Most private AI conversations assume a head office with a server room. Plenty of Nigerian organisations do not work that way. They have branch offices with no IT staff, project teams on construction or energy sites, NGOs running programmes in several states, audit and consulting teams that work from client premises for weeks, and event teams that set up and tear down in a day. They still handle sensitive data, and they still want an AI assistant that does not send that data abroad.
The portable AI server is our answer: a compact, private AI server built into a rugged transport case with its own UPS, which plugs into whatever network is available and serves a ChatGPT-style assistant to everyone on it. It is built to order from our AI Series, not a fixed-price off-the-shelf product. This article explains what it is, who it suits, and the honest trade-offs. For the wider context, see why Nigerian companies need private AI.
What is inside the case
A portable AI server is the same private AI stack we install in offices, rearranged to travel.
- Compute · A compact GPU server built around the AI Research or AI Professional core: a single RTX 4070 Ti Super or RTX 4090, a high-core-count CPU and generous memory.
- Case · A rack-style transport case with shock mounting, front and rear lids, and airflow planned for operation with the lids off.
- Power · A rack UPS in the same case to ride through outages and switchover from generator or inverter, with automatic safe shutdown if power is lost for too long.
- Networking · A small switch and, where useful, a Wi-Fi access point, so the case can create its own private network on a site with no infrastructure.
- Software · Open models such as Llama, Qwen, Mistral or Gemma, served by Ollama or vLLM, with Open WebUI as the chat interface and optional document collections for retrieval.
- Security · Encrypted drives, a locked case and an interface reachable only on the local network.
Who it is for
| Use case | Why portable | Typical setup |
|---|---|---|
| Branch offices without a server room | Arrives as one unit, plugs into the branch LAN, no build-out needed | Plugged into the branch switch on the office inverter |
| Construction, oil and gas, and engineering site offices | Moves with the project; works with unreliable or no internet | Own Wi-Fi network; UPS fed from site generator |
| Audit, consulting and legal teams on client premises | Client data is processed in your own locked case, not on a client's network or a cloud service | Own isolated network for the team |
| NGOs and development programmes | Beneficiary data stays on organisation hardware across field locations | Moved between state offices by road |
| Events, training and workshops | A private assistant for participants without exposing the organisation's data | Set up on the day, packed away after |
| Firms that want AI locked away | The whole system can physically go in a safe room overnight | Office LAN by day, locked storage by night |
The honest trade-offs
A portable server is a compromise, and we would rather you know where the compromises sit before you buy one.
Size limits capacity
A compact chassis that travels safely is realistically a single-GPU build. That means the portable server's capacity follows its computing core: roughly 5–15 light concurrent users of 7–14B models on an AI Research core, or roughly 15–50 staff on quantised 14–32B models on an AI Professional core. Company-wide, 70B-class workloads belong on a multi-GPU AI Lab in a proper server room, not in a case.
Heat is the real engineering problem
A GPU under sustained load produces a lot of heat, and a closed case on a hot site office floor is a poor place for it. We design the airflow for operation with the lids removed, specify cooling with Nigerian ambient temperatures in mind and advise on placement. Running it in direct sun or a sealed cupboard is the quickest way to shorten its life.
Weight and handling
A rack case with a server and a UPS is a two-person lift. Battery choice and case size are a balance between runtime and portability, so we start from how you will actually move it: a vehicle between sites, a courier, or a trolley between rooms.
Local models are not frontier models
As with any private AI, open models on a single GPU handle drafting, summarising, extraction and document Q&A well, but trail the leading cloud models on the hardest reasoning tasks. For field teams with poor connectivity, the fact that it works offline often outweighs that.
Pricing: built to order
We do not publish a single price for the portable AI server, because the case, UPS runtime, networking and chassis vary with the job. What we can give you is the reference point: the computing core follows our published AI Series tiers, and the portable build is quoted on top of that after a consultation.
| AI Series tier | Core specification | Price (inc. VAT) | Rough fit for private AI |
|---|---|---|---|
| AI Research | Core i7-14700K, 64GB DDR5, RTX 4070 Ti Super (16GB VRAM) | ₦8,800,000 | Pilot or small team: roughly 5–15 light concurrent users of 7–14B models |
| AI Professional | Core i9-14900K, 128GB DDR5, RTX 4090 (24GB VRAM) | ₦18,600,000 | Department scale: roughly 15–50 staff on 14–32B models, quantised |
| AI Lab | Threadripper or Xeon W, 256GB ECC, 2× RTX 4090 or RTX A6000, 1600W redundant PSU, server chassis | From ₦25,000,000 | Company-wide, 70B-class models; consultation only. Not suited to a portable case |
The AI Research tier at ₦8,800,000 and the AI Professional tier at ₦18,600,000 are the tower-format prices for the same computing class. A portable build changes the chassis, adds the transport case, UPS and networking, and is priced accordingly; ask us for a quote against your use case.
Setting it up on site
- Place it · On a stable surface in a shaded, ventilated spot, lids off, near a protected power source.
- Power it · Into the inverter, generator or mains circuit through its built-in UPS.
- Network it · Into the office switch with one cable, or switch on its own Wi-Fi network on a site with none.
- Use it · Staff open the internal address in a browser and log in with their accounts. Nothing else is installed on their machines.
- Pack it · Shut down from the interface or power button, close the lids, lock the case.
For how the same software stack works in a fixed office, read How to Give Every Employee a Private AI Assistant. For loading project documents so the team can query them, see local RAG for Nigerian businesses. If you are weighing several sites against one central server, our sizing guide and our earlier piece on compute infrastructure for small AI teams help frame the decision.
Why it matters for data
For field teams, the alternative to a portable server is usually staff uploading documents to public AI tools over mobile data. That is exactly the shadow AI data leak most organisations are trying to close. A portable server gives those teams the same sanctioned, private tool as head office, and keeps beneficiary, client or project data on hardware the organisation owns and can lock away.
Sephora Systems builds portable AI servers to order from our AI Series, from Abuja with nationwide delivery and support. Book a private AI consultation to describe where and how you will use it, and we will specify the build. You can also ask Kitan, our site assistant, or WhatsApp us on +234 707 096 6669.