AMD launches Helios, its first rack AI system to rival Nvidia, adding Microsoft as newest buyer

After a notable resurgence, Advanced Micro Devices (AMD) is set to launch Helios, its inaugural rack-scale system designed for artificial intelligence applications. This significant development positions AMD as a direct competitor to Nvidia's widely acclaimed Grace Blackwell and Vera Rubin systems. The tech giant Microsoft has recently announced its intention to incorporate Helios into its data centers, joining other major players like Meta, OpenAI, and Oracle in the race for enhanced computational power. Beginning later this year, AMD will start shipping Helios to its clients, including Microsoft. Although specific financial details and compute capacity have not been disclosed, Microsoft CEO Satya Nadella emphasized the importance of this addition to Azure's infrastructure, stating it would provide the performance and scalability necessary for next-generation AI applications. Helios is poised to enhance model inference for Microsoft and its suite of AI products while supporting Azure AI services. Additionally, Microsoft will introduce two new computing instances leveraging AMD's latest Venice CPUs, aimed at optimizing AI and data processing tasks, as well as semiconductor design. This partnership builds on a longstanding relationship, with AMD providing chips for various Microsoft products, including Surface PCs and Xbox consoles. Earlier this year, Microsoft was also the first to utilize AMD's MI300X GPUs, which are seen as rivals to Nvidia's offerings. The increasing demand for computational resources has prompted companies like Microsoft to seek out AMD's solutions, especially as they expand their AI initiatives. AMD claims that eight of the top ten AI companies use its Instinct GPUs, including notable organizations like OpenAI and SpaceXAI. In a strategic move, Meta plans to utilize up to 6 gigawatts of AMD GPUs, starting with 1 gigawatt on Helios racks. Moreover, OpenAI and Oracle have committed to deploying Helios this year, alongside Tata Consultancy Services from India. Recently, CNBC gained exclusive access to a Helios system, showcasing its development at AMD's Texas data center. Named after the Greek sun god, Helios represents a synthesis of AMD's capabilities in GPUs, CPUs, networking, and software. According to Forrest Norrod, AMD's data center head, the company is focused on delivering an optimal total cost of ownership and competitive pricing for its customers. AMD's CEO, Lisa Su, previously highlighted Helios's advantages over Nvidia's systems, particularly in terms of inference performance and memory capabilities. While AMD has not disclosed pricing, estimates suggest Helios could range from $5 million to $5.5 million, in contrast to Nvidia's Vera Rubin, which is estimated to cost between $3.5 million and $4 million. With Nvidia holding a commanding 95% share of the data center GPU market, AMD's current 4.5% share could see a significant boost with the introduction of Helios. Analysts predict that AMD could capture 20-25% of the market, which could translate into substantial revenue potential. In the first quarter of 2026, data centers were reported to account for a majority of AMD's revenue, growing 57% year-over-year, and the company anticipates tens of billions in AI revenue starting in 2027, largely attributed to Helios. While Intel currently leads the CPU market, AMD has been steadily improving its share. The development of Helios, powered by advanced EPYC CPUs and a robust networking infrastructure, illustrates AMD's strategic efforts to reclaim its standing in the data center market. Analysts have noted that Helios's performance is comparable to Nvidia's offerings, but the real differentiator may lie in AMD's software optimization. As the market evolves, the success of Helios will hinge on its early deployment outcomes and whether AMD can establish itself as a technological leader amidst growing demand for AI capabilities.