Does a local AI rig save money compared to the API?
A local AI rig does not save money against the DeepSeek API. Two DGX Spark machines cost $9,398 and the same output costs about $30 a month through an API, a 26-year break-even.
The main idea
If you want to buy a local AI rig to save money, I say don't. This is a strong opinion, and I hold it as of right now. If API prices go up, I might change my mind. I made a video before on whether you should buy a local AI rig. This time I made the video because I keep seeing posts on Twitter, LinkedIn, and Reddit that tell people to buy these rigs to save money.
In the video I run the numbers. DeepSeek V4 Flash through OpenRouter starts at 8 cents per million input tokens and 17 or 18 cents per million output tokens. Two DGX Spark machines cost $9,398 before tax. At about 41 tokens per second, running 24/7, they give you about 106 million output tokens per month. The same tokens through an API cost about $30 a month, even with the machine running nonstop. The break-even is 26 years, and that is with free electricity.
API is expensive only for frontier models from OpenAI and Anthropic. Open models like DeepSeek stay cheap through OpenRouter.
There are good reasons to buy a local rig. You can run uncensored models from Hugging Face. You own the models. You have no weekly usage limits, no service outages, and you can use AI on a plane with a 128 GB MacBook. Your questions are not profiled by a provider. Those are privacy and ownership reasons. Cost savings is not one of them.
Key points
- DeepSeek V4 Flash through OpenRouter starts at 8 cents per million input tokens and 17 to 18 cents per million output tokens.
- Two DGX Spark machines cost $9,398 before tax and give about 41 tokens per second.
- Running 24/7 gives about 106 million output tokens per month, which costs about $30 a month through an API.
- The break-even against the API is 26 years, even with free electricity.
- Good reasons to buy a local rig are uncensored models, ownership, no usage limits, offline use, and privacy.
Chapters
- 0:00Don't buy a local AI rig to save money
- 1:46The $9,398 math: 2x DGX Spark at 41 tokens/sec
- 3:28Why API stays cheap for open models
- 6:00The verdict: a privacy play
Save my free seat →