Dedicated Proxy Plans: What You Pay For and When It Pays Off

Buy dedicated when a changing IP breaks something. Logins, allowlists that a partner configured for you, long-lived account sessions, API keys tied to a source address, anything where the address is a credential rather than a transport detail. In those cases there is no break-even to calculate, because rotation does not do the job at any price.

Buy shared, semi-dedicated or rotating when the address is disposable and the only thing you need is more of them. That is almost every scraping workload. Paying the exclusivity premium on a job that throws each address away after forty requests is money set on fire.

The grey zone is the workload that rotates but hits a target with a tight per-IP tolerance. That is where the arithmetic in the break-even section earns its keep, and where the answer flips on one measurement you probably have not taken.

Three things. Vendors describe them in many ways, but the substance is short.

Exclusive use for the term. Nobody else routes traffic through that address while you hold it. Oxylabs states it as "100% Exclusive IPs" on the dedicated tier of their datacenter page. IPRoyal draws the same line explicitly, describing competitors who "create the cheapest datacenter proxies by letting any number of users use the same IP" and positioning their product against that.

A history you control from today forward. This is the real product. A shared address carries whatever the previous tenant did on your target. Exclusivity does not erase the past, it just stops new damage arriving from strangers.

Stability. The address does not change under you mid-session, mid-day or mid-month.

Now the harder list, the things the premium does not buy and which no pricing page says out loud.

It does not buy a clean neighbourhood. Exclusivity is per address, never per prefix. You can hold the only lease on 198.51.100.17 while the other 254 addresses in that /24 belong to customers hammering the same target you are. Reputation systems score prefixes, not just addresses, which is why our piece on subnet proxies matters more to a dedicated buyer than to a rotating one. Ask every vendor how many distinct /24s your order will span. Most will answer. None publish it.

It does not buy immunity from your specific target. An address can be technically healthy, route fine, pass every connectivity check, and still return 403 on the one site you bought it for. Read the replacement policy carefully, because "we replace non-working proxies" and "we replace proxies your target has blocked" are different promises and only the first one is usually in writing.

It does not buy bandwidth, unless the plan says so. Oxylabs sells dedicated datacenter per IP and shared datacenter per GB as separate products with separate meters. IPRoyal bundles 100 GB per proxy per monthly cycle and pools it across the order, so ten proxies carry 1,000 GB between them. Those are materially different deals at the same nominal per-IP price.

The word covers four different things on the market, and the failure mode differs in each.

Semi-dedicated deserves a note, because it is the tier buyers most often get wrong. Oxylabs prices their shared datacenter product at "Max 3 Users per IP", which is the classic semi-dedicated shape, at a sample $1.20/IP against $2.25/IP for the fully dedicated tier. Three tenants is not a small risk if all three are scraping retail, and it is close to no risk if the address is only holding an allowlist entry. We break the tier down further in semi-dedicated proxies, and the head-to-head in shared vs dedicated datacenter proxies covers the behavioural differences.

Read from each vendor's own page on 23 September 2026. Verify before you buy.

That last row is the honest one. SparkProxy's four public plans (Starter $75/mo at 100 threads, Core $140/mo at 250, Boost $240/mo at 500, Plus $440/mo at 1000, all unlimited bandwidth on 30-day validity) sell rotation and concurrency. If what you need is an address that stays yours, buy it from a vendor who sells that, and come back to a rotating plan for the parts of your workload that scale on volume instead of identity. Our dedicated provider checklist is vendor-neutral for exactly this reason.

Two vendors publish both tiers on the same page, which makes the exclusivity premium directly measurable rather than a matter of opinion.

Bright Data, at matching quantities: $2.20 against $1.40 at 10 IPs is a 57% premium, $1.70 against $1.00 at 100 IPs is 70%, $1.30 against $0.90 at 1,000 IPs is 44%. Oxylabs, comparing the sample configurations shown on their datacenter page: $2.25 against $1.20 is an 87% premium, although the two products also differ in pool size and country count, so it is not a clean like-for-like.

Call it 1.5x to 1.9x. That is the number to hold in your head, and here is the uncomfortable way to read it. The premium buys exactly one thing the shared tier lacks: nobody else's request history on your address. If your target does not track per-address behaviour tightly, that history is worth nothing to you, and you have paid roughly 60% extra for a property your target never inspects.

The test is cheap and almost nobody runs it. Take ten shared addresses and ten dedicated ones from the same vendor, point both sets at your real target for an hour at identical pacing, and compare success rates. If the two sets land within a couple of points of each other, buy shared and spend the difference on more addresses. If the shared set collapses, you have just justified the premium with evidence rather than intuition.

This is the section every dedicated-proxy page skips, and it turns on one variable most buyers have never measured.

Call it R: the number of requests a single address survives on your target per day before the status codes change. R is a property of your target, not of your provider. It is not on anyone's pricing page and you cannot infer it from a datasheet.

Everything else follows. For N requests a month you need roughly N divided by 30R addresses. At price P per address, the dedicated bill is:

addresses = N / (30 * R)
dedicated_monthly = addresses * P

Set that equal to a flat concurrency plan priced at F and solve for the volume where they cost the same:

N* = 30 * R * F / P

Worked through with P at $1.70 (Bright Data's published 100-IP dedicated rate) and F at $140 (SparkProxy Core, 250 threads, unlimited bandwidth), the crossover is about 988,000 requests a month at R of 400. Below that the per-IP plan is cheaper because you buy fewer addresses. Above it the per-IP plan loses, because addresses scale linearly with volume and threads do not.

The table below is an illustrative model built on those two published prices, not a measurement from any test we ran. Substitute your own R and your own quoted rate and the shape holds.

Three things fall out of this that are worth more than the table itself.

The tighter your target, the worse dedicated gets. That is the reverse of most people's instinct. Buyers reach for dedicated addresses precisely when a site is aggressive, and aggression is what drives R down, which is what makes per-IP pricing expensive. Hostile targets want more addresses, and more addresses is the one thing the premium tier makes costly.

Retries move the answer. A 20% block rate does not change the per-IP bill directly, because you are renting addresses rather than bytes, but it does change effective R, since failed attempts consume the same per-address budget as successful ones. Our notes on retry and backoff strategies apply to the invoice, not only to the code.

None of this applies when the address is a credential. If you are holding logins or sitting on an allowlist, R is irrelevant, the flat plan is not a substitute at any volume, and you should stop reading the table.

Four failure modes, in the order they actually bite.

Burned addresses, and who pays for the swap. Your target blocks four of your eighty addresses in week two. Under most replacement policies, an address that still routes traffic is a working address, so the swap is on you. Get the replacement trigger in writing before purchase, in the vendor's words, and note whether replacements are capped per month.

Bandwidth sold separately. A per-IP rate with a metered bandwidth product behind it is two bills. IPRoyal's bundled 100 GB per proxy per cycle is the friendlier shape here, and at roughly 150 KB per HTML response that allowance covers on the order of 700,000 plain pages per proxy, which is far more than a single address will survive on a defended target anyway. On a per-IP plan, bandwidth is rarely your binding constraint. Addresses are.

Term lock. IPRoyal's published ladder rewards longer commitments, $1.39 on 90 days against $1.57 on 30. That is an 11% discount for accepting a three-month bet on a target that can change its defences in an afternoon. Take it on a workload you have already run for a quarter, not on a new one.

Buying addresses to get concurrency. This is the expensive mistake. Some buyers add IPs because their throughput is too low, when the actual constraint is simultaneous connections, worker count or DNS. Read concurrent connections in proxies before you enlarge an order, and check where the queue is actually waiting.

R is worth an afternoon. Hold a single address and keep requesting your real target at a realistic pace until the status codes change. The count at which they change sizes every per-IP plan you will ever price.

import time
import requests

PROXY = "http://USER:PASS@gateway.sparkproxy.io:11002"
proxies = {"http": PROXY, "https": PROXY}

ok = 0
for i in range(1, 3001):
    try:
        r = requests.get(TARGET_URL, proxies=proxies, timeout=20)
    except requests.RequestException:
        print(f"transport error at request {i}")
        break
    if r.status_code == 200:
        ok += 1
    else:
        print(f"status {r.status_code} first seen at request {i}, {ok} succeeded")
        break
    time.sleep(1.2)

Two details make the result usable. Pace the loop the way your production job paces, because R measured flat out is not R measured at your real cadence. And run it three times on three different addresses, because the first address you draw may already be damaged and one sample tells you nothing.

Where the gateway rotates per request, as SparkProxy's does by default on port 11000, use the sticky session port 11002 to hold one exit for the duration of the measurement. The IP whitelisting guide covers the authentication side if your runner has a dynamic egress address.

Send these to sales in one message. The speed and specificity of the reply tells you as much as the answers.

1. How many distinct /24 subnets will my order of N addresses span?
2. Under what exact conditions do you replace an address, and is a target-specific block one of them?
3. Are replacements capped per month, and does the clock reset on renewal?
4. Is bandwidth included, metered, or a separate product? If included, how much and is it pooled?
5. Do I keep the same addresses at renewal, or is the order re-provisioned from the pool?
6. What is the concurrency limit per address, and is it enforced or advisory?
7. Which authentication methods are supported, and how many whitelist entries do I get?
8. Can I see the actual addresses before payment, or at minimum their ASN and country?

Question 1 is the one that separates vendors. A provider who can answer it has subnet-level inventory management. A provider who cannot has a list of addresses and a hope. Question 5 catches the renewal surprise that costs teams their allowlist entries at exactly the wrong moment.

Buy the smallest possible order, usually five to ten addresses, and run this before you scale it.

First, confirm the addresses are what the invoice says. Check each one's exit IP and the ASN behind it, and count how many share a /24.

for p in 1 2 3 4 5; do
  curl -s -x "http://USER:PASS@proxy$p.your-vendor.example:8080" \
    "https://api.ipify.org?format=json"
done

Second, run the shared-versus-dedicated comparison from the premium section. One hour, identical pacing, both sets, your real target. Record success rate, median latency and the first status code that is not 200.

Third, measure R on two of the dedicated addresses using the loop above, then price the plan with the break-even formula instead of with the vendor's suggested tier.

If you want to skip the proxy layer entirely for the reachability question, one managed call answers it without any of this setup:

curl -G "https://scrape.sparkproxy.io/api/v1" \
  -H "X-API-Key: YOUR_API_KEY" \
  --data-urlencode "url=https://example.com/account/orders" \
  --data-urlencode "render_js=false" \
  --data-urlencode "premium_proxy=true" \
  --data-urlencode "country_code=US"

Plain fetches cost 1 credit, JavaScript rendering costs 5, and the free tier is 1,000 credits with no card, which is enough to establish whether a target blocks by ASN before you commit to renting addresses from anyone. The response carries X-Credits-Used and X-Duration-Ms, so the cost per successful page is visible from the first call.

Three measurements and a formula beat any vendor's recommended tier, including a recommendation from us. If you also want the broader price landscape across proxy types, how much proxies cost has the full ladder.