Glossary
What is a cloud AI model?
In short
A cloud AI model is a neural network that runs on the servers of a provider. You use it through a website, an app or an API, and you pay for the volume of requests.
Infrastructure · cloud model
MMOOZZGG.com editorsChecked: 2 min read
For a child
There are two ways to hear a fairy tale. You can buy a book and keep it on a shelf at home. Or you can call your grandmother, and she will read a story over the phone from her big library. The second way is like a helper who lives far away, on someone else’s big computer. You ask a question over the internet, and it answers. Without a connection, this helper is silent.
In plain words
Most people use neural networks exactly this way: they open a chat in a browser or an app, type a question and get an answer. The model does not run on your phone but in the data center of the company that built it. You do not buy GPUs or update software, because the provider does that. You pay by subscription or per request. The downside: the text of your question goes to someone else’s server, and nothing works without the internet. For example, an accountant asks a chat to shorten a letter to a partner, and that letter goes to the service provider for a while.
For business
It is a fast start: connection in days, access to strong models and payment per token used. Costs grow with volume, so under a large, steady load your own server is sometimes cheaper. The main question for a company in Russia is where the data goes. If the requests contain personal data and the provider’s servers are abroad, this is a cross-border transfer. Under article 12 of 152-FZ, Russia’s personal data law, the company must notify the regulator (Roskomnadzor) in advance. The second risk: the provider changes prices, quotas and model versions, or shuts the service down. So before choosing, check the jurisdiction of the servers, the terms for using your data in training and the option to switch to another model. A map of AI solutions sorts tasks into what goes outside and what stays in-house.
For a specialist
A cloud model (cloud-hosted model) is a model that a provider hosts on its own infrastructure and offers as a service: through a web interface as SaaS or through an API billed per token. NIST SP 800–145 (Mell, Grance, 2011) defines cloud computing as a model for ubiquitous, convenient, on-demand network access to shared resources that are provisioned and released with minimal effort. Hosting options: the model developer’s public API, the same model at a large cloud provider in a chosen region, or a dedicated deployment with reserved throughput. Weights are usually closed, so the client controls only the prompt, the generation parameters and sometimes fine-tuning through the API. Engineering limits: request and token quotas, network latency, versioning and model retirement on the provider’s schedule, and the request retention policy. The opposite option: open weights on your own hardware.
Example
A real estate agency connects a cloud model to its CRM to write property descriptions. It removes owners’ addresses and phone numbers from each request, so only the area, the district and the features of the apartment go outside.
Related terms
Questions and answers
How is a cloud model different from a local one?
Can you send personal data to a cloud model?
Sources
Not sure whether your company needs this?
AI solutions map