What ZB Cloud runs

#Hosting

Your site, your app and your data, live. The store, the app, the database and the files a business runs on, hosted on the cloud closest to its customers and backed up.

• Web, API, databases, file storage
• Autoscale to zero and back
• Backups on by default

#Model inference

Open models, answering at any volume. Inference means running a trained model to answer a request. ZB Cloud serves open models on whichever GPUs are free and best priced right now.

• Open models on H100 and H200
• Text, image and speech models
• More GPUs as requests grow

#Training and fine-tuning

Teach a model your trade. Fine-tuning trains an open model further on your own data, so it learns your language, your products or your field. A LoRA fine-tune adjusts a small part of the model; a full fine-tune retrains all of it.

• LoRA and full fine-tunes
• Multi-GPU runs on H100, H200 and B200
• Progress saved as it trains

#Batch jobs

Overnight work at the night's best price. A batch job is work with a deadline but no one waiting on it: reports, data processing, rendering, long builds. It runs on spare capacity, which clouds sell for less.

• Spot and decentralized capacity for overnight runs
• Finishes by the deadline you set
• Continues from its last save if moved

#GPU scheduling

One fair queue for every GPU. When many jobs want the same GPUs, the scheduler decides which runs next. Urgent work goes first, and every AI team gets its fair share.

• Fair-share GPU queue across AI teams
• Priorities and quotas per project
• NVIDIA and AMD in one pool

#Build, release and monitoring

Every change built, checked and watched. Each change your AI team makes is built, tested and released the same way, then watched once it is live.

• Every release checked, rollback ready
• Logs, metrics and alerts
• Cost per job in the record