MMOOZZGG.com

Who it is for

Local AI for small business: no server and no engineer

3 min read12 sources

Local AI for small business does not need a server. One Mac or a laptop with a lot of memory covers the tasks of a small team, and if your data may leave the company, a cloud service for a few tens of dollars a month is cheaper than any hardware. Below is what to buy, what it solves and when not to buy at all.

On this page · 4 sections

What to buy

One user: a MacBook Pro with 36–64 GB of memory; the 128 GB configuration cost $8,100 on 29 September 2026. A small team: a Mac Studio M4 Max with 128 GB for $11,000 on a desk, drawing up to 145 W in the base M4 Max configuration, according to Apple. Models with 8–32 billion parameters: open Russian models and open models from around the world. Software: LM Studio or Ollama. By our calculation, the full cost is about $300–$400 a month with no engineer, because an employee who can update software handles maintenance.

What it solves

Drafts of emails and business proposals from your templates, answers to routine customer questions from a knowledge base, sorting incoming requests into a spreadsheet, checking contracts against company details, and meeting minutes from transcripts. A bank IT head’s personal agent on a Mac works with email, meeting transcripts and messengers at about 70 tokens per second.

When the cloud is cheaper

If your data holds no customer personal data, code or finances, or the volume is under 60 million tokens a month, the cloud wins: a mid-range cloud model costs $92–$250 a month, a cheap one $12–$48, and a Russian cloud with a data processing agreement $47–$900. By our calculation, a Mac Studio pays off in 40 months against a $350 monthly bill and in 71 months against $250, and against a bill under $94 it never pays off. For a small business, the local option is about customer data and independence from a provider, not about savings.

How to start in a week

Collect 50 real requests. Install LM Studio on any Mac with 16 GB and an 8-billion-parameter model, and run them. If the quality suits you, buy the memory the task needs, using the calculator. If not, try a 32-billion model on a rented GPU from about $0.50 an hour. Our pilot: one agent in 14 days from $1,499, on your Mac or on rented GPUs in Russia, with a metric in the contract.

Questions and answers

Is an ordinary laptop with 16 GB enough?

For an 8-billion-parameter model in 4 bits, yes: 6.3 GB with an 8,000-token context. For 32 billion in 4 bits, about 23 GB, which means a Mac with 36 GB of memory or more.

Do you have to pay for the model?

The weights of open models download for free; check the commercial terms in the license of each model: one open Russian 32-billion model, for example, uses Apache 2.0, which allows commercial use. You pay for hardware and for implementation.

Sources and fact-checking

Checked by the editors on 29 September 2026.

  1. apple-store.net.ru, MacBook Pro 14 M5 Max 128 GB 2 TB, 29 September 2026 (apple-store.net.ru)
  2. store-apple.msk.ru, Mac Studio M4 Max 128 GB, 29 September 2026 (store-apple.msk.ru)
  3. Apple Support, Mac Studio power consumption (support.apple.com)
  4. iXBT, a Russian bank released an open 32B model under Apache 2.0, 18 July 2025 (ixbt.com)
  5. A Russian search company, open weights of its 8B model, 25 February 2025 (yandex.ru)
  6. Tproger, cloud GPUs for ML in 2026, 21 May 2026 (tproger.ru)
  7. Claude Platform, API pricing (checked 29 September 2026) (platform.claude.com)
  8. A large Russian bank’s developer portal, API rates for companies, 16 September 2026 (developers.sber.ru)
  9. GPU Index, RTX 4090 rental in Russia: 173 offers, from about $0.40 an hour (checked 30 September 2026) (gpuindex.ru)
  10. Digital Platforms, case no. 2447, October 2025: a large Russian bank (platforms.su)
  11. Digital Platforms, case no. 3855, 2 September 2026: a large Russian financial group (platforms.su)
  12. Digital Platforms, case no. 4067, 2 September 2026: Alfa-Bank, head of IT (platforms.su)

Found a mistake? Tell us

Publications

Mike Mozg

Agentic AI use cases at AT&T: nearly 500 processes now run on agents

AT&T no longer uses AI only to help staff work faster: it is rebuilding whole business processes around agents. SiliconANGLE reported on 14 November 2026 that nearly 500 agentic AI use cases at AT&T are already in production. Here is how the company rebuilds them, where a human still checks the agent, and what other large companies can take from it.

News4 min

Cases on this topic: 3

  • Market case · Russia

    A large Russian bank in-house LLMs solve 45% of requests; support costs estimated down at least 25% a year

    The bank put its own family of language models into its chatbot for LLM customer support: about 45% of customer requests are solved without an operator. In October 2025 it launched AI copilots for first-line operators, and more than 90% of support staff use them every day. The models run inside the bank’s network perimeter, so data does not leave the bank.

    Banks and finance

    source: platforms.su

    Read the case
  • Market case · Russia

    A large Russian financial group AI financial assistant on in-house language models, beta since 20 July 2026

    The group built an AI financial assistant for tasks linked to money and shopping, on its own family of large language models. To avoid made-up answers, the system relies on verified data sources, not only on what the model has learned. The beta was presented on 20 July 2026, and the full launch is planned for the autumn.

    Banks and finance

    2 September 2026 · source: platforms.su

    Read the case
  • Market case · Russia

    Alfa-Bank, IT head local agent on the Qwen 3.6–35B-A3B model: about 70 tokens per second on a Mac

    A local AI agent on Mac does autonomous work on a model with 35 billion parameters, about 3 billion of them active per token. It runs through LM Studio at about 70 tokens per second. The agent first ran in Docker on the Mac, then moved to a virtual machine, and the author calls that move a turning point for ease of use.

    Banks and finance

    2 September 2026 · source: platforms.su

    Read the case
Contact us