← back to blog

What A Thousand Profiles Actually Cost To Run

The number that matters isn’t the profile count

Everyone who gets into multi-accounting asks the same first question: how many profiles can I run? That’s the wrong question, and it’s the one that gets people into trouble. The right question is what does one profile cost, fully loaded, per month, when it’s part of a thousand instead of part of ten? Those two numbers are not close to each other, and the gap between them is where most budgets quietly blow up.

A single profile on a laptop, using a free trial of an anti-detect browser and a proxy you already had lying around, costs almost nothing. That number is useless for planning. It tells you nothing about what happens when you need proxy diversity across a thousand distinct IPs, storage for a thousand distinct cookie jars and cache states, and enough human attention to notice when profile 640 starts throwing captchas at double the rate of the other 999.

Scaling browser profiles cost isn’t a single line item you can look up on a pricing page. It’s a stack: software licensing, proxy bandwidth or IP rental, compute and storage, the labor to set up and babysit accounts, and the replacement cost every time a profile gets flagged and has to be rebuilt from scratch. Most people budget for the first item and get surprised by the other four. This piece walks through each layer, builds a model for what a thousand profiles actually runs, and looks at where operators cut corners and what breaks when they do.

What a browser profile actually contains

Before the cost math makes sense, it helps to be precise about what you’re actually paying to maintain. A “profile” in an anti-detect browser like Multilogin, GoLogin, Kameleo, AdsPower, or Dolphin Anty is a bundled identity: a set of browser fingerprint attributes plus a persistent storage state, wrapped around a Chromium or Firefox engine that’s been patched to report those attributes consistently.

The fingerprint side includes things like canvas rendering output, WebGL renderer strings, installed fonts, screen resolution and pixel density, hardware concurrency (CPU core count as reported to JavaScript), audio context fingerprint, timezone, locale, and the WebRTC-exposed local IP. Some of these are spoofed at the browser engine level; some are just configuration values passed to the OS. The point is that all of them need to be internally consistent, because a fingerprint that claims to be a MacBook with 8 cores while reporting a GPU string from a low-end Android device is a red flag on its own, no proxy required.

The storage side is simpler but heavier on disk: cookies, local storage, IndexedDB, cached login tokens, browsing history. This is what makes a profile “warm” over time, the accumulated signal that looks like a real, aged account rather than a browser that opened for the first time an hour ago. Both halves cost something to maintain at scale, and both are why “just buy more licenses” is not actually a scaling strategy.

Licensing: the line item people obsess over

Licensing is the cost operators think about first because it’s the one with a visible price tag. Most anti-detect browser vendors sell in tiers, either per number of profiles or per number of team seats, with enterprise tiers unlocking API access, team roles, and cloud-hosted profile storage instead of local-only. At a thousand profiles you are almost certainly on an enterprise or custom tier, which usually means direct negotiation rather than a self-serve checkout.

Here’s the part that surprises people who are new to this: licensing is usually the smallest line in the whole stack once you’re past a few hundred profiles. A per-profile license fee, even at the higher end of what vendors in this category charge, is a fixed and predictable cost. It doesn’t scale with usage, it doesn’t spike when something goes wrong, and it doesn’t disappear when a profile gets banned. Compare that to proxy bandwidth, which is metered and can double in a bad week, or labor, which grows with every account that needs manual attention.

The mistake operators make is optimizing hard for the cheapest license and then discovering that the money they saved on software is a rounding error next to what they’re about to spend keeping a thousand distinct residential or mobile IPs alive. Pick the tool based on whether its team features, API, and cloud profile hosting actually fit how you work, not based on shaving a few dollars off the per-seat price. The license is not where the budget lives.

Proxies are where the real money goes

If licensing is the line people obsess over, proxies are the line that actually determines whether the operation is financially sane. Every profile that needs to look like a distinct, independent user needs its own IP, ideally one that doesn’t share a subnet with dozens of other profiles logging into the same platform. That means the proxy bill scales almost linearly with profile count, in a way licensing doesn’t.

There are three broad tiers here, and they have very different cost and risk profiles. Datacenter proxies are the cheapest per IP and the easiest to detect as non-residential, because the ASN and reverse DNS give them away to anyone doing IP reputation scoring. Residential proxies route through real ISP-assigned IPs, usually via a peer network, and cost more, typically billed per gigabyte rather than per IP, which means usage patterns matter as much as raw profile count. Mobile or 4G proxies route through actual carrier SIMs and cellular networks, which is the most expensive tier but also the one where IP reputation tends to be cleanest, since carrier-grade NAT means thousands of legitimate phones share the same IP block already.

At a thousand profiles, proxy selection isn’t a single decision, it’s a portfolio decision. Different platforms and different account types tolerate different proxy tiers differently, and mixing tiers based on what each batch of accounts actually needs is usually cheaper than putting everything on the most expensive option out of caution.

Compute and storage: local rigs vs cloud-hosted profiles

Every open Chromium instance, even a lightweight one, eats real RAM and CPU. Run a hundred profiles concurrently on a local machine and you’ll find the practical ceiling fast, usually somewhere between RAM exhaustion and the machine’s disk I/O falling over from a hundred simultaneous cache writes. A thousand profiles run concurrently on local hardware is not realistic for most setups, which is why cloud-hosted profile infrastructure exists as a category in the first place.

Cloud-hosted profiles shift the compute cost from a capital expense, buying and maintaining machines, to an operating expense, renting VM or container capacity from the anti-detect vendor or a separate cloud provider. This is usually billed by concurrent running profiles or by compute hours, separate from the base license fee. The tradeoff is that you’re now also paying for network egress and storage on someone else’s infrastructure, and you’re dependent on that infrastructure’s uptime.

Storage itself is a smaller but non-trivial cost at scale. A thousand profiles, each with months of accumulated cookies, cache, and session data, adds up to real disk space, and if you’re paying for cloud storage per profile per month, that number compounds the same way proxy bandwidth does. The practical takeaway is that compute and storage need their own line in the budget, separate from licensing, because they scale with concurrency and profile age, not with the number of licenses you bought.

The labor line nobody puts in the spreadsheet

This is the cost category most new operators leave out entirely, and it’s often the largest one by the time a thousand profiles are actually running. Every profile needs to be created, which usually means an email, sometimes a phone number for verification, and a period of “warmup” activity that makes the account look like it belongs to a real, gradually-engaged user rather than something that logged in once and immediately started doing volume actions.

Warmup isn’t automatable in the naive sense, not if you want it to hold up, because the whole point is that the behavior needs to look organic: varied session lengths, plausible browsing patterns, realistic delays between actions. Some of this can be scripted, but scripted behavior that’s too uniform across a thousand profiles is itself a pattern, so someone needs to be watching for that uniformity and breaking it up. That’s human hours, and it doesn’t scale down linearly, because reviewing flagged accounts, solving captchas that resisted automated solvers, and triaging which accounts are worth saving versus abandoning all take a person’s actual attention.

Budget for this honestly. If you model a thousand profiles as a software and proxy cost with zero labor, you will hit a wall the first week something goes wrong at scale, because nobody is watching for it.

Churn: the cost of profiles that get flagged

No anti-detect browser is undetectable, and none of the tools in this category, Multilogin, GoLogin, Kameleo, AdsPower, Dolphin Anty, or any other, can honestly claim otherwise. Platforms invest heavily in detection, and some fraction of any profile pool is going to get flagged, limited, or banned over time. The question that actually matters for budgeting isn’t whether that happens, it’s how often, and what it costs each time it does.

Every flagged profile is a sunk cost on the proxy, the warmup labor, and any verification steps that went into it, plus a new cost to replace it: a fresh IP, a fresh identity, and another warmup cycle before that profile is useful again. If your churn rate is five percent a month, you’re effectively rebuilding fifty profiles every month out of a thousand, and that rebuild cost needs to be in the model from day one, not discovered after the fact.

Churn rate isn’t fixed. It depends on proxy quality, on how careful the warmup process is, on how consistent the fingerprint configuration is with the proxy’s claimed location, and on how aggressively the platform you’re operating on has invested in detection. Two operators running the same anti-detect software can have wildly different churn rates because everything around the software differs. This is also why nobody honest in this space will promise you a specific ban rate or guarantee an account survives. The software is one input among several, not a shield.

Building the model: what a thousand profiles actually costs

To make this concrete, here’s a worked model. Treat the numbers as an illustrative framework for how the categories interact, not as verified pricing from any specific vendor, since actual rates vary by provider, region, and negotiated volume.

Say licensing for a thousand profiles on an enterprise tier lands somewhere in the low thousands of dollars a month, a fixed cost regardless of usage. Proxies are the biggest variable: if you’re running a mixed portfolio, some datacenter, mostly residential, a slice of mobile for the accounts that need the cleanest reputation, and each profile uses a modest amount of bandwidth per month, the proxy bill alone can easily run several times the licensing cost, simply because it’s metered and scales with both profile count and activity level. Compute, if you’re cloud-hosting concurrent sessions rather than running everything locally, adds another meaningful chunk, scaling with how many profiles are active at once rather than the total pool size.

Then labor: even a lean team spending a modest number of hours a week on warmup, captcha triage, and monitoring turns into a real monthly cost once you price out the hours. And churn adds a recurring tax on top of all of it, because every replaced profile drags its proxy and labor cost with it.

Add those five categories up and the total for a thousand profiles is routinely several times the licensing fee alone. Anyone budgeting off the license price sheet is looking at a small fraction of the real number.

Where the corners get cut, and what breaks first

Under budget pressure, the first corner operators cut is proxy quality, moving from residential or mobile to cheap datacenter IPs to save money. This is also usually the first thing that breaks, because datacenter ASNs are trivial to flag and platforms that care about multi-accounting have had years to build IP reputation lists that catch this exact pattern.

The second corner is fingerprint diversity. Some setups reuse near-identical fingerprint templates across large batches of profiles to save configuration effort, which creates a cluster of profiles that look suspiciously alike to anything checking for fingerprint similarity across sessions, even if each individual fingerprint looks plausible on its own. A platform doesn’t need to catch one profile to catch the pattern, it needs to notice that forty profiles share the same unusual combination of screen resolution, font list, and hardware concurrency value.

The third corner is warmup time, skipping or shortening it to get accounts productive faster. This is often invisible until it isn’t, because a rushed account can look fine for weeks and then get caught in a batch review that platforms run periodically rather than in real time. None of these shortcuts are dramatic on their own. They’re cost-saving decisions that quietly raise churn, and because churn has its own cost, the “savings” from cutting corners often get eaten by the replacement cost of the accounts that get caught.

How platforms are actually looking at this

It’s worth understanding what’s on the other side of this, because the cost model only makes sense in light of it. Platforms that care about multi-accounting aren’t relying on a single fingerprint check. They’re combining IP reputation scoring, TLS and network-level fingerprinting like JA3, browser-level entropy from canvas and WebGL and font enumeration, behavioral signals like typing cadence and mouse movement, and cross-session consistency, meaning whether the same account’s fingerprint and IP stay coherent over time or jump around in ways a real device wouldn’t.

The part that matters most for a thousand-profile operation specifically is cluster detection. A single well-configured profile with a clean residential IP can look, on paper, indistinguishable from a real user. A thousand profiles that all came from the same anti-detect tool, provisioned in the same session, sharing subtle statistical similarities in how their fingerprints were generated, are a different problem entirely, because detection at scale doesn’t need to catch the individual, it needs to notice the batch.

This is why cost and detection risk are tied together rather than separate concerns. Spending less on proxy diversity, spending less on fingerprint variation, spending less on staggering account creation over time, all of these show up as statistical patterns to a platform doing cluster analysis, regardless of how good the underlying anti-detect browser is at spoofing any single attribute.

Cost and risk don’t move in a straight line

It would be convenient if more spend simply meant less risk, but that’s not how this works, and no operator who’s actually run a large pool should claim otherwise. Spending more on the most expensive mobile proxies for every single profile, regardless of what that profile is actually doing, is often wasted money, because plenty of low-stakes accounts don’t need that level of IP reputation to survive.

Conversely, spending less doesn’t necessarily mean getting caught faster. Detection is probabilistic and platform-dependent, and plenty of underfunded, corner-cut operations run for a long time without incident, simply because they didn’t attract enough volume or the right kind of scrutiny to trigger a review. That’s not a reason to run that way, it’s a reason not to trust survival as evidence that a setup is sound. Absence of a ban is not proof of a good setup, any more than a ban is always proof of a bad one.

The honest way to think about the relationship is that spend on proxy quality, fingerprint diversity, and warmup time reduces the probability of detection at the margins, without ever taking it to zero, and every dollar past a certain point buys smaller reductions in that probability. There is no tier of anti-detect browser, and no proxy budget, that makes an account unbannable. Budgeting for a thousand profiles has to include the assumption that some percentage will be caught regardless of how well the operation is run.

What changes between 10 profiles and 1000

The gap between a small pool and a thousand profiles isn’t just more of the same thing, it’s a change in what actually determines success. At ten profiles, an operator can hand-configure each one, personally track which accounts are warm and which are fresh, and notice a problem within a day just by looking at the dashboard. None of that is possible at a thousand.

At scale, the operation needs systems that don’t exist at small scale: automated monitoring that flags anomalies like a spike in captcha rate or a sudden drop in session length, a proxy rotation and allocation system that tracks which IPs are assigned to which profiles and avoids overlap, and some form of staggered provisioning so that a thousand accounts aren’t all created and warmed up on the same timeline, which is itself a detectable pattern. Team structure changes too, moving from one person doing everything to defined roles, someone owning proxy health, someone owning account creation, someone triaging flags.

Costs shift accordingly. Small-scale operations are dominated by whatever the person running them is willing to do manually. Large-scale operations are dominated by infrastructure and process costs that exist specifically because manual attention no longer scales. This is also why cost per profile at a thousand is not simply the small-scale cost per profile multiplied out. Some things get cheaper per unit, like negotiated proxy rates. Some things get more expensive per unit, like the monitoring and process overhead needed to catch problems that used to be obvious at a glance.

What to actually change this week

If you’re already running a meaningful pool of profiles, the useful move isn’t to rebuild the whole stack, it’s to audit the categories that are usually invisible. Start by separating your actual monthly spend into the five buckets from this piece, licensing, proxies, compute, labor, and churn-driven replacement, even roughly. Most people find their spreadsheet only tracks the first one, and once the other four are visible, the real cost per profile is usually two to four times what the license price sheet suggested.

Next, check your proxy-to-profile ratio and tier mix against what your accounts actually need, rather than defaulting everything to the cheapest or most expensive option. Not every profile needs mobile-grade IP reputation, and not every profile can survive on datacenter. Match the tier to the stakes.

Track churn as an actual number, not a feeling. If you don’t know your monthly replacement rate, you can’t know whether your proxy or warmup process is actually working, and you can’t forecast next month’s real cost. And budget labor honestly, including the time spent triaging flagged accounts, because that time is real even when it doesn’t show up on an invoice.

Finally, resist the instinct to treat the anti-detect browser itself as the lever that fixes detection problems. It’s one part of a stack, and the parts around it, proxy quality, fingerprint diversity, warmup discipline, and monitoring, usually matter more than which vendor’s logo is on the software.

The honest bottom line

Running a thousand browser profiles is not a software purchase, it’s an operation, and it has an operation’s cost structure: a small fixed cost for tools, a large variable cost for infrastructure, a real and ongoing labor cost, and a tax on top of all of it for the accounts that don’t make it. Anyone selling you a single number for “cost per profile” without asking about your proxy mix, your churn rate, and your labor setup is selling you a license price, not a real answer.

None of the tools in this category are undetectable, none are unbannable, and running any of them, well configured or not, doesn’t remove the risk that comes with operating multiple accounts on a platform that doesn’t want you to. What good tooling and a properly budgeted operation actually buy you is a lower and more predictable rate of loss, not a guarantee of zero loss. Plan for the categories this piece walked through, price them honestly, and treat the churn number as the real scorecard, because it tells you more about whether your setup is working than any single tool’s marketing ever will.

We test this stuff hands-on and write up what actually holds up and what doesn’t. For more breakdowns like this one, head back to the Anti-Detect Review home page.

Get new guides and videos first — join the Telegram channel.

need infra for this today?