AI deployment
What server should I buy to run Dify or an AI assistant?
Short answer: Dify recommends at least 2 CPU cores and 4 GB RAM. The cheapest fit is Baidu AI Cloud 2 vCPU / 4 GB at ¥39/year (≈$5.5); for more headroom, Tencent Cloud Lighthouse 4 vCPU / 4 GB at ¥109 (≈$15) for the first year. That is plenty if you call model APIs (ChatGPT, DeepSeek, etc.) instead of running a large model on the server.
Prices checked 2026-10-22; the provider's checkout page is final.
Cheap and good enough
Running open-source LLMs locally (7B and up) needs a GPU or 16 GB+ RAM — these budget servers can't do that.
| Provider | Plan | Price | Notes | Link |
|---|---|---|---|---|
| Baidu AI Cloud e1 | 2 vCPU / 4 GB, 3 Mbps | ¥39/yr | Just meets Dify's minimum | Baidu AI Cloud |
| Tencent Cloud Lighthouse | 4 vCPU / 4 GB, 3 Mbps | ¥109 first year | Smoother; room for a few more services | Tencent Cloud |
| Vultr | 4 GB and up | Billed hourly | Delete the server and billing stops | Vultr |
| Oracle Free | Arm 2 OCPU / 12 GB | Free forever | Credit card needed; capacity can be scarce | Oracle |
FAQ
Can 2 vCPU / 2 GB run Dify?
Not recommended. Dify runs several Docker containers, and 2 GB of RAM tends to freeze or fail to install.
Can a server in mainland China call the OpenAI API?
Access to OpenAI's official API is restricted from mainland China servers; use Chinese model APIs such as DeepSeek, Qwen or ERNIE instead — or choose a server outside China.
Do I need ICP filing?
Not for personal use over an IP address. To serve a public website on your own domain from a mainland China server, yes.
What's the cheapest setup?
Baidu AI Cloud 2 vCPU / 4 GB at ¥39/year plus pay-as-you-go DeepSeek API.
Keep reading
Cheapest servers 2026 · Alibaba vs Tencent Cloud · Free-forever servers · All deals compared