Upgradeable to AI — Add 16GB RAM
303 refurbished laptops, desktops & mini PCs for running Ollama, LM Studio, Llama 3, Qwen, and Mistral locally — no cloud, no API key. Ships same day from Reno, NV.
Filter by AI capability
303 devices · Upgradeable
Tablet/2-in-1Microsoft
MS Surface Pro 4 1724 12.3"2736x1824 TOUCH m3-6Y30 900MHz 4GB 128GB SSD Win10Pro
Microsoft
MS Surface Book 2 13.5"3000x2000 TOUCH i5-8350U 1.7GHz 8GB 256GB W11P Great Batt
Tablet/2-in-1
Microsoft
Microsoft Surface Book T S 13.5" i7-6600U 8GB GeForce 940M 256GB SSD Silver W10h
MacBook
Tablet/2-in-1Tablet/2-in-1
Apple
Apple MacBook 12"EARLY-2016 A1534 M3-6Y30 1.1Gh 8GB 256GB MLHA2LL/A SSD Monterey
How to run AI locally
Pick a device
Filter by tier, brand, or type.
Install Ollama
Free. ollama.com. 2 minutes.
Pull a model
ollama run llama3
Chat privately
Nothing leaves your machine.
Common questions
What laptop do I need to run Llama 3.1 locally?
To run Llama 3.1 8B comfortably, you need at least 16GB RAM. For Llama 3.1 70B, you need 32GB+ RAM or a GPU with 16GB+ VRAM. Apple Silicon MacBooks (M1/M2/M3/M4) with 16GB unified memory run 8B models at 20–40 tokens/sec with no setup beyond installing Ollama.
Can I run local AI on a laptop without a GPU?
Yes. CPU-only inference works well for 7B–8B models on any laptop with 16GB+ RAM. Expect 5–15 tokens/sec, which is readable in real time. For faster speeds, a laptop with a discrete NVIDIA GPU (6GB+ VRAM) or Apple Silicon is recommended.
How much RAM do I need to run AI models locally?
8GB RAM is the minimum — runs Phi-4 Mini and Llama 3.2 3B only. 16GB RAM runs Llama 3.1 8B and Qwen3 7B at usable speeds. 32GB runs Qwen2.5 14B and 32B Q4 models. 64GB runs 70B models at full quality.
Do desktops and mini PCs work for local AI?
Yes. A Dell OptiPlex or Intel NUC with 16GB RAM runs 7B models well. Gaming desktops with RTX GPUs (8GB+ VRAM) handle 13B–32B models at 20–50 tokens/sec.
Is Apple Silicon (M1/M2/M3/M4) good for running local AI?
Apple Silicon is excellent for local AI. Unified memory means the GPU and CPU share the same RAM pool, so a 32GB M2 Pro can offload a full 32B model to the GPU. Metal acceleration is automatic via Ollama. M1 16GB runs 8B models at 20–30 tokens/sec. M2/M3/M4 32GB runs 32B Q4 models at 15–25 tokens/sec.
How do I start after buying a device?
Install Ollama from ollama.com (free, 2 minutes). Open Terminal and run: ollama run llama3.1:8b — it downloads the model and starts automatically. For a ChatGPT-style interface, install LM Studio from lmstudio.ai. No cloud account, no subscription, no data leaving your machine.