Shared vs Dedicated ISP Proxies: What Each One Costs

Holding a login, an allowlisted address, or anything where a block has to be traceable to your own behaviour? Buy dedicated. Exclusivity is not a comfort feature here, it is the only way to debug what happens to the address.

Doing read-only work where a residential ASN is required but no state persists between requests, such as regional price checks, ad placement snapshots or localisation spot checks? Shared is fine, and the money you save is real.

Not sure whether the target checks the ASN at all? Do not buy ISP addresses yet. Run the target against ordinary datacenter IPs first. The most common ISP purchase we see is one nobody needed, made because a scraper was failing for a reason that had nothing to do with the exit network.

Vendors use the word "shared" for three arrangements with completely different risk profiles. Before comparing prices, work out which one a page is selling.

Rotating ISP pool. You send a request to a gateway and get back a static residential address chosen from a pool that every customer draws on. You never hold an address, and the same address may serve you, then three other customers, inside a minute. Our explainer on rotating ISP proxies covers the mechanics. This is the cheapest ISP product and the least useful for account work.

Shared or semi-dedicated static ISP. A specific address is assigned to you for the term, and simultaneously to a small number of other customers. You keep it, so does everyone else on it. Vendors usually cap the co-tenant count at two or three, though very few publish the number. Our post on semi-dedicated proxies covers how those caps are enforced, or not.

Dedicated static ISP. One address, one customer, for the length of the term. Nobody else routes through it.

Webshare's own product navigation, read on 23 September 2026, is unusually clear about the middle tier. It lists three static residential tiers side by side, and the accompanying copy on that page states that private static residential proxies "may be shared with a limited number of users" while dedicated proxies "provide maximum exclusivity". That is the distinction vendors usually bury.

IPRoyal takes the opposite approach. Their ISP proxy page, read the same day, sells dedicated addresses only and describes the entire product as dedicated static residential IPs, with no shared tier offered at all. Two reputable vendors, two different views on whether shared ISP should exist as a product.

A datacenter proxy vendor who runs out of addresses buys another block, announces it, and is selling by the end of the week. The address space is plentiful and the cost per address is close to noise.

ISP proxies do not work like that. An ISP address has to be allocated by an actual consumer internet provider, announced under that provider's ASN, and kept there. Getting a range requires a commercial relationship with the ISP, justification for the allocation, and usually a minimum term. Vendors cannot conjure a new /22 of residential-looking space on a Tuesday because demand spiked. If you want the background on why the ASN is the whole point, see what ISP proxies are.

Two consequences follow, and both cut against the intuitions you build buying datacenter proxies. The first is that shared tiers exist on ISP products specifically because supply is tight. On datacenter, shared proxies are a price-discrimination product: the vendor has plenty of addresses and sells cheap slots to customers who would otherwise buy nothing. On ISP, shared slots exist because the vendor genuinely cannot give every customer their own address.

The second is that the thing you are renting is more fragile. The entire value of an ISP address is that it resolves to a consumer ISP network and scores as residential in the IP intelligence databases that targets query. Burn that reputation and you have a slow datacenter IP. On an ordinary hosting address you started from a low base, so a co-tenant's bad behaviour costs you less in relative terms. Our piece on IP reputation walks through how those scores are built and how long they take to recover.

Here is what two vendors published on their own pages on 23 September 2026. Every figure below was read on the date shown, and both vendors change rates without notice.

Webshare also publishes a volume ladder on the shared static residential page. On 23 September 2026 it ran flat at $0.30 per proxy from 20 through 250 addresses, then stepped down: $0.285 at 500, $0.27 at 1,000, $0.255 at 2,000, $0.24 at 3,000 and 5,000, and $0.225 at 10,000. The monthly totals shown alongside match those unit rates exactly, so 100 addresses listed at $30.00 and 1,000 at $270.00 after a 10% volume discount off the $300.00 list.

Now the number nobody puts in a comparison table. Compare the shared-to-dedicated multiple inside each product family, using the same vendor so the comparison is fair:

• Datacenter: $0.018 shared to $0.77 dedicated is roughly a 43x step.
• Static residential: $0.23 shared to $1.47 dedicated is roughly a 6.4x step.

Those two numbers explain why the datacenter playbook misleads people. On datacenter, shared addresses cost so little that buying a hundred and discarding half of them is a sane strategy, and the dedicated tier is a serious commitment by comparison. On ISP, you are already paying real money for the shared tier, and the exclusivity upgrade is a proportionally much smaller step.

One caveat on the arithmetic, because it matters. The $0.23 shared rate is a volume floor, and the $1.47 dedicated rate is the headline. Compared at a matched count of 250 addresses, where the published shared ladder reads $0.30, the multiple is about 4.9x. We did not find an equivalent published volume ladder on the dedicated tier from the same navigation, so treat 4.9x as an upper bound at that count and price your own quantity at checkout rather than trusting either multiple.

This is the section the datacenter comparison does not need, and the one that should decide your purchase.

When a dedicated datacenter address gets blocked, most vendors swap it quickly and without argument, because they have thousands of spares sitting in the same subnets. The swap costs them almost nothing.

ISP addresses are scarce, so vendors ration replacements. The policies we see fall into four shapes, and you want to know which one you are buying before you spend anything:

1. Quota. A fixed number of swaps per address per month, or per account per month. Once you burn through it, you buy new addresses.
2. Cooldown. Unlimited swaps in principle, with a minimum holding period per address. The practical effect is a cap.
3. Evidence gate. The vendor swaps only when you demonstrate the address fails an objective test, usually a blacklist lookup rather than "my target blocks it". This one catches people out, because most ISP addresses that stop working for scraping are not on any public blacklist.
4. Paid. Swaps cost money, sometimes at the full per-IP rate.

Now apply each of those to the shared tier, and notice that they mostly stop making sense. You cannot ask for a replacement of an address you do not own. The best a vendor can do is move you to a different shared address, which arrives with its own co-tenants and its own unknown history. On a rotating ISP pool the concept does not exist at all, because you never held an address to begin with.

That gives you one question to put to any ISP vendor in writing before you buy, and the answer tells you more than the price does:

On a dedicated ISP address that stops working against my target, what is your replacement policy: how many swaps per month, what evidence do you require, and what is the turnaround from request to working address?

A vendor who answers precisely is selling you a replacement entitlement. A vendor who answers with "just open a ticket" is selling you an address and hoping. Our guide to reading a proxy provider SLA covers where that commitment usually lives in the contract, and what the exclusions typically swallow.

Four things go wrong with co-tenants, and only the first is obvious.

Your rate limit is not yours. Targets count requests per IP. If two other customers are hitting the same platform from the same address, your effective budget is a third of what you planned, and it moves without warning as their workloads change. You cannot see their traffic, so you cannot forecast your own headroom.

Account clusters get linked. If you and a co-tenant both log into the same platform from one residential address, the platform observes two unrelated account clusters sharing a consumer IP. That is an unusual pattern for a genuine home connection and a well-known linkage signal. You will never see the co-tenant's accounts, and their suspension can take your accounts with it. Our post on why proxy accounts get suspended covers the signal in more detail.

Debugging becomes guesswork. When a dedicated address starts throwing challenges, the cause is in your traffic, and you can find it. When a shared address does the same thing, you cannot separate your requests from theirs, so every hypothesis you form is untestable. Teams burn days on this.

You cannot make promises about the address. Allowlisting a shared IP at a partner API, or listing it in a compliance record as the origin of a set of requests, are both things you have no standing to do. Somebody else's traffic will arrive from that address under your allowlist entry.

Against those four, the shared tier gives you one genuine advantage besides price: if an address does get burned, you did not lose an asset you were depending on.

The numbers below are an illustrative model built on Webshare's published rates as read on 23 September 2026. These are not measurements from any test we ran, and your own counts and rates will differ.

Scenario: 40 long-lived logins on a platform that ties sessions to the originating IP, held for a full year.

The gap between the dedicated option and the shared one is $525.60 over the year. That is the entire decision, and it is a smaller number than most people expect before they run it.

So the real question is not "which is cheaper", it is "how many account recoveries does $525.60 buy". Only you know what re-establishing one burned account costs your team once you count the operator's time, the re-verification, and the gap in whatever the account was collecting. Take your own figure and read it off this table:

Each row is $525.60 divided by the recovery cost, rounded up to a whole incident. If your honest estimate of a single recovery is in the hundreds, and you expect co-tenant trouble more than twice in a year, the shared tier is not saving you anything. That is the arithmetic, and it is why dedicated wins nearly every account-holding scenario despite looking five times more expensive per address.

Flip the scenario and the answer flips with it. For read-only regional checks where nothing persists, a burned address costs you one retry, the recovery cost is close to zero, and no number of incidents justifies paying $525.60 more.

Four cases where we would spend the money on the shared tier and put the difference somewhere more useful.

Read-only geo verification. Checking what a page shows in a given country, where the residential ASN is the requirement and no session survives the request. A co-tenant can cost you a single blocked fetch, which you retry elsewhere. This is the same shape of work covered in our post on geo-pricing audits.

Proving the ASN matters at all. Before committing to dedicated addresses for a year, buy twenty shared ones for a month and establish that a residential ASN actually changes the target's behaviour. Quite often it does not.

Short campaigns. If your work finishes in six weeks, the reputation you would have built on a dedicated address has no time to pay off.

High address counts with low per-address value. If your design needs 500 exits and each one carries a small, replaceable amount of state, the co-tenant risk is spread thin and the price difference at that count is substantial.

Three situations where neither tier is the right purchase, which is worth saying plainly given that we do not sell ISP proxies and have no reason to push you toward them.

The target does not check the ASN. Test this before you assume it. Plenty of sites that "block proxies" are actually blocking a TLS fingerprint, a header order, or a request rate, and will serve a plain datacenter IP happily once those are fixed. Our comparison of ISP proxies and datacenter proxies sets out how to tell the two failure modes apart.

The target blocks static addresses regardless of ASN. Some platforms treat any address that has been sending machine traffic for weeks as suspect, whatever network it sits on. Exclusivity does not help here, and rotating residential is the honest answer. See ISP proxies vs residential proxies for that trade-off.

It is high-volume read-only collection. Holding addresses is the expensive way to solve a problem that a managed endpoint solves per request. The SparkProxy Scraping API gives 1,000 free credits with no card, which is enough to find out. A plain fetch through the rotating pool costs 1 credit and a rendered one costs 5. Routing through the residential pool with premium_proxy=true replaces that base rather than adding to it, at 10 credits plain and 25 rendered, and the only true add-ons are country_code, stealth, js_scenario and screenshot or PDF output at 5 credits each. A residential rendered fetch with country targeting is therefore 30 credits, not 35.

What SparkProxy does sell is datacenter proxies and that API: 1M+ datacenter IPs across 80+ countries including 50,000+ US addresses, reached through gateway.sparkproxy.io on port 11000 for HTTP and HTTPS, 11002 for sticky sessions and 13000 for SOCKS5, on plans from Starter at $75 a month for 100 threads to Plus at $440 for 1,000 threads, all unlimited bandwidth with 30 days validity. If you need ISP addresses, buy them from a vendor who sells them, and our ISP provider comparison is written on that basis.

Two weeks and under $50 of spend will tell you more than any comparison table, including this one.

1. Buy both tiers from one vendor. Five shared addresses and two dedicated, same country, same product family. One vendor keeps the network constant so the only variable is sharing.
2. Run an identical probe on all seven. Every hour for fourteen days, fetch the same known page on your actual target and log status code, response size and latency. Keep the request pattern identical across all seven addresses.
3. Look for blocks that do not correlate with your own behaviour. On a dedicated address, failures should track your request rate. On a shared one, you will see challenges appear during hours when your probe did nothing unusual. That divergence is the co-tenant tax, measured on your target rather than in the abstract.
4. Time a replacement. On day ten, tell the vendor one dedicated address has stopped working and ask for a swap. Record what evidence they demand and how long the new address takes to arrive. Our post on how long proxy IPs last sets out how often you should expect to need this.

Step four is the one almost nobody runs, and it is the one that separates vendors. Two providers can publish the same per-IP price while one replaces a burned address in an hour and the other takes a week and a screenshot of a blacklist you are not on.