Budget
BudgetHome AI Server
Budget Home AI Server
$1,162
16 GB VRAM250W TDP~25dB8 components
Inside This Build
Components
| Category | Component | Price | Rationale | Buy |
|---|---|---|---|---|
| gpu |  NVIDIA GeForce RTX 4060 Ti 16GB | $449 | 16GB VRAM at only 160W TDP — the efficiency king for always-on servers. Runs 7-8B models at full speed and 14B models at Q4. At ~$15/month in electricity (24/7), this is the cheapest card to operate as a server GPU. The RTX 3060 12GB was considered but 4GB less VRAM limits you to 7B only. | Buy |
| cpu |  AMD Ryzen 5 7600 (non-X) | $179 | 65W TDP. 6 cores are more than enough for serving one model at a time to a household. The non-X variant runs 20W cooler than the 7600X at near-identical inference throughput. Low idle power (~25W at socket) matters when the box runs 24/7. | Buy |
| motherboard | GigabyteGigabyte B650M DS3H | $109 | Budget B650 micro-ATX with 2.5GbE LAN. Two M.2 slots for OS + models. PCIe 4.0 x16. No WiFi needed — this is a wired server. BIOS supports Wake-on-LAN for remote power management. | Buy |
| ram | 32GB DDR5-5200 (2x16GB) | $69 | 32GB is sufficient when the GPU handles inference. System RAM serves Ollama, Open WebUI, and any RAG indexing. DDR5-5200 is the cheapest DDR5 tier and bottleneck-free for this CPU. | Buy |
| storage |  500GB Kingston NV2 NVMe (OS) + 4TB Seagate IronWolf HDD (data) | $129 | Split OS/models on NVMe for fast boot and model loading. 4TB IronWolf HDD for NAS-grade data storage — designed for 24/7 operation with 180TB/year workload rating. Models live on NVMe, documents and media on HDD. | Buy |
| psu |  Corsair RM550x 550W 80+ Gold | $79 | 550W with 80+ Gold efficiency. Total system draw under load is ~250W, idle ~60W. Running at 45% load gives peak PSU efficiency. Fully modular — clean cabling in a server closet matters for airflow. | Buy |
| case | Fractal DesignFractal Design Node 304 | $99 | Compact Mini-ITX case with 6 HDD bays — expandable to true NAS later. Fits a 310mm GPU. Designed for quiet operation with included 140mm fans. Small footprint for shelf or closet placement. | Buy |
| cooler | NoctuaNoctua NH-L9a-AM5 | $49 | Low-profile AM5 cooler at just 37mm tall. Near-silent at 23 dB. The 65W 7600 never stresses this cooler, keeping fan speed minimal 24/7. | Buy |
| Total | $1,162 | |||
Shop Components
Search links — prices and availability vary by retailer.
Buy Used GPU
Prices and availability vary. Inspect hardware before purchasing.
What This Runs
AI models tested on this build's hardware.
| Model | Quant | Speed |
|---|---|---|
| Llama 3.1 8B Instruct | Q5_K_M | 40 tok/s |
| Qwen 2.5 7B Instruct | Q5_K_M | 38 tok/s |
| Phi-4 14B | Q4_K_M | 22 tok/s |
| Gemma 3 12B | Q4_K_M | 24 tok/s |
| Mistral 7B Instruct v0.3 | Q8_0 | 45 tok/s |
| nomic-embed-text v1.5 | FP16 | — |
| Whisper Large V3 | FP16 | — |
AI Builder Workflow: Home AI Server
Always-on local AI server for a household or small team. Runs Ollama + Open WebUI accessible from any device on the network. Serves chat, coding assistance, document Q&A, and transcription to multiple simultaneous users — with zero API costs and complete data privacy.
Tools in this workflow
OllamaOpen WebUIAnythingLLMWhisperLibreChat
Concurrent VRAM Usage
7 GB of 16 GB
9 GB headroom for additional workloads
API Cost Replacement
If you're paying ~$80/month for cloud API access, this build pays for itself in 10 months.
Based on a 3-person household using ChatGPT Plus ($20/mo each = $60/mo) plus occasional API calls for document processing (~$20/mo). Budget Home AI Server at $1,162 including electricity (~$15/mo for the budget tier at 250W average). Break-even includes electricity. After break-even, savings are $65+/month indefinitely. Privacy benefit: no family conversations, documents, or voice recordings leave your network.
Upgrade Path
Swap GPU to a used RTX 3090 24GB (~$800) for 8GB more VRAM and 3x the memory bandwidth (936 GB/s vs 288 GB/s), enabling 32B models. TDP jumps from 160W to 350W — upgrade PSU to 750W. Add more IronWolf HDDs for expanded NAS storage.
Target Use Cases
ChatSelf-HostingEmbeddingsTranscription
Last updated: 2026-03-14.