Many compact AI PCs use soldered unified memory, so you cannot add RAM later and the SSD becomes your main upgrade path. Check the M.2 size (2280 or 2230), PCIe generation, number of M.2 slots and whether the system has SODIMM slots. Then upgrade to 2–4TB of internal NVMe for models, and add an external SSD for backups.
Two Kinds of AI Mini PCs
| Unified-Memory AI Mini PCs | Mini PCs with SODIMM Slots | |
|---|---|---|
| Memory | Soldered LPDDR5X shared by CPU and GPU, often 64–128GB | Standard DDR5 SODIMM, upgradeable |
| Can you add RAM? | No. Choose capacity at purchase. | Yes, usually two modules |
| Storage | One or two M.2 NVMe slots, often 1–2TB from the factory | One or two M.2 NVMe slots |
| Main upgrade path | SSD capacity and external storage | RAM and SSD |
Unified-memory systems are popular for local AI because the GPU can use most of the memory pool, often around 96GB of a 128GB system, to load large models. But because that memory is fixed, many owners fill the factory SSD with models within weeks.
Step 1: Check Your M.2 Slots
- Form factor: most mini PCs use M.2 2280. Some ultra-compact systems and handhelds use the shorter M.2 2230.
- PCIe generation: Gen4 x4 is common. A Gen5 drive works in a Gen4 slot, but at Gen4 speeds.
- Number of slots: if a second slot is free, add a dedicated model drive instead of replacing the system drive.
- Cooling and clearance: check whether the slot has a thermal pad, and whether a heatsink-equipped SSD will physically fit.
Step 2: Pick the Right Capacity
| How You Use It | Suggested Model Drive |
|---|---|
| A few chat and coding models | 2 TB |
| Comparing many models, including 65–80GB MoE models | 4 TB |
| Fine-tuning, datasets, diffusion models and RAG archives | 4 TB internal + external SSD |
Step 3: Choose Your Silicon Power SSD
UD90 NVMe SSD (2TB / 4TB)
The most capacity per dollar for a dedicated model drive.
US75 NVMe SSD
Faster model loads and switching for daily use.
US85 NVMe SSD
For mini PCs with a Gen5 M.2 slot.
Silicon Power M.2 2230 2TB
For ultra-compact systems and handhelds that use the short M.2 2230 size.
Step 4: Add External Storage for Backups and Portability
Models are easy to download again, but your fine-tunes, LoRA adapters, prompt libraries and RAG document sets are not. Keep a copy on an external drive.
DS72 2TB Portable SSD
Both USB-C and USB-A connectors, so it works with any mini PC.
PD60 M.2 SSD Enclosure
Turn the SSD you replaced into an external backup drive.
If Your Mini PC Has SODIMM Slots
Then memory is your first upgrade. Two DDR5-5600 32GB SODIMMs (64GB total), or the DDR5-5600 64GB (2×32GB) kit, let you run 30B–35B MoE models and long-context coding tools. The same rules as laptops apply; see Laptop Local AI: SODIMM Upgrade Guide.
- Unified memory cannot be upgraded, so storage is the main upgrade path for many AI mini PCs.
- Check M.2 size, PCIe generation, slot count and clearance before buying.
- 2TB suits a few models; 4TB suits large MoE libraries.
- Back up fine-tunes and RAG data to an external SSD.
Frequently Asked Questions
Can I add more RAM to a unified-memory AI mini PC?
No. Unified memory is soldered to the board, so choose the capacity you need when you buy. Storage is the part you can still upgrade.
Can I move my models to a new SSD without downloading them again?
Yes. Copy the model folder to the new drive and point your app to it. In Ollama, set the OLLAMA_MODELS environment variable. In LM Studio, change the models folder in its settings.
Will a faster SSD make my model answer faster?
Once a model is loaded into memory, generation speed depends on memory bandwidth, not the SSD. A faster SSD shortens load times and helps if you use SSD expert streaming.
Will a heatsink SSD fit in my mini PC?
Not always. Mini PCs have tight internal space. Check the clearance above the M.2 slot, and choose a drive without a heatsink if space is limited.