Local vs Subscription
Running AI on your own machine has two costs: the machine, spread over however long you keep it, and the electricity for the hours you actually use it. Put your plan on one side and those two on the other, and the decision stops being about ideology and starts being a number you can check against your own bill.
Enter 0 if you already own it — that single change decides most of these.
Not hours switched on — hours under load.
The subscription
$20.00 /mo
Your machine, all in
$74.77 /mo
$66.67 hardware + $8.10 electricity
Cheaper over the life of the machine
The subscription
Spread over its life, the machine costs more per month than the plan does.
Once bought, the machine pays for itself in
202 months
$20.00 plan − $8.10 electricity = $11.90 saved every month, against $2,400.00 up front.
This compares money, not quality. The model you run at home is not the model behind the subscription — it is smaller, and on hard tasks it will show. Treat a win here as “worth trying”, not “settled”.
The number that decides it is the purchase price. Already own a capable machine and the running cost is a few pounds of electricity a month, so local wins almost immediately. Have to buy one and the plan usually wins on money alone — which means what you are really buying is privacy and independence, and it is worth knowing the price of those rather than pretending they are free.
How to read it: one input decides most of these, and it is the purchase price. If you already own a capable machine, running a model on it costs a few pounds of electricity a month and beats almost any subscription immediately — the money argument is over before it starts. If you would have to buy the machine, the plan usually wins on money alone, often by a wide margin, and the honest conclusion is that you are not buying savings. You are buying privacy and independence, and this tool tells you what they cost.
Where it stops working: this compares money, not capability. The model you run at home is smaller than the one behind the subscription, and on hard tasks that gap is real — a win here means “worth trying”, not “settled”. It also assumes you keep the machine for the life you entered and that the plan price holds, which is exactly the assumption the rug-pull record keeps disproving.
Sizing a machine rather than pricing one? The Local AI Hardware Sizer answers whether a model fits at all. Deciding for a company rather than a person? Build vs API runs the same question at fleet scale.
Embed this tool
Paste this snippet into any page to run the calculator there.