Private LLM or a Claude business plan: what does a small business actually need?
Updated
A private LLM for a small business usually means one of four things: a business plan for an AI assistant such as Claude, its Enterprise plan with tighter controls, a model like Claude run inside your own Azure, AWS or Google cloud, or an open model on servers you run. Most small businesses that ask for a private LLM need the first: a company account where the vendor does not train on your content, at $20 a person a month.
The other three cost more and do less out of the box. They are worth it when a client contract, a regulator or a protective order says your data has to stay in your own cloud.
The four options side by side
Prices are list prices as of November 2026 and change. Imhauser Technologies is independent and not affiliated with Anthropic, Microsoft, Amazon or Google.
| Option | Where your data is processed | What it costs | What you get |
|---|---|---|---|
| Claude Team plan | Anthropic's cloud, in a company account. Anthropic states it does not train on your content by default | $20 a person a month billed annually, or $25 monthly | The full Claude app: chat, Cowork, connectors to email and files, Word and Outlook add-ins, admin controls and single sign-on |
| Claude Enterprise plan | Anthropic's cloud, with audit logs, custom data retention, IP allowlisting and network controls | $20 a seat a month billed annually plus usage at API rates, 20-seat minimum | Everything in Team, plus the controls an IT or compliance team may require |
| Claude in your own cloud | Your Azure, AWS or Google Cloud subscription. On Microsoft Foundry's Hosted on Azure option, Microsoft documents that prompts and outputs never leave Azure | Usage billed by the cloud provider, plus building and running an application on top | The model only. Chat screens, document search and connections to your systems have to be built |
| Open model on your own servers | Your servers or your cloud GPUs. Nothing leaves your network | GPU servers, which can run to thousands of dollars a month, plus the people to run them | A model that is usually less capable than the leading commercial ones, and everything around it to build |
What "private" usually needs to mean
When owners ask for a private LLM, the worries behind it are usually three: that the AI company will train on their documents, that employees are pasting company work into personal accounts, and that a client or regulator will ask where the data went.
A business plan answers the first two. On Claude's Team and Enterprise plans, Anthropic states that it does not train its models on your content by default, and everyone signs in to the company's account, so the work stays with the business. Admins choose which connectors and features are allowed. Anthropic's security documents and certifications are at trust.anthropic.com.
The third worry is the one that can call for more. If a contract or regulation says data must stay in your own cloud tenant or in a specific region, a business plan in the vendor's cloud may not meet it, and one of the other options does.
Claude in your own Azure, AWS or Google cloud
Claude models are offered through Microsoft Foundry on Azure, Amazon Bedrock and Google Cloud Vertex AI. You call the model from your own cloud subscription and pay the cloud provider for usage. On Microsoft Foundry, the Hosted on Azure option processes prompts and outputs on Azure infrastructure; Anthropic is still the seller of the model and the data processor under its own terms, so read both Microsoft's and Anthropic's data terms before relying on it. Check the equivalent terms for Bedrock or Vertex AI if you are on AWS or Google.
What you get is a model, not an assistant. The Claude app's chat, Cowork, connectors and Office add-ins do not come with it. Someone has to build the application your people will use: sign-in, a place to upload or search documents, and the connections to your systems. That is a software project, which is why this option usually starts above $10,000, not at $20 a person.
For most small businesses that cost buys less than a business plan does: fewer features, more to maintain, and a project to fund before anyone gets value. Unless a contract or regulation leaves no other choice, start with a business plan set up properly, which is the work we do in Claude Implementation.
An open model on your own servers
Open-weight models can run entirely on hardware you control, with no outside provider seeing a word. The trade-offs are real: a server with GPUs large enough for a capable model is expensive to buy or rent, someone has to keep it patched and running, and the open models a small business can afford to run are usually behind the leading commercial models on writing, reasoning and long documents.
This option fits organizations with strict air-gap requirements and the technical staff to run it. For most small businesses it costs far more than a business plan and gives the team a weaker tool.
Which one fits
- Your concern is training on your data or staff using personal AI accounts
- You want the team using AI this month, at a predictable price
- No contract or regulation requires data to stay in your own cloud
- A client contract, protective order or regulator requires data to stay in your tenant or region
- You need AI inside software you already run or sell, not a chat app
- You have a budget for a software project and someone to own it afterwards
Start with the business plan
Even where one client or contract does require your own cloud, it rarely covers all of your work. Keep everyday work on the business plan and limit anything more restrictive to the work that truly needs it. That keeps the expensive part small.
Related questions
What is a private LLM?
A large language model used in a way that keeps your data away from other customers and out of the vendor's model training. It can mean a business plan on a vendor's cloud with no training on your content, a model run in your own cloud subscription, or an open model on your own servers.
How much does a private LLM cost a small business?
From $20 a person a month for a Claude Team plan, as of November 2026, to a software project usually above $10,000 for an application built on a model in your own cloud, plus usage. Running an open model on your own GPU servers adds hardware or rental costs that can run to thousands a month.
Is a private LLM API the same as a business plan?
No. An API, through Anthropic directly or through Azure, AWS or Google Cloud, gives developers the model to build with. A business plan gives your team a finished app with chat, connectors and admin controls. Most teams want the app; the API is for building something of your own.
Does Anthropic train on our data?
Anthropic states that it does not train its models on your content by default on its Team and Enterprise plans. Its security documents are at trust.anthropic.com.
Can Claude run inside our Azure tenant?
Claude models are available in Microsoft Foundry, and with the Hosted on Azure option Microsoft documents that prompts and outputs are processed on Azure infrastructure. Anthropic remains the seller and data processor. You still need an application on top for your team to use.
Next step
Tell us what you need. We reply within 24 hours.
Imhauser Technologies
McLean, VA
United States
Available on-site across the DC metro, Tysons, Reston, Arlington, and Washington, DC, as well as hybrid and remote worldwide.