Explainer · 10
What does it really cost?
Cloud AI is cheap to start and bills you every month. A private machine is expensive to start and then mostly doesn't bill you at all. Which costs less depends on how many people use it and for how long — and that's easy to work out.
Two ways to pay
-
Cloud: rent
A subscription per person, per month — and for heavier or automated use, charges based on how much text is processed. Low to start, and it grows with every person you add.
-
Local: own
You buy the hardware once. After that, the running costs are electricity and the occasional upgrade. The fifth person using it costs the same as the first.
What a build costs
Builds start at $2,500 for a Desktop for one person, $5,000 for a Workstation for professional work, and $9,000 for a Server for a household or small team. Each is quoted individually after the consult — these are honest starting points, not teasers.
How to compare them honestly
Add up what everyone who'd use it would pay for cloud AI each month, and multiply by the years you'd keep a machine. Compare that with the build price. Then adjust for what a spreadsheet misses:
- Electricity. The machine works hard while it's answering and idles quietly the rest of the time. For most offices it's a small line item, but it isn't zero.
- Capability. The largest cloud models can do things a desk machine can't. If you'd still need a cloud subscription for some of your work, count it.
- Ownership. The hardware is yours. It can be upgraded, repurposed, or sold.
When cloud is cheaper
For one person using AI occasionally, cloud almost always wins on cost — by a lot. If that's you and your work isn't confidential, keep your subscription. We'll say the same on the call.
Local starts to make financial sense when several people use AI every day, or when the real reason is control rather than savings — in which case cost is the second question, not the first.