IBM, AMD, and Zyphra Partner on Scalable AI Infrastructure

• Zyphra will train multimodal AI models using AMD’s MI300X GPUs on IBM Cloud, marking a milestone in open-source superintelligence development.

IBM and AMD have entered a multi-year collaboration with Zyphra, a San Francisco-based open-source AI company, to deliver advanced infrastructure for training frontier AI models. The agreement includes the deployment of AMD Instinct MI300X GPUs on IBM Cloud, forming one of the largest generative AI training clusters to date. Zyphra will use this setup to develop multimodal foundation models across language, vision, and audio domains. These models will power Maia, a general-purpose AI agent designed to assist enterprise knowledge workers.

Supporting Open-Source Superintelligence Research

Zyphra recently secured $1 billion in Series A funding to build a research lab focused on neural network innovation, long-term memory, and continual learning. The company selected IBM and AMD for their ability to deliver GPU accelerators at scale and pace, aligning with Zyphra’s rapid development goals. This marks the first large-scale training cluster on IBM Cloud using AMD’s full-stack platform, including Pollara 400 AI NICs and Ortano DPUs. Initial deployment began in September, with further expansion planned for 2026.

The infrastructure will support Zyphra’s mission to advance open-science AI systems that are both transparent and scalable. By integrating compute and networking components, the collaboration aims to streamline model training and reduce latency. IBM and AMD are positioned to scale resources as Zyphra’s needs evolve. Their combined expertise offers a foundation for sustained innovation in multimodal AI.

Enhancing Performance and Efficiency for AI Workloads

IBM Cloud’s secure and scalable environment complements AMD’s high-performance computing capabilities. The MI300X accelerators were introduced as a service last year, targeting generative AI and HPC applications. According to IBM Cloud GM Alan Peacock, faster and more efficient scaling of AI workloads is essential for maximizing return on investment. AMD’s Philip Guido added that the partnership sets a new standard for infrastructure speed and scale.

Together, the companies aim to support both startups and enterprises in deploying smarter AI solutions. Hybrid multi-cloud options are part of the offering, helping clients optimize cloud value and performance. The collaboration reflects a broader shift toward modular, scalable AI environments. Zyphra’s use case may serve as a blueprint for future deployments across industries.

Quantum-Centric Supercomputing on the Horizon

IBM and AMD have also announced plans to co-develop quantum-centric supercomputing architectures. This initiative will combine IBM’s advancements in quantum hardware and software with AMD’s leadership in AI acceleration. The goal is to create systems capable of solving complex problems beyond the reach of classical computing. If successful, these architectures could redefine the boundaries of AI and computational science.