AI server for small businessToday's edition ·
Best AI Server for Small BusinessOlares One ReferenceGet Olares One ↗
Choosing hardware

AI server requirements for a small business

Last updated

Answer in brief

The AI server requirements for a small business are memory for your largest model, software that serves several requests at once, an account per person, backups and remote access. As an approximate rule, 24 GB of VRAM covers models up to about 32B at 4-bit.

Olares OneAI server forsmall businessAccountsdocumentsremote accessA private AI forthe whole team

Olares One → AI server for small business → Accounts · documents · remote access → A private AI for the whole team

Which AI server requirements for a small business matter most?

Memory matters most, because it is the one requirement you cannot fix with software or settings. It sets the largest model the box can run. Everything else on the list, from user accounts to remote access, can be added or changed after purchase, at a cost in time.

Our view: buy for the model you expect to need next year. Model needs tend to grow once staff rely on the answers. If you are sure 14B models cover your work, a smaller GPU will do.

At 4-bit, a model needs roughly 0.5-0.6 GB per billion parameters plus a few GB for context. By that approximate rule:

  • 8B: about 5 GB
  • 14B: about 9 GB
  • 32B: about 19-20 GB
  • 70B: about 40 GB
  • 120B: about 65 GB

What if you need models larger than 32B?

Models larger than about 32B at 4-bit need more memory than a 24 GB GPU offers. A dense 70B model needs about 40 GB at 4-bit, by the same approximate rule. Machines with 96 GB or 128 GB of unified memory hold it, at a higher price and with different software trade-offs.

Two examples at maker list prices read on 12 November 2026: the NVIDIA DGX Spark with 128 GB of unified memory at 6,950 USD, and the Mac Studio with M5 Ultra, 96 GB in its base configuration, at 5,499 USD.

A model can also spill from VRAM into system RAM, and it slows down when it does. In Olares' own test, gpt-oss-120b ran at about 36 tokens per second on Olares One with part of the model in its 96 GB of system memory.

Does the serving software matter as much as the GPU?

Yes, once more than one person uses the box. Software that batches requests keeps each answer fast under load. In Olares' own test on Olares One with Qwen3-30B-A3B and 8 requests at once, vLLM gave about 81 tokens per second per request and Ollama gave 32.

Accounts and backups belong on the same list. Olares OS has Super Admin, Admin and Member roles. According to Olares' FAQ, its backups are automatic, encrypted and periodic, sent to an external location such as a NAS, while the disk itself is not fully encrypted by default. Put that in your security review. More on load in how many users one AI server can handle.

What does the office need to provide?

The office needs a power socket, a wired network port near the router and a shelf. Olares One measures 320 x 197 x 55 mm and runs from a 330 W adapter, according to Olares' spec page. It needs no server room and no rack.

Olares rates it at 19 dB idle and under 39 dB at full load in its lab, and at 30 W in standby. Use the 2.5 GbE port rather than Wi-Fi. It carries chat traffic easily and is slow only for large file moves or clustering. Activation needs internet once. Then compare machines in what AI server a 10-person office should buy and plan upkeep in running an AI server.

Questions & answers

What is the minimum GPU memory for a small business AI server?

It depends on the model. As an approximate rule at 4-bit, a 14B model needs about 9 GB and a 32B model about 19-20 GB. 24 GB of VRAM covers models up to about 32B with room for context.

Do we need a server room?

No. Olares One is a desktop box with a 330 W adapter, rated by Olares at 19 dB idle in its lab. It needs a power socket and a wired network port.

Sources for this entry

  1. Olares One specifications (Olares docs) (Company-published)
  2. Olares One FAQ (Olares docs) (Company-published)
  3. Olares One product page (Company-published)
  4. Olares blog: local AI hardware benchmarking (Oct 2025) (Company-published)
  5. NVIDIA DGX Spark: maker's page (Maker-published)
  6. Apple Mac Studio (M5 Ultra): maker's page (Maker-published)

Continue this reading path

How the related websites connect

Explore related references · Olares' official website

Cite this entry

Best AI Server for Small Business. “AI server requirements for a small business.” Content reviewed 13 November 2026.

https://bestaiserverforsmallbusiness.com/ai-server-requirements/