Back to Blog

How to protect a free compute tier from throwaway accounts

Free compute tier abuse takes capacity from paying customers. How to price an hour, set earned limits and see where basic signup checks break.

How much money is left? $96,000, your bank app says on the 3rd, and you'd budgeted for $105,000. A month ago it was $121,000, and the plan was to spend $16,000. That's the number your co-founder and your one investor have been hearing. You spent $25,000. Recent statements ran a little over, and you put that down to growth. This one's $9,000 over.

$96,000 ÷ $16,000 is six months of runway. $96,000 ÷ $25,000 is 3.84. More than two months of your company just left the forecast.

This is a scenario, so you can check every number. Follow the money backwards. The extra $9,000 is one debit: your cloud provider took $14,300 on the 1st, and five months ago that line was $5,300.

Monthly compute bill, illustrative0200040006000800010000120001400016000JunJulAugSepOctNovUS dollarsMonth
Monthly compute bill, illustrative050001000015000JunJulAugSepOctNovUS dollarsMonth

You split the invoice by plan. Paying customers cost $4,100. Free accounts you recognize cost $1,200. That leaves $9,000, and when you sort the other free accounts by hours used, the answer is at the top of the export. Nine hundred accounts used all 500 of their free hours, on small instances that cost you $0.02 an hour. Most never opened your dashboard after the day they were created.

900 accounts x 500 hours = 450,000 hours
450,000 hours x $0.02    = $9,000 a month
$9,000 x 12 months       = $108,000 a year

That's free compute tier abuse: accounts created to take the free hours and nothing else. At $108,000 a year, it costs more than you have left in the bank.

The bill turns out to be the smaller problem. Those hours have to run somewhere, and 450,000 hours ÷ 720 hours in a month is 625 instances running day and night. Your provider's quota is 1,100. Your busiest moment of the week used to need about 500. Now it needs 1,125, so every weekday around 3 p.m. the last 25 workloads have nowhere to start. The customers who deploy most often keep landing in that gap.

At 3:07 p.m. a customer on your $99 plan, 14 months in, writes: "Our deploy has been pending for 22 minutes. Third time this week. Are you down?" You aren't down. Your servers are busy running accounts nobody's watching. You refund the month. Two more customers ask by Thursday, which makes $297 in refunds, and a fourth stops answering and cancels on the 30th. That one was worth $1,188 a year.

So now you're looking at the free tier. Most of your paying customers started on it. You built it so a student could ship something without asking anyone for money, and you put it on the homepage because you were proud of it. It's also the obvious $9,000 to cut, with under four months to decide. Turn it off and next year's customers go with it. Cut it from 500 hours to 100 and a real hobby project gets about 3 hours and 20 minutes a day (100 ÷ 30 days), while the accounts you were aiming at just need five times as many signups.

Those numbers are invented. These aren't:

From our early beta

A team whose cloud bill normally ran $200 to $300 a month woke up to a $12,000 bill. Automated registrations had triggered thousands of expensive serverless functions, database operations and background jobs overnight.

That's 40 to 60 times a normal month, and it started at the registration form.

Why free compute tier abuse hits harder than other free plans

A free seat in a project tool is only useful inside that tool. A free CPU hour is useful for almost anything: crypto mining, proxies, scraping, someone else's batch jobs. So it's worth taking even if the person has no interest in your product. And where a real hobbyist works for an evening and goes to bed, a process uses all 500 hours and costs you the full $10.

Your limits are per account, which is the natural place for them, and that's why the obvious fix doesn't hold. A 500-hour cap works as designed: the account stops at 500. Then a new account arrives with 500 fresh hours, and nothing in the cap knows about the last one. Rate limits don't see it either.

How to set earned limits on free compute

Each of these steps limits what a single free account can cost you, and that's as far as an account-level rule can go.

Price one free account. Use your real instance price.

monthly cost per account=free hours×cost per hour\text{monthly cost per account} = \text{free hours} \times \text{cost per hour}

At 500 hours and $0.02, that's $10.

Give free accounts their own capacity. Your real free users average about 83 instances ($1,200 ÷ $0.02 = 60,000 hours, then ÷ 720), so a separate pool of 150 gives them room. The most it can cost in a month is 150 × 720 × $0.02 = $2,160, which takes the bill from $14,300 to $6,260 ($4,100 + $2,160). The other 950 instances in your quota now belong to paying customers, so the 3 p.m. gap closes today.

The catch shows up a day later. Those 625 instances of throwaway work are competing for 150 places, and your real free users wait behind them.

Start small and let people earn more. Give a new account 20 hours a month instead of 500. More hours come from something a real user does anyway, like deploying on five separate days.

Stop work that never stops. Sleep idle instances and cap how long a free run can last. A process that's always busy never goes idle, so the run cap does the work there, and real users with long jobs will hit it too.

Run nothing expensive at registration. Wait until the email is confirmed and the person asks for an instance. See how to stop expensive work running on unverified signups.

NoYesNoYesNew accountEmail confirmed?No compute yet20 hours a monthSleeps when idleDeployed on 5 separatedays?100 hours a monthSleeps when idleMore hours need a paidplan
NoYesNoYesNew accountEmail confirmed?No compute yet20 hours a monthSleeps when idleDeployed on 5 separatedays?100 hours a monthSleeps when idleMore hours need a paidplan

Run the scenario through that. At 20 hours, each account costs $0.40 a month (20 × $0.02), so 900 accounts cost $360 instead of $9,000. If all 900 earned 100 hours, that's 900 × $2 = $1,800.

A real user who needs more than 20 hours in the first week gets a slower start, and some will leave before they've earned anything. And the limit still belongs to the account, so every new account starts with its own 20 hours. See how to set free plan limits that don't invite extra accounts.

The signup checks you'll try next, and how they break

Earned limits make each account cheaper. To get fewer of them, most teams add checks at signup, starting with email. A unique-email rule compares strings, and one mailbox can have many. Gmail ignores dots, so jane.doe@gmail.com and janedoe@gmail.com reach the same inbox, and mail to jane+anything@gmail.com lands there too. That's simply how email works.

Basic checkWhat it catchesWhat it missesWho it wrongly blocksWhat it costs to keep working
Email verificationTypos, dead inboxesAny real second inboxPeople whose code lands in spamDeliverability work
Unique emailsThe same address twiceEvery different addressAlmost nobodyVery little
Disposable-domain blocksKnown throwaway servicesNew domains, ordinary mailboxesPrivacy forwarding usersA list updated forever
Gmail dot and plus rulesOne inbox written several waysA second real mailboxPeople who sort their mail with plus addressesUpkeep as mail services change
CAPTCHASimple scriptsPeople signing up by handScreen reader usersFees, lost signups
Phone requiredCasual second accountsAnyone with another numberPeople without a reliable mobileA charge per text
Card requiredCasual second accountsAnyone with another cardStudents and hobbyistsFewer signups
One account per IPRepeats on one connectionAny other networkUniversities, offices, householdsExceptions and tickets
VPN blocksKnown VPN rangesEverything off your listDevelopers on company VPNsAnother list
Rate limitsBursts of job startsOne process running steadilyReal users in a busy hourTuning per endpoint
Usage capsOne account overrunningEach new account's fresh capAnyone at the cap mid-projectDeciding what stops
Shorter trials, fewer hoursSome cost per accountMore accounts make up the differenceEvery honest userFewer people reach paying
Manual reviewWhat a person can spotAnything after hoursWhoever looks odd to a tired reviewerHours every month

Stack these and your free tier becomes a form with a puzzle, a text message and a card field, and the student you built it for gives up at the card field. Manual review doesn't rescue it: 900 accounts at two minutes each is 30 hours, and next month's accounts are different ones. A signup check only runs at signup, so the 900 accounts already running never meet it.

All of them ask about the account in front of them. You want to know whether the person behind that account already had their 500 hours, and nothing in your tables can tell you that.

Deciding who gets the next 500 hours

Every row in that table needs someone to keep it alive, and you already have a company to run. You don't have to build your own abuse prevention system. AbuseZero is made for the moment that table runs out: an account asks for free compute, your server makes one API call, and you get a clear allow or deny under rules you control.

Your 500-hour cap resets for every signup, but with AbuseZero, creating a fresh account isn't enough to erase previous activity. It recognizes repeat devices across new accounts, even when email addresses keep changing. Your email rules stop at the string, while AbuseZero checks mailbox aliases, disposable domains and mail routing. Your VPN list stops at whatever's on it, and AbuseZero sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. With multi-layer AI risk analysis, AbuseZero examines 140,000+ signals, detects linked claims and flags abuse that single checks miss. There's also a proprietary algorithm, with AI that checks the claim again.

Each level of your earned limits can be its own benefit with its own eligibility, so the first 20 hours can be once per user. Observe mode shows you what would be denied before anyone is blocked, and the quickstart gets your first benefit running. Your platform still meters the hours and enforces the result.

Frequently asked questions

What is free compute tier abuse?

It's creating accounts to take free compute hours more times than you meant to give them.

Should I remove my free tier because of abuse?

Try separate capacity and earned limits first. If it still loses money after that, here's how to decide whether to remove it.

How do I stop free accounts slowing down paying customers?

Give free accounts their own pool and budget, so paying customers' work never waits behind free work. AbuseZero's evaluation comes earlier, when the free compute is claimed.

Keep your free tier running

Keep reading

Copying your prompt…

Copied the to your clipboard. Give it to your coding agent & it will handle the rest.