Purchases and Costs
Last updated
Local LLM ownership cost over one or two years
Add ownership costs, then subtract actual resale and savings
The total cost of ownership (TCO) of local AI hardware is its purchase price plus electricity and maintenance, less its estimated resale value at the end of the ownership period. If your goal is to reduce subscriptions, also include the bills you can actually eliminate. Start with the new expenditure and compare running the same work for one and two years.
If you already own a computer, adding a GPU1 and buying a complete new system start with different costs. Comparing hardware sticker prices alone hides that difference. Use keeping your existing PC as the baseline, record the new cards, parts and installation each option needs, then add running costs and subscriptions to see what you would pay to move which work.
List the money you would spend now
For a GPU upgrade, add the card price and any power supply, case or installation you would actually need. For a new Mac or prebuilt PC, use the complete purchase price. Include all required new parts and exclude components you will keep using.
Calculate one- and two-year costs separately. If you plan to sell at the end, subtract the estimated proceeds; if you will keep the system, also keep a cash-outlay view without resale. Compare lower resale proceeds to check whether the purchase remains affordable.
| Aspect | Amount to include |
|---|---|
| Purchase | New hardware, required parts and installation |
| Operation | Working and idle energy, cooling and maintenance |
| Recovery | Estimated resale value at period end |
| Subscription savings | Bills actually cancelled or reduced |

Cancelling a KRW 30,000 monthly plan saves KRW 360,000 a year
Suppose you pay KRW 30,000 per month. Moving the work locally and cancelling the plan saves KRW 360,000 over 12 months or KRW 720,000 over 24 months. If you switch to a cheaper plan, include only the difference between the old and new bills.
If you keep the same subscription while doing some work locally, subscription savings are zero. Privacy and offline use may still justify the purchase. Separate money saved from new convenience to understand whether cost savings or another need drives the decision.

Calculate working time and idle time separately
Use a wall power meter to measure the complete system where possible: the CPU2, storage and fans use electricity too. Separate model-running hours from idle hours and calculate each as average watts multiplied by hours, divided by 1,000.
Summer cooling adds energy to remove the heat your hardware produces; enable the calculator’s cooling option to include it. Enter your household’s existing monthly use too, since the progressive tariff affects the cost of an extra 10kWh. The site divides annual calculations into seven other months, two summer months and three winter months.
How much do assumptions change the result?
The following is an illustrative calculation. Assume KRW 2 million for the purchase and additional parts, KRW 200,000 for a year of electricity and maintenance, and KRW 1.2 million resale proceeds after a year. Net ownership cost is KRW 1 million. Cancelling a KRW 30,000 monthly subscription reduces the net burden by KRW 360,000 to KRW 640,000.
Lower resale proceeds to KRW 800,000 and keep the subscription, and the net burden becomes KRW 1.4 million. The same hardware now differs by KRW 760,000 between the two assumptions. Vary resale and cancellation independently, and reconsider options that look cheap only under optimistic conditions.
| Aspect | Cancel plan; KRW 1.2m resale | Keep plan; KRW 0.8m resale |
|---|---|---|
| Purchase and parts | KRW 2,000,000 | KRW 2,000,000 |
| Electricity and maintenance | KRW 200,000 | KRW 200,000 |
| Subtract resale proceeds | −KRW 1,200,000 | −KRW 800,000 |
| Subtract subscription savings | −KRW 360,000 | KRW 0 |
| Net burden | KRW 640,000 | KRW 1,400,000 |
Complete the costs, then check the actual work
Save purchase, running cost, estimated resale and subscription savings for both one and two years. Add additional-system, repair and financing costs yourself where applicable; the calculator does not include them automatically. Consider whether installation and management reduce the time available for your work.
If the costs are affordable, finish by processing your usual documents and code. The system must deliver useful results within an acceptable wait. If your existing PC already does that work, keep tuning and retaining it as an option in the same cost comparison.

Terminology notes
GPU — A processor designed to handle many calculations in parallel. It performs model computations during AI inference.
Back to the textCPU — The central processor that runs the operating system and general-purpose instructions; in AI workloads it also handles tasks such as preprocessing and data movement.
Back to the text
Frequently asked questions
- What should you check first in this article?
- The total cost of ownership (TCO) of local AI hardware is its purchase price plus electricity and maintenance, less its estimated resale value at the end of the ownership period. If your goal is to reduce subscriptions, also include the bills you can actually eliminate. Start with the new expenditure and compare running the same work for one and two years.
Read next
read first
Choosing local LLM hardware: memory, speed and power
Executable programs and extensions
Running local LLMs across two or more GPUs
memory and model
How much VRAM or unified memory does a local LLM need?
Response speed and acceleration
LLM memory bandwidth vs compute: how to find the speed bottleneck
