Enthusiast Infra Package
Your own AI lab. Owned outright. 2× NVIDIA DGX Spark running GLM 4.7 around the clock, 60 TB local NAS, and a cluster of 10 Raspberry Pi 5 nodes handling scraping and agentic workflows.
What's in the rack
Five hardware categories, pre-integrated, pre-configured, rack-mounted, and ready to run.
2× NVIDIA DGX Spark
Dual AI compute units
- GB10 Grace Blackwell Superchip
- 128 GB unified LPDDR5X per unit
- 1 PFLOPS AI (FP4) per unit
- Pre-configured: GLM 4.7 + Llama 3.3 70B
- 400 W TDP peak, 65 W idle
Local NAS (60 TB+)
Persistent on-premises storage
- Synology DS1823xs+ or TrueNAS Scale
- 8-bay, 8 TB NAS HDDs in ZFS RAID-Z2
- 10 GbE + 1 GbE built-in
- S3-compatible API for agent access
- Snapshot, versioning, auto-backup enabled
Orchestrator Node
Cluster brain and API gateway
- Mini-ITX: AMD Ryzen 9 / Intel i9
- 64 GB ECC DDR5 RAM
- 2 TB NVMe (OS) + 4 TB NVMe (data)
- Kubernetes master node
- n8n, Traefik reverse proxy, auth gateway
10× Raspberry Pi 5 Cluster
Distributed agentic worker nodes
- Raspberry Pi 5 (8 GB RAM), 128 GB NVMe per node
- Roles: scraper, agent, monitor, bridge
- Playwright / Puppeteer for JS-heavy scraping
- n8n workflow runners, custom Python agents
- PoE+ powered, auto-restart on failure
Rack & Networking
Cable-managed, ops-ready
- 12U desktop rack with cable management
- 10 GbE managed switch (Pi cluster uplink)
- UPS battery backup — 1,500 VA
- Segmented 1 GbE + 10 GbE internal network
- WireGuard VPN + firewall pre-configured
Live Stack View
Ideal for
Designed for operators who want AI leverage without recurring license fees.
Power Users
Run private LLMs around the clock, automate personal research pipelines, build knowledge graphs over your own files, and schedule multi-site scraping jobs — all without subscriptions or data leaving your premises.
Business Owners
Process confidential documents locally, automate competitor intelligence, run agentic customer-support pipelines and financial reporting workflows fully on-premises. One package replaces several SaaS subscriptions.
Developers and Researchers
Fine-tune open-weight models on proprietary datasets, schedule overnight batch experiments, and store all checkpoints on NAS. Full reproducibility with zero cloud cost spikes.
Digital Agencies
White-label the stack to run 10+ simultaneous scraping projects, generate content drafts via local LLM, and deliver faster at lower per-project cost — with full isolation between client data.
Why own it now
Cloud GPU costs, hardware prices, and AI API fees are all climbing. The window for a favorable own-versus-rent calculation is narrowing.
Hardware Price Index (2021 = 100)
Normalized price per unit relative to Q1 2021 — DDR5 $/GB vs GPU Compute $/TFLOP
3-Year Total Cost of Ownership
Equivalent compute capacity — own stack vs public cloud, USD thousands
Own stack: $16K hardware + $2K setup + $1.5K/yr maintenance · Cloud: equivalent GPU hours at current list pricing
Power Pack — Cost Breakdown
All-inclusive at $19,990
Recent hardware and pricing news
AWS GPU instances up 30% in 2026
Amazon Web Services raised per-hour pricing on G5 and P4de instances citing sustained AI infrastructure demand. A single always-on A10G instance now costs over $26,000/year.
DDR5 $/GB index +38% vs H1 2025
HBM3e supply constraints and accelerating AI server deployments drove DDR5 module prices to a four-year high, per DRAM eXchange. Analysts project further increases through Q3 2026.
+25% US tariffs on server hardware
The latest Section 301 tariff tranche covers network switches, DRAM modules, and certain compute boards imported from China. DIY self-hosted builds are now 15–25% more expensive off-the-shelf.
DGX Spark on selective allocation
Consumer demand for NVIDIA DGX Spark exceeded production capacity within weeks of launch. Units are on selective allocation with 12–16 week lead times. AlgoVectra secures stock for confirmed orders.
Procurement window: DGX Spark units are on selective allocation with 12–16 week lead times. AlgoVectra secures hardware stock for confirmed orders. Contact us early to lock in current pricing before the next tariff adjustment.
Pricing packages
Fixed-price packages. No recurring hardware fees, no per-token billing.
Starter Pack
Single DGX Spark build
- 1× NVIDIA DGX Spark (128 GB)
- 1× Orchestrator Node
- 40 TB NAS (ZFS RAID-Z2)
- 6× Raspberry Pi 5 worker nodes
- 8U rack + networking
- 30-day setup support
Power Pack
Full enthusiast configuration
- 2× NVIDIA DGX Spark (128 GB each)
- 1× Orchestrator Node (64 GB ECC DDR5)
- 60 TB NAS + S3 API
- 10× Raspberry Pi 5 (scraper + agent roles)
- 12U rack, 10 GbE switch, 1,500 VA UPS
- 60-day setup support + 6-month maintenance
- Pre-configured: GLM 4.7, Llama 3.3, n8n pipelines
Enterprise Custom
Tailored to your scale
- Custom multi-rack hardware configuration
- Model fine-tuning on proprietary data
- Full cluster orchestration and CI/CD
- 12-month SLA with on-site support
- Custom agent and workflow development
- Staff training and documentation
Technical Architecture
Inference Layer
The two DGX Spark units run separate models concurrently — one dedicated to GLM 4.7 for general-purpose reasoning and one to Llama 3.3 70B for code and structured output. A local load balancer routes requests across both, with failover to a quantized fallback model.
Orchestration & API Gateway
The orchestrator node runs Kubernetes (k3s lightweight) as cluster master, n8n for workflow automation, Traefik as reverse proxy, and a custom FastAPI gateway exposing LLM, scraping, and storage endpoints under a unified auth layer.
Scraping Cluster
Ten Pi 5 nodes run headless Chromium via Playwright, rotated across residential proxy pools. Jobs are queued by the orchestrator. Results land in MinIO on the NAS for downstream agent processing.
Storage & Observability
All model outputs, scraping results, and agent memory land on the ZFS RAID-Z2 NAS with 5-year retention. Prometheus and Grafana run on the orchestrator for cluster health, inference throughput, and job queue dashboards.
Looking for managed cloud or enterprise AI infrastructure?
Our Infrastructure Package covers Managed Kubernetes, CI/CD, BYOD on-prem, and edge fleet deployments alongside the Enthusiast stack.
Ready to own your AI infrastructure?
We handle hardware procurement, rack assembly, OS configuration, model deployment, and pipeline setup. You get a fully operational AI lab — without the cloud bill.