# AI PC builder: pick parts for local LLMs | What Can I Run AI

> Build a PC for local AI part by part, PCPartPicker style: compatibility and power checks, VRAM, bandwidth, estimated tokens/sec, context fit, total price and a shareable build link.

Source: https://whatcanirunai.vercel.app/planner/

# AI PC Builder

Pick a GPU or a complete system, then the rest of the parts. We check compatibility and power, and estimate which models run and how fast.

Quick start

Budget / usedBalancedMaxClear build

Share link for this buildCopy linkCopy parts list[Compare](https://whatcanirunai.vercel.app/builds/compare/)

| Component | Selection | Specs | Price | Where | Actions |
| --------- | --------- | ----- | ----- | ----- | ------- |

Prices are dated snapshots: GPUs at list price (MSRP) unless marked used; CPUs, boards and SSDs from Newegg and PSUs, cases and RAM from Corsair on 2026-10-23 (new CPU platforms and RM650e priced 2026-11-05). We may earn a commission when you buy through these links, at no extra cost to you.

## What this build runs

Workload Quant 

Chat (Qwen3-14B)Coding agent (Qwen3-Coder-30B-A3B)70B (Llama 3.3 70B)

Every model on this build (656)

| Model | Q4\_K\_M | Q8\_0 | FP16 | Est. tok/s at Q4 |
| ----- | -------- | ----- | ---- | ---------------- |

Memory = weights + 15% + 1 GB at about 4k context; longer context adds the KV cache from each model's config (FP16). Speed = bandwidth ÷ weight size × an efficiency band calibrated on llama.cpp scoreboards; mixture-of-experts models get only an upper bound ("up to"). Multi-GPU adds memory, but speed is about one card's. [How we calculate](https://whatcanirunai.vercel.app/method/).

---
Agent guide: https://whatcanirunai.vercel.app/llms.txt · JSON API: https://whatcanirunai.vercel.app/developers/