Zyphra has announced the availability of 15 megawatts of AMD Instinct™ MI355X GPU capacity through Zyphra Cloud, its full-stack "neocloud" platform. This expansion transitions Zyphra Cloud from offering serverless inference for open-source models to providing robust, bare-metal AMD infrastructure. Designed for AI-native startups, enterprises, and hyperscalers, the platform leverages software-driven infrastructure to accelerate the deployment of large-scale pretraining, reinforcement learning, and agentic inference workloads.
Massive Scale: 15MW of AMD Instinct MI355X GPU capacity is now live and available.
Deployment Models: Offers on-demand bare-metal GPU clusters and custom hyperscale infrastructure.
Software-Driven Infrastructure: Integrates optimizations from Zyphra Research, including the ZAYA1 model series and TSP inference algorithm.
High-Performance Networking: Utilizes IBM-partnered clusters featuring AMD Pensando Pollara 400G networking.
Agentic Inference: Powers serverless endpoints for leading models like DeepSeek and Kimi on MI355X hardware.
Future Roadmap: Plans are already in place to expand to next-generation AMD MI450 series architectures.
Zyphra Cloud distinguishes itself by being deeply integrated with Zyphra’s own research and model development. Unlike generic cloud providers, Zyphra has validated its infrastructure by training its own models, such as ZAYA1-8B and ZAYA1-74B, end-to-end on AMD hardware. This "research-to-production" pipeline ensures that the systems are pre-optimized for the specific demands of modern AI, particularly in long-context and agentic workflows.
"AMD has rapidly emerged as a major compute platform for AI," said Krithik Puthalath, Founder and CEO of Zyphra. "Zyphra Cloud is not just another AMD neocloud. We've demonstrated expertise across the full AI lifecycle... and now deliver that as software-driven infrastructure to enable customers to go-to-market faster on AMD."
The platform’s strength lies in its vertical integration. By combining AMD’s latest silicon with proprietary inference algorithms like TSP (Temporal Sequence Processing), Zyphra Cloud addresses the efficiency bottlenecks often found in large-scale deployments. For enterprise clients, this means more reliable scaling and reduced latency for production-grade AI agents. Zyphra’s collaboration with IBM and the use of Pollara networking further solidify the platform’s capacity to handle the industry’s most intensive compute requirements.
Zyphra is an open superintelligence research and product company based in San Francisco, CA, building full-stack AI systems across research, infrastructure, and applications. Zyphra's mission is to build human-aligned AI that helps individuals and organizations reach their fullest potential. With a focus on the AMD ecosystem, Zyphra is accelerating the path to scalable, high-performance AI production.
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.