Blog · 17 September 2026 · 6 min read

How much does it cost to install a local AI in your company in 2026

A local AI doesn't have per-user fees like cloud services, but it does have other costs: the hardware it runs on and the work of installing it, connecting it to your data and training your people. Let's look at them one by one.

In short: the cost of a local AI is made up of three items. 1) Hardware, bought once and owned by the company: in 2026 a Mac Studio starts at €3,049 (M5 Max, 36 GB) and at €6,699 with the M5 Ultra (96 GB); a GPU server depends on how many cards and how much memory you need. 2) Installation and training: with Dabryx, consultancy starts at €100 per hour. 3) Maintenance: billed by the hour or through a support contract. There are no per-user subscriptions and no charges for each question you ask the AI.

Free PDF guide · 9 pages · updated September 2026

Want an AI that's all yours, inside your company? We'll show you how.

  • What hardware you really need (Mac or server) and what it costs
  • The best models for Italian and how to install them step by step
  • Security, GDPR and a 20-point checklist
Download the free guide →

You'll receive it by email within seconds. No spam.

1. Hardware: the cost that varies most

Artificial intelligence runs on memory: the larger the model, the more memory it needs. As a rough guide, a model with around 30 billion parameters compressed to 4 bits needs about 20 GB, and one with 70 billion over 40 GB (source: technical guides on Ollama, see below). Almost the entire hardware cost depends on this.

ConfigurationStarting price (Italy, Sept. 2026)Suited to
Mac Studio M5 Max, 36 GB3.049 €Small and mid-sized models, few users, automations
Mac Studio M5 Ultra, 96 GB6.699 €Large models, more users, reading many documents
GPU serverOn quotationMany simultaneous users, heavy workloads, data centre

Mac prices are the starting list prices announced at launch: they go up if you add more memory and storage. We don't give figures for GPU servers, because in 2026 memory and graphics card prices are very volatile; we quote them at the time.

You often don't need to buy everything straight away: you can start with a configuration suited to the first use cases and scale up once you've measured real usage.

2. Installation, data integration and training

The second item is the work. In a typical project it includes:

  • Analysis: which tasks to speed up, which documents to use, how many people will use the AI.
  • Installation: models (for example with Ollama), a chat interface with personal accounts (for example Open WebUI), backups and secure remote access over a private network.
  • Data integration: manuals, procedures, contracts and shared folders, with permissions for who can see what.
  • Automations: workflows that use AI to read emails or PDFs and pass the data on to your business software.
  • Staff training, which also helps with the AI Act (see below).

Dabryx works on a consultancy basis, from €100 per hour (VAT not applicable – Italian flat-rate tax scheme). After the initial analysis we prepare a quote with the estimated hours for each phase, so the total cost is clear before we start.

3. Maintenance and updates

Open models improve quickly: updating them from time to time means better answers on the same hardware. Maintenance covers software and model updates, backup checks and intervention if problems arise.

It can be handled by the hour, when needed, or with a support contract with dedicated tickets and responses within 2 hours on working days.

4. How to compare it with a cloud AI service

The right comparison isn't only about money. With a cloud service you usually pay a per-user subscription or a usage-based fee, and data is processed on the provider's servers. With local AI you pay for the hardware once plus the installation work, and the data stays in the company.

To run the numbers for your own company, start from three figures: how many people will use the AI, how often and with what data. If the data is confidential (contracts, customer data, projects), the value of local AI goes beyond a simple cost comparison.

5. Training and the AI Act: a cost to budget for

Article 4 of the AI Act, applicable since 2 February 2025, requires those who use artificial intelligence systems to take measures to support the AI literacy of their staff. With Regulation (EU) 2026/1744 (Digital Omnibus), in force since 27 July 2026, it has become an obligation of means: you don't have to guarantee a specific level, but you must take effective, proportionate measures and be able to demonstrate them. Documented training at the time of installation goes in this direction.

Frequently asked questions

How much does a local AI cost in total for a small business?

It depends mainly on the hardware and on how much work is needed to connect your data. The easiest fixed cost to estimate is the hardware: a Mac Studio starts at €3,049 (M5 Max, 36 GB). On top of that come the installation and training hours, which Dabryx quotes after an initial analysis, from €100 per hour.

Are there monthly costs?

There are no per-user fees and no charges for each request to the AI. What remains is the hardware's electricity consumption and, if you wish, a support contract for updates and assistance.

Do you need to buy a GPU server?

Not always. For offices and small to medium teams, a Mac with Apple Silicon and plenty of memory is often enough. A GPU server makes sense with many simultaneous users or heavy workloads.

Want an estimate for your company?

Tell us how many people will use the AI and what for: we'll tell you which configuration you need.

Response times

Within 2 hours on business days (9:00–18:00) for clients with a support contract.

AI consulting

From €100 per hour, on your company's servers or Macs.

Where we work

All over Italy, remotely and on site.