Zettabyte has launched a managed cloud service that gives organisations API access to open-source and open-weight AI models, reducing the infrastructure needed to deploy and run them.
Zettabyte has launched Model-as-a-Service (MaaS) on zCLOUD, turning access to open-source and open-weight AI models into an on-demand cloud service through a single API. The service gives organisations access to models including Kimi, GLM, Llama, GPT-OSS, Gemma, and Qwen without requiring them to independently deploy and manage each model.
The models run on Zettabyte’s NVIDIA H100, H200, B200, and B300 GPUs. Teams can call them through APIs without sourcing, installing, or tuning the models themselves. Organisations requiring greater control can instead rent raw GPU capacity on zCLOUD and run their own models.
NVIDIA H100 capacity is available immediately, starting at US$1.99 per GPU-hour depending on requirements and scale. zCLOUD targets 99.5% uptime, while reserved clusters are designed for deployment within six hours of confirmed requirements.
Zettabyte offers short-term GPU capacity, reserved clusters for sustained training and inference, and private cloud deployments for procurement, security, or regional requirements.
“AI computing will be everywhere, from hyperscale data centers to the edge. The supporting AI infrastructure and software must be compatible, resilient, and agile. zCLOUD was built to deliver Model-as-a-Service, bringing more than 40 hardware providers together and managing the underlying AI infrastructure through our zSUITE full stack,” said Dr. David Ku, Technology Leader at Zettabyte and former Corporate Vice President and CTO of AI at Microsoft.
The service creates a middle ground between proprietary AI APIs and self-hosted open models, giving organisations a managed way to adopt open AI without first building the underlying infrastructure.
