Renting the always-on machine your product runs on, instead of buying and running one yourself.
Why it exists
A server has to stay powered, cooled, physically secured, and connected to the internet every hour of the year, and running that yourself means a room, spare hardware, and staff on call for failures. Hosting exists so you rent that capability from a provider who already runs it at scale, and pay only for what you use. That choice is not just plumbing: cheaper hosting can buckle when traffic climbs, and where the machines physically sit affects both speed for your users and which data residency laws you must follow.
How it actually works
You will not keep a server in your kitchen. You rent space on someone else's machines from a hosting provider, who keeps them powered, cooled, and connected. You pay for what you use.
Hosting ranges from a simple shared plan for a static page up to full cloud platforms that give you hundreds of machines on demand.
A senior PM walks you through it
A junior PM, stuck
Finance forwarded me the hosting bill and asked why it has to roughly double before the Eid campaign, and all our engineer said in the thread was "we need a bigger box." I do not actually know what we are renting today or why a sale forces us onto something bigger. I have to answer Finance by end of day.
"A bigger box" has a real meaning you can see, and reading it takes about five minutes. What we rent is one always-on machine in a data center, and it has a fixed ceiling of CPU and memory; when Friday lunch traffic climbs, the machine's own metrics show it running out of room. I pulled the host metrics from that single server across last Friday, from the quiet night through the lunch peak, and trimmed them so we can read every line.
Host metrics for the single rented app server, Friday 2026-03-06
one machine, 2 vCPU, Singapore region (ap-southeast-1)
12026-03-06T03:07:11Z INFO host cpu 3% mem 38% orders/min 1 last-req 42ms
22026-03-06T08:00:04Z WARN host cpu 61% mem 90% orders/min 12 breakfast-promo
32026-03-06T11:30:00Z INFO host cpu 40% mem 66% orders/min 9
42026-03-06T12:30:00Z INFO host cpu 74% mem 80% orders/min 22 lunch-peak-begins
52026-03-06T12:41:00Z WARN host cpu 92% mem 89% orders/min 34
62026-03-06T12:52:00Z WARN host cpu 100% mem 95% orders/min 38 requests-queueing
72026-03-06T12:41:03Z INFO edge request from Dhaka served from region ap-southeast-1
Click a step to see the lines it points at.
Mistakes I've seen
Approving "a bigger box" without asking for the peak metrics. If cpu and memory both sat near 50% during the last sale, the higher tier buys headroom you will never touch.
Treating the hosting bill as a fixed cost line. It scales with the ceiling you rent, so a campaign that doubles traffic can double the bill for those weeks; that is a planning number, not a surprise.
Forgetting the machine has a location. "Where is it hosted" changes both speed for far-away users and which country's data laws apply, and both are product calls, not just infra ones.
Assuming a bigger box removes the risk. One machine is still one machine; a higher tier lifts the ceiling but the whole product still rides on a single point of failure until you add more machines.
Tell Finance: "We rent one machine with a fixed CPU and memory ceiling; last Friday's lunch peak pushed it to 100% CPU with requests queueing, and Eid is a larger sale, so we pre-rent a higher tier for those weeks and step back down after. The increase buys headroom for the campaign, it is not a permanent doubling." You read a host's own metrics and turned "a bigger box" into a cost you can defend.
Where a PM meets this
Hosting choices affect cost and reliability. Cheaper hosting can mean the site goes down when traffic climbs.
"Where is it hosted?" matters for speed (closer to users is faster) and sometimes for law (some data must stay in-country).
Hear it in a meeting
"We're moving hosting to a bigger plan before the campaign."