Sovereign ATX · Private AI for Alpha School

School-owned infrastructure. Frontier AI models. Zero cloud dependency. Managed by one of your own students.

Schedule a Call Full Proposal →
$10K
Cloud AI spend per student/year
Unlimited
Queries — flat monthly rate
$0
Data Leaves Campus
116×
Cheaper per token vs cloud
What Alpha School currently spends on cloud AI
$0
per year across 40 students at $10K/student
↓
Unlimited
queries with Sovereign — same budget, no token rationing, data stays on campus
The Cluster

4 Mac Studios. 2TB Unified Memory.
Running Frontier Models On-Campus.

Every machine connected. Every query answered locally. Nothing leaves the building.

SOVEREIGN AI CLUSTER UBIQUITI USW-Pro-24-PoE Mac Studio M5 Ultra · 512GB exo node 1 Mac Studio M5 Ultra · 512GB exo node 2 Mac Studio M5 Ultra · 512GB exo node 3 Mac Studio M5 Ultra · 512GB exo node 4 2TB Pooled via exo STUDENT WORKSTATIONS — 15× MAC MINI Mac Mini Mac Mini Mac Mini Mac Mini Mac Mini Mac Mini Mac Mini Mac Mini +7 more
4× Mac Studio M5 Ultra 2TB unified memory pooled exo distributed inference 15× Mac Mini workstations Ubiquiti enterprise networking UPS power protection
Savings Calculator

How Much Does Alpha School Save?

Drag to set your current monthly cloud spend.

Current monthly cloud AI cost $11,500/mo
Year 1
+$31,000
saved vs. cloud
Year 2
+$96,000
cumulative saved
Year 3
+$223,000
cumulative saved

Based on $65K upfront + $3,500/mo maintenance. Cloud midpoint used as baseline. Hardware retains resale value.

From Today to Go-Live

4–5 Weeks. Turnkey. Done.

Kavin handles every step personally. The school receives a ready-to-use AI cluster.

Today
Decision & Agreement
Service agreement signed. Hardware ordered same day. 10–14 day lead time for Mac Studios.
Week 2
Network Deployment
Ubiquiti router and switch installed. VLANs configured. Rack location prepared. Cable runs complete.
Week 3
Hardware Arrives & Racks
Kavin personally racks and cables 4 Mac Studios + 15 Mac Minis. UPS connected. All hardware powered up and named.
Week 4
Software & Models Go Live
exo framework configured. All frontier models loaded and tested. Open WebUI deployed. Student accounts provisioned.
Week 5
Training & Handoff
Live training session with staff and students. Full runbook delivered. Remote monitoring active. Cluster live.
Month 4–6
🏆 Hardware Pays for Itself
Break-even vs. cloud spend. From here on: $3,500/month instead of $8K–$15K. Compounding savings every month after.
Frontier Models. Local.

No Cloud. No Limits. No Excuses.

Frontier-class models running entirely on-campus. Students get unlimited AI — no token rationing, no data leaving the building.

Qwen 3.5-397B
397 billion parameters
World-class coding, math, reasoning and deep research. Frontier-class on MMLU, MATH, and HumanEval benchmarks.
Best for Code + Research
Kimi K2.5 Thinking
1T MoE · Extended reasoning
Deep reasoning with extended chain-of-thought. Scores 88.4% on AIME 2025 — outperforming frontier reasoning models on math and logic.
⚡ Top Reasoning Model
Llama 4 Maverick
400B MoE · Multimodal
Meta's frontier multimodal model. Images, documents, and text in one model. 1M token context. Fastest large model for classroom workloads.
Best for Multimodal + Speed
DeepSeek R1
671B MoE · Reasoning
Academic writing, scientific analysis, structured research. 79.8% on AIME 2024. Exceptional for STEM coursework and long-form analysis.
Best for STEM + Writing
Live — All 4 Models Running Simultaneously
Each row is a simultaneous inference request on a different model
Benchmark Spotlight
Kimi K2.5 Thinking — AIME 2025
88.4%
Kimi K2.5 Thinking
vs
86.7%
Claude Opus 4.6
AIME (American Invitational Mathematics Examination) is the industry standard for measuring frontier-level mathematical reasoning. Kimi K2.5 Thinking runs entirely on the Sovereign cluster — no cloud, no API, no token cost.
Infinite Scale

Dozens of Models. Writing Your Curriculum.
Right Now.

When 2TB of memory is available, you don't run one model — you run a fleet. Small 8B models spin up instantly to generate practice questions, check answers, adapt difficulty, and explore new topics simultaneously.

0
Active Instances
0
Questions Generated
0
Subjects Covered
The Human Behind It
"I'm 16. I'm your student.
I've already deployed this.
Let me build it here."

Kavin Lingham is an Alpha School student who used his school time to build Sovereign ATX — a private AI infrastructure company. He personally installs every system, manages all deployments remotely via Tailscale, and writes the service agreements himself. His first client, Black Sheep Coffee, went live in March 2026. Alpha School is next.

16 years old · Alpha School student
Black Sheep Coffee · $2,500/mo paying client
Site walk · March 25, 2026
Manages remotely via Tailscale
Ready to Move?

Alpha School's AI.
Private. Sovereign. Yours.

30-minute call to review specs. Site walk on campus. Deploy in 4–5 weeks. Interim setup active immediately, permanent cluster built when M5 Ultra ships.

Schedule a Call Full Proposal →

Kavin Lingham  ·  kavin@sovereignatx.ai  ·  +1 (858) 336-3468  ·  sovereignatx.ai