New Delhi: IBM and AMD have entered a multi-year partnership with San Francisco-based open-source AI company Zyphra to provide advanced infrastructure for training next-generation artificial intelligence models. The collaboration brings IBM’s cloud computing platform together with AMD’s Instinct MI300X GPUs to power one of the largest deployments for training multimodal foundation models.
Zyphra, valued at $1 billion after its recent Series A funding round, is positioning itself as an open-science lab focused on superintelligence research. The company is working on novel neural architectures, systems for long-term memory, and continual learning methods. Its central project, Maia, is designed as a general-purpose AI agent intended to support productivity tasks across industries.
The deployment began in September and will expand through 2026. According to Zyphra, the collaboration marks the first integration of AMD’s full-stack training platform at scale within IBM Cloud. Alongside GPUs, the infrastructure incorporates AMD’s networking technologies, including Pensando Pollara AI NICs and Ortano DPUs, to improve training efficiency.
Executives from the three companies framed the partnership as a step toward creating enterprise-grade AI systems in open-source environments. Zyphra’s CEO Krithik Puthalath described the effort as key to advancing open-source superintelligence. IBM highlighted scalability and cost efficiency as central to the agreement, while AMD pointed to the growing role of its high-performance computing platforms in AI workloads.
The collaboration reflects a broader trend of alliances between cloud providers, semiconductor companies, and AI startups as they seek to build computing systems capable of supporting increasingly complex models. It also ties into IBM and AMD’s ongoing exploration of quantum-centric and hybrid multi-cloud computing, an area seen as integral to the future of AI development.
