Most articles about the cloud conclude that you should move to the cloud. This one does not. Some systems move to the cloud, end up costing twice as much, and run no faster.
The wrong question and the right one
The wrong question is whether the cloud is better than physical servers. The right question is whether your workload fluctuates, and by how much.
The cloud charges for what you use. Physical servers charge for what you bought. If your load sits steadily at 60% capacity around the clock, you are paying for elasticity you never use. If your load arrives in a four-hour window and the machine idles for the other twenty, physical hardware forces you to buy for the peak and waste most of the day.
Four cases for moving
1. Load swings widely and predictably
An e-commerce platform during a sale, an admissions system during application week, accounting software at quarter close. These systems peak at 5 to 20 times normal. Buying hardware for the peak means paying to keep it idle for most of the year.
2. You do not know how big this will get
A new product with no idea whether it will serve 100 or 100,000 users. The cloud lets you start small and grow with reality. Buying the wrong hardware at this stage is money you will not get back.
3. You need presence in several regions
Clients in Japan, the United States and Vietnam served from a single server in Hanoi will see round-trip latency above 200 milliseconds from the US. Building physical infrastructure on three continents is a large-enterprise problem. In the cloud it is a configuration change.
4. A small team with nobody dedicated to infrastructure
Physical servers need someone to swap failed drives, apply security patches, watch the temperature in the server room, and handle power failures. If nobody in your company does that work, the real cost is not the hosting fee — it is the risk when something breaks at two in the morning.
Three cases for staying on physical servers
1. Load is steady and well understood
An internal system serving 200 employees from 8am to 6pm, with load that barely changes all year. This is where physical servers win on cost, usually by a wide margin. For an equivalent configuration, 36 months of cloud typically costs 1.8 to 2.5 times the combined cost of buying hardware, colocation and maintenance.
2. Data residency is legally required
Some sectors require data to stay within national borders. Domestic cloud providers can meet this, but the field is narrower and the pricing is rarely as attractive as the international providers.
3. You move a lot of data outbound
This is the most common cost trap. Major cloud providers usually do not charge for data coming in, but they do charge for data going out. Video platforms, high-resolution image libraries and large file downloads can generate bandwidth bills that dwarf the compute cost. Estimate this before you commit, not when the first invoice arrives.
How to cost it properly
A fair comparison runs over 36 months and includes everything below. Many comparisons divide monthly rent by hardware price and stop there, and that conclusion is nearly always wrong.
| Item | Physical servers | Cloud |
|---|---|---|
| Hardware | One-off, depreciated over 3–5 years | Included in the monthly fee |
| Rack space, power, cooling | Monthly colocation | Included |
| Outbound bandwidth | Usually flat rate | Per GB, easy to underestimate |
| Backup and recovery | Build it yourself | Available, billed separately |
| Operations staff | Someone must be on call | Substantially reduced |
| Hardware failure | Downtime plus replacement | Provider absorbs it |
| Scaling up | Buy and wait for delivery | Minutes |
One number worth remembering
For systems with steady load, the break-even point between cloud and physical usually falls somewhere between month 18 and month 24. If you are confident the system will run for more than three years at constant load, the maths favours physical hardware. If you are confident about nothing, the cloud buys you the right to change your mind.
The hybrid option, usually overlooked
You do not have to choose all or nothing. A common and effective pattern is to put the steady baseline on physical servers and let the cloud absorb the peaks. The primary database runs on hardware in Hanoi while web servers scale out in the cloud when traffic climbs. The architecture is more complex, but it usually gives the best cost profile for mature systems.
Mistakes people make during migration
- Lifting and shifting unchanged. Moving virtual machines from your server room to the cloud without changing the architecture just changes where they sit. You gain no elasticity and pay cloud prices for it.
- No cost alerts. One bad loop in your code can generate a four-figure bill overnight. Turn on budget alerts on day one.
- Forgetting the cost of leaving. Pulling all your data out to move to another provider can be expensive. Consider this before you lock yourself into an ecosystem.
- Migrating everything at once. Move component by component, running each in parallel before cutting the old one off. Slower, but nobody loses a night of sleep.
Where to start
Before discussing providers, answer three questions: how does load vary across a year, how much longer will this system run, and what does an hour of downtime cost the business. Those three numbers determine the answer — not the provider feature list.
Written from the team's hands-on experience delivering infrastructure and software. No sponsorship, no product placement.