A CLEARER START

Cloud basics, in plain English.

Understand the few ideas that change your choice. Start with a question, or look up a term when you meet it.

Quick definitions.

What the term means, and when it matters.

GPU
A processor that can perform many calculations in parallel. Many AI workloads use one.When choosing compute
VRAM
The GPU’s working memory. Your model and active workload need enough usable space.When checking model fit
Cloud instance
A rented computing environment with a defined amount of compute, memory and other resources.When comparing whole machines
Inference
Running a trained model to produce an answer or other output.When choosing a serving route
Model API
An interface your app calls to use a model hosted by a service.When you want less infrastructure work
Fine-tuning
Further training a model on selected data to adapt its behaviour or task performance.When adapting a model
Serverless
The provider manages the servers and capacity behind your workload. Billing and scaling rules vary.When comparing scaling and idle costs
Dedicated hosting
Capacity allocated for your workload rather than a shared model-serving pool. Check exactly what is dedicated.When comparing isolation and control
On-demand pricing
Pay for capacity as you use or provision it, without a reserved-term commitment. Check billing minimums.When estimating flexible usage
Spot pricing
Discounted spare capacity that can be interrupted. Your workload needs a recovery plan.When a job can tolerate interruptions
Data egress
Data leaving a service or network boundary. A provider may charge for it.When checking transfer costs

Where the pieces fit.

A GPU does calculations. VRAM holds working data.

More VRAM can help a model fit. It does not, by itself, prove that one setup is faster or suitable for your workload.

Understand GPU memory →

You still decide what to manage.

A hosted model, a managed endpoint and your own runtime put different responsibilities on you.

Compare the three routes →
Official sources behind these definitions

OFFICIAL DOCUMENTATION / FURTHER READING

Hugging Face: endpoint configuration (opens in a new tab)AWS: Spot Instances and interruptions (opens in a new tab)Hugging Face: quantization overview (opens in a new tab)

Editorial references checked 11 October 2026. Product features and terms can change; confirm the current documentation before purchasing.

YOUR NEXT STEP

Put the terms to work.

Tell us what you want to run. We’ll bring in the technical details when they change the choice.

Try the Cloud Finder