CloudCostCompareCloud vs on-premises cost: the inputs people forget
Short answer: a fair comparison counts everything on both sides. Most mistakes come from comparing a cloud bill with only the purchase price of hardware. The cloud wins when load is spiky, uncertain or small relative to the team. Owned or committed capacity wins when load is steady and high and you can run it well.
Key points
- Count space, power, network, licences, staff and spare capacity on the on-premises side.
- Count data transfer, support and extras on the cloud side.
- Spread hardware over its real useful life.
- Utilization is the biggest lever: idle owned hardware costs the same as busy hardware.
- Test the result with a peak double your forecast.
What does the on-premises side include?
- Hardware, spread over its useful life, with a replacement plan.
- Space, power and cooling, or the rent for a colocation rack.
- Network equipment and links.
- Software licences and support contracts.
- Staff time to install, patch, monitor and replace parts.
- Spare capacity. You buy for the peak, so average use is lower than capacity.
- Backup and recovery, including a second site if you need one. See disaster recovery cost tiers.
What does the cloud side include?
- Compute, storage and managed services at list price, or at your committed rate. See reserved instances and savings plans.
- Data transfer, logs and the other hidden charges.
- Support plan. See cloud support plan costs.
- Staff time, which is smaller for managed services and larger for self-managed servers in the cloud.
What does a worked example look like?
Made-up round numbers to show the method, not real prices. A workload needs 10 servers’ worth of capacity at peak but averages 4.
| Item | On-premises | Cloud (on demand) |
|---|---|---|
| Servers | 10 bought at 5,000 each, over 4 years = 50,000 / 48 = 1,042 per month | 4 on average x 730 h x 0.20 = 584 per month |
| Rack, power, cooling | 300 per month | included |
| Network and licences | 250 per month | 100 per month |
| Staff time | 0.3 of a person at 6,000 = 1,800 per month | 0.1 of a person = 600 per month |
| Total | 3,392 | 1,284 |
Here the cloud wins because the load averages 40 percent of peak. Now make the load flat at 10 servers all day, and assume a 40 percent commitment discount. Cloud compute becomes 10 x 730 x 0.20 = 1,460, less 40 percent = 876, so the cloud total is 876 + 100 + 600 = 1,576. If your existing team can run the hardware with no extra staff, the on-premises total falls from 3,392 to 1,592, and the two are close. The inputs that change the answer are utilization and staff time, not the server price.
Where does each side tend to win?
| Situation | Leans toward |
|---|---|
| Steady, predictable load running all year | On-premises or committed cloud capacity |
| Spiky or uncertain load | Cloud, since you pay only for what runs |
| Small team with no operations staff | Cloud managed services |
| Large data leaving the site every day | On-premises, or a design that limits egress |
| Need for another region quickly | Cloud |
How should I run the comparison?
- List workloads with their CPU, memory, storage and traffic.
- Price the cloud side per month with an estimator.
- Price the on-premises side per month: hardware over its life, plus running costs.
- Add staff time to both.
- Test with a peak double your forecast, and with a year of no growth.
Which mistakes are common?
- Comparing the cloud bill with the purchase price only.
- Ignoring the spare capacity bought for peaks.
- Using list prices in the cloud and discounted prices on-premises, or the reverse.
- Counting staff on one side only.
- Spreading hardware over seven years.
Decision checklist
- Is utilization measured, not guessed?
- Are staff costs counted on both sides?
- Is hardware replacement included?
- Have I tested a doubled peak and a flat year?
Next steps
Use the web app estimator for the cloud side. If a move is likely, read the cloud migration cost estimate and planning a cloud budget. Your own purchase prices and staff costs are inputs we do not have, so they have to come from you.
Frequently asked questions
Is the cloud cheaper than on-premises?
For spiky or uncertain loads and small teams it often is. For steady, predictable load that runs all year, owned hardware or committed cloud capacity can cost less. The answer depends on utilization, staff cost and how long the hardware lasts.
What costs do people forget when pricing on-premises?
Space, power and cooling, network equipment, software licences and support, staff time, spare capacity for peaks, backup and a second site for recovery.
What is total cost of ownership (TCO)?
The full cost of running a system over its life, including purchase, operation, staff and replacement, expressed per month or per year so two options can be compared.
How long should I spread hardware cost over?
Use the realistic useful life, often three to five years for servers, and include a replacement plan. Spreading it over too long makes on-premises look cheaper than it is.
Does moving to the cloud remove the need for operations staff?
It reduces some work, such as hardware and power, but adds cloud design, cost control and security work. Managed services cut staff time more than self-managed servers in the cloud.