← Back to AI for School

Executive Summary

We propose building a school-owned AI server that brings artificial intelligence directly into the classroom — no VPNs, no paid subscriptions, no data leaving the school network. Every teacher and student gets instant access to lesson planning assistants, text simplification tools, vocabulary builders, reading assessments, and educational games. All running on hardware the school owns and controls.

Why This Matters for Your School

The Problem Teachers Face

• Blocked access — AI tools like ChatGPT, Gemini, and Claude are often blocked by school firewalls
• Data privacy — Teachers cannot safely enter student information, assessment data, or internal curriculum into public AI services
• Cost barriers — Premium AI subscriptions are expensive; teachers don't know where to start
• Digital divide — Not all staff have equal technical skills or confidence with AI

The Solution: A School-Owned AI Server

• Internal access only — Works on the school LAN. No internet required after setup
• Data never leaves — All processing happens inside the school. Fully compliant with GDPR, FERPA, or your local data protection laws
• Open-source — Zero licensing fees. Uses free tools: Ubuntu Linux, Ollama, and open-weight AI models
• Educational — Designed specifically for teaching, not general chat

Proof of Concept

This proposal is based on real-world testing. A proof-of-concept machine was built using personal funds and has been running successfully, serving educational content and AI tools to the school network. This demonstrates that the technology works, the concept is viable, and the benefits are tangible before any institutional budget is committed.

Recommended Server — Two Budget Tiers

Both configurations use professional server-grade hardware sourced from used enterprise markets. These are significantly more cost-effective than consumer PCs while delivering enterprise reliability.

Standard Configuration (~¥12,050 / ~$1,700 USD)

2U enterprise server. Sourced from used/server refurb markets. Tested before shipping. Typical 1-year seller warranty.

Component	Specification	Est. Price
Chassis	GIGABYTE G292-Z20 2U Rackmount Server (used)	¥4,700
Motherboard	MZ22-G20 (SP3 socket, 8x DDR4, 10x PCIe x16)
CPU	AMD EPYC 7532 — 32 cores / 64 threads, 2.4GHz
RAM	256GB DDR4 ECC RDIMM (8x 32GB, used) — expandable to 512GB	¥2,400
GPU	2x NVIDIA Tesla A2 (16GB GDDR6 each, 60W, datacenter AI inference) — expandable to 4x	¥4,000
Storage (System)	1TB NVMe SSD (used enterprise)	¥400
Storage (Data)	4TB HDD (used enterprise)	¥350
Power	2200W Dual Redundant PSU (hot-swap, included)	Included
Cooling	4x Server-grade fans (included)	Included
Network	Dual 10GbE SFP+ ports (onboard) + IPMI remote management	Included
Rails / Rack Kit	Sliding rail kit for server cabinet (recommended)	¥200
OS & Software	Ubuntu 24.04 LTS (free, open-source)	¥0
Total	32 cores · 256GB RAM · 32GB VRAM · Rackmount · 10GbE · Redundant PSU	~¥12,050

Why a Rackmount Server?

For a school environment with a server room, the rackmount server is the best long-term investment:

• Enterprise form factor — 2U rackmount sits in the server room, not on a desk. Professional, secure, out of the way.
• Redundant power — dual 2200W hot-swap PSUs. If one fails, the other keeps running. Zero downtime for students.
• 10GbE networking — dual SFP+ ports onboard. Future-proof when the school upgrades network infrastructure. 10x faster than 1GbE for large AI model transfers.
• Maximum expansion — 10 PCIe slots (can add 2–4 more Tesla A2s later), 8 hot-swap drive bays (add storage without opening the case).
• Enterprise reliability — built for 24/7 datacenter operation.
• 32 cores is enough — AI inference is GPU-bound, not CPU-bound. The Tesla A2s will be the bottleneck long before 32 EPYC cores max out.

When a Rackmount Isn't Possible

If your school has no server room or rack cabinet, a full-tower dual-CPU workstation is an alternative. Expect to pay roughly ¥13,650 (~$1,900 USD) for 64 cores, 256GB RAM, and 2x Tesla A2 — but you lose rackmount, redundant PSU, and 10GbE networking. It sits on the floor or a desk like a very large PC.

What the Server Will Host

For Teachers — Professional Development

• AI Lesson Plan Assistant — Generate differentiated activities, rubrics, and assessments
• Text Simplifier — Adapt reading passages for ESL students by level
• Discussion Starter Generator — Create engaging prompts for any topic
• Vocabulary Builder — Auto-generate word lists with definitions and example sentences
• Reading Assessment Creator — Build comprehension questions with answer keys

For Students — Learning Tools

• Educational Games — Grammar, vocabulary, and critical thinking games
• Creative Project Generator — AI-assisted story writing and project ideas
• AI Tutor — Students can ask questions and get explanations without leaving the school network
• Listening Practice — AI-generated listening exercises with transcripts

For the School — Infrastructure

• Internal AI Resource Website — A professional portal for all AI tools
• Content Management — IT controls what content is available; everything is curated for education
• No External Dependencies — Works even if the school's internet connection is slow or filtered

Benefits to Your School

What IT Needs to Provide

To make this operational, the school's IT team or network administrator needs to allocate:

1. Static IP Address — Reserve one IP on the school LAN so the server is always reachable at the same address (e.g., 10.x.x.x or 192.168.x.x)
2. DNS Entry — Map a subdomain like ai.yourschool.edu or ai.yourschool.cn to the static IP so teachers and students access it by name, not by numbers
3. Firewall Rule — Allow internal LAN traffic to the server (typically ports 80 and 443). No external internet exposure required
4. Rack Space — One 2U slot in the server cabinet, plus power outlets. If no server room is available, a full-tower alternative can sit in an office or classroom

Next Steps

1. Review this proposal — Share with your Principal and IT team
2. Adapt the budget — Adjust hardware choices and pricing to your local market
3. Build a demo — Even a modest desktop PC can run a proof-of-concept to prove the concept works in your environment
4. Pilot with teachers — Start with a small group of early adopters
5. Scale based on feedback — Add more GPUs, RAM, or features as usage grows