst Ashish Nadkarni said AMD is the only company that can match what Nvidia offers in data center and AI infrastructure, including GPUs and DPUs. “They are making steady progress to be the strong No. 2 in the market,” he said. “There is nobody else in the semiconductor industry today that can offer what AMD is offering other than Nvidia.” The two rivals are competing on rack-scale AI system designs powered by their chips as data center operators seek the most powerful, but efficient infrastructure to support their demanding training and inferencing workloads. Nvidia set the pace years ago through its integrated, single vendor designs like GB200 NVL72 and more recently Vera Rubin NVL72. Nvidia detailed /data-center-chips/gtc-2026-nvidia-unveils-vera-rubin-ai-platform-eyes-1t-by-2027 its Vera Rubin AI platform, including its Vera CPU, at its GTC conference in March. AMD’s Helios is a direct answer to Nvidia’s NVL72 platforms, said Matt Kimball, vice president and principal analyst at Moor Insights & Strategy. “Helios is a very strong response to that,” he said. Helios, powered by AMD’s Instinct MI450 Series GPUs, is already gaining traction. This week, Microsoft said it will deploy https://newsroom.amd.com/news/microsoft-azure-ai-infrastructure/ Helios to power AI workloads, including frontier model AI inference and to support Azure AI services. Anthropic this week also announced https://ir.amd.com/news-events/press-releases/detail/1292/amd-and-anthropic-announce-strategic-partnership-to-deploy-up-to-2-gigawatts-of-amd-instinct-mi450-series-gpus it will deploy up to 2 GW of the GPUs in Helios solutions. “That is big traction. You’re quietly seeing Instinct market share increase quarter over quarter, year over year, so it's starting to establish,” Kimball said. AMD said its MI455X GPU delivers 34 times higher token throughput and up to 18 times lower token cost than its previous MI355X GPU. While AMD is competing with Nvidia in GPUs and rack-scale systems, Nvidia is taking on AMD in CPUs. During the keynote, Su said AMD’s CPUs reached 46% revenue share of the server market last quarter with more than 60% of the Fortune 100 running on Epyc. But Nvidia isn’t the only one eyeing the CPU market. To meet AI demand, Arm /data-center-chips/arm-enters-data-center-chip-fray-with-agi-cpu-for-ai-infrastructure and Qualcomm /data-center-chips/qualcomm-lands-meta-cpu-deal-unveils-ai-data-center-platform have also introduced their own Arm-based CPUs to compete against Intel and AMD’s x86 CPUs. “Nvidia wants you to look at Vera independently as a standalone CPU and worthy of being competitive with Intel, AMD, and Arm's AGI CPU,” Kimball said. Through benchmarks, however, AMD today is claiming that its sixth-generation CPUs outshine Vera in performance. “It doesn't matter how you measure Epyc versus Vera,” Kimball said. “AMD is saying across any of those measurement vectors, not only do we win, but we win decisively. That's a strong story.” Epyc Gains During the keynote, Su said AMD’s Epyc CPUs reached 46% revenue share of the server market last quarter with more than 60% of the Fortune 100 running on Epyc. AMD’s move to offer a family of Epyc chips tailored for specific workloads is new for the company, Kimball said. Nvidia takes a similar approach with Vera, targeting analytics, HPC and agentic AI, he said. Nadkarni said AMD's x86 architecture gives it an edge with customers reluctant to rewrite their software stacks for Arm. Overall, with today’s announcements, AMD is well-positioned to win more business. “They’re very disciplined in what they’re doing,” he said of AMD’s approach to the AI market. “As the market has changed, AMD has grown with it.” Next-Gen Chipsets The company’s sixth-generation Epyc lineup includes four processors: AMD Epyc 9006 SP7: Built for high-density agent sandbox execution, this CPU has up to 256 cores. It will ship in the fourth quarter of 2026. AMD Epyc 9006 SP8: Available in 8- to 128-core configurations with low-power options, this processor is built for general-purpose enterprise workloads and agent sandbox execution where efficiency is a top priority. It will ship in the first half of 2027. AMD Epyc 9006X SP7: Designed for high-density agent sandbox execution, this CPU supports up to 256 cores. AMD Epyc 9006 LP: Formerly code-named Verano, this processor is built to serve as an AI host node CPU for next-generation rack-scale AI systems, with up to 72 cores. It will ship in the second half of 2027. At the end of the keynote, Su previewed future technology. She stated that seventh-generation CPUs, codenamed Florence, are on track for 2028, while eighth-generation CPUs are slated for 2030. AMD also plans a new GPU and Helios system every year with the Instinct MI500 planned for 2027 and the MI600 slated for 2028. On the software front, AMD also announced ROCm.ai https://ir.amd.com/news-events/press-releases/detail/1294/aai-2026-amd-delivers-full-stack-compute-for-the-agentic-ai-era , offering an AI-native software experience for AI development across AMD platforms. It combines ROCm CLI, AMD Skills and AI-powered software optimization into a unified developer experience, the company said.