The argument this blog keeps making about proxy pricing is that most vendors advertise a rate almost nobody pays, and that we check rather than repeat. An argument like that is worth very little until it survives being pointed at our own records.
On 10 August 2026 we re-read the pricing page of all 20 residential vendors we track, one week after the original capture, and compared each page against what our database said about it. Four records did not survive. Two held figures that were incorrect. One held two figures that were each real and could never apply to the same buyer. One held a price that was real, purchasable, and filed under the wrong product. None of the four looked wrong from the inside, which is the part worth dwelling on.
Four records, as first captured and as corrected
| Vendor | As first captured | As corrected, 10 August 2026 |
|---|---|---|
| IPBurger | $8.17/GB as both entry and floor, recorded as a 1x spread, alongside a $139 Pro plan | $5.90/GB entry, $4.48/GB floor on the $269 Pro plan, a 1.3x spread |
| SOAX | $0.35/GB advertised floor | $0.25/GB floor, needing Tier 3 geography, the $3,000/month Enterprise plan and 15+ TB, a 20x spread against a $5/GB entry |
| Bright Data | $8.00/GB list entry beside a $2.50/GB promotional floor | $8/GB to $5/GB on the list ladder only, a 1.6x spread, promo stored separately |
| Infatica | $0.30/GB recorded as the residential floor | $0.30/GB kept as their advertised claim, recorded as a Shared Datacenter rate that no published residential plan reaches |
Each vendor's own pricing page, read 10 August 2026. The corrected column renders from the price record itself.
IPBurger, the error that ran in our favour
Our record gave IPBurger an entry rate of $8.17/GB and an advertised floor of $8.17/GB. Identical figures mean something specific in this dataset: a vendor charging exactly what it markets, which only two of the sixteen providers in the price study manage. We recorded it as a 1x spread and treated it as a point in IPBurger's favour.
Re-reading their pricing page on 10 August 2026, $8.17 is not on it. Not under residential, not under any other product, not struck through beside a discount. Their published residential ladder has three rungs: $59 for 10 GB, $149 for 30 GB and $269 for 60 GB, which works out at $5.90/GB, $4.97/GB and $4.48/GB. The same record also carried a $139 Pro plan. There is no $139 plan. Pro is $269.
The corrected spread is 1.3x, not 1x. Mild for this market, and still a spread. We had overstated IPBurger's entry rate by 38% and then complimented them for a discipline our own data does not show. The second half of that is worse than the first: a wrong number sits in one cell, but an argument built on a wrong number travels into every page that cites it. The corrected version now runs in our IPBurger comparison.
An error in this direction flatters us. IPBurger left our database pricier than they really are, which makes our own rate look better by comparison. A mistake shaped like that is the last one anyone inside the building thinks to question, and most of why we went looking in the first place.
SOAX, an understated extreme
Our record put SOAX's advertised floor at $0.35/GB. That is a real cell in their pricing matrix, which crosses geography tier against monthly volume: $0.35 is what Tier 2 traffic costs an Enterprise customer in the 15+ TB band, and also what Tier 3 costs an Enterprise customer between 0 and 5 TB. It is a price on their page. It is not their lowest one.
The floor is $0.25/GB, and reaching it takes three conditions at once: Tier 3 geography, the $3,000 per month Enterprise plan, and 15+ TB of monthly usage. Set against a no-commitment entry rate of $5/GB, that is a 20x spread, the widest in the study.
This one ran the other way. We had made SOAX's pricing look less extreme than it is, inside an article whose entire subject was extreme pricing gaps. The related claim we made about them did survive re-checking: a buyer who needs US or UK addresses cannot reach the headline rate at any level of spend, because Tier 1 geography bottoms out at $0.50/GB and still requires Enterprise plus the 15 TB bucket.
Bright Data, two real prices that never met
Our Bright Data record held an entry rate of $8.00/GB beside a floor of $2.50/GB. Both were read off their own pages, and both were correct.
They came from different pricing regimes. The $8.00 is the list rate a new customer pays with nothing signed. The $2.50 was the bottom of the promotional ladder, unlocked by the code RESIGB50, a 50% discount that does genuinely revert after three months. Bright Data runs two ladders at once: at list you move from $8.00 down to $5.00 as you commit, and under the promotion you move from $4.00 down to $2.50. All four of those numbers are real.
No customer has ever faced the pair we recorded. Shopping at list, the best rate available is $5.00. Shopping during the promotion, the entry point is $4.00. Taking the top of one ladder and the bottom of the other produced a spread nobody could experience, which makes it a construction rather than a measurement.
The record now carries the list ladder alone, $8/GB down to $5/GB, a 1.6x spread, with the promotional rates stored as their own regime rather than blended into it. A promotional price is worth publishing. It is not worth publishing in the same row as a standard one.
Infatica, a real price under the wrong heading
The subtlest of the four, and the only category error in the set.
We recorded Infatica's residential floor as $0.30/GB. The number is real and you can buy it: 2,000 GB for $599. On their site it sits under the heading Shared Datacenter IPs. Their residential ladder is a different set of numbers entirely, running from $4/GB down to $2.60/GB. We had filed a datacenter price as a residential one, and that is the kind of error that walks straight through a sanity check, because $0.30/GB is a perfectly plausible residential floor for a vendor marketing aggressively.
There is a reason it happened, and it is the interesting part. Infatica's own navigation dropdown, under Products and Pricing, reads 'Residential IPs from $0.3 per GB'. The vendor attaches that datacenter figure to a residential label themselves. Our record had a source. Having a source is not the same as being right.
The correction was not to delete the number, because it is what Infatica advertises to a residential buyer and a buyer will meet it. It stays in the advertised column, and the record now states what it actually is: a Shared Datacenter rate that no published residential plan reaches. Naming the condition attached to a headline figure is the entire subject of the study this post audits, and we had managed to land on the wrong side of it inside our own database.
What the corrected study says
None of the four changed the finding, which is worth stating plainly. Across the sixteen providers with a verifiable headline, the median gap between the advertised rate and what a new customer actually pays is 2.5x. The widest is SOAX at 20x, a $0.25/GB headline against a $5/GB Tier 1 entry rate. Two of the sixteen have no gap at all: DataImpulse and PacketStream, both advertising $1/GB and both charging it.
What the IPBurger correction did remove is an example we had been leaning on elsewhere, since IPBurger no longer belongs in the small group whose advertised price equals what it charges. Within the study's sixteen that group is unchanged at two, and both of them were always doing it at a real price rather than a symbolic one.
Why the second pass caught what the first did not
All four survived the original capture for the same reason: each was individually plausible. An $8.17/GB residential rate is believable. A $0.35/GB floor is believable for a vendor with an enterprise tier. A $2.50 promotional price beside an $8.00 list price looks like a wide spread, which is exactly what the study was collecting. A $0.30/GB headline reads like the unreachable marketing figure the study exists to document. A wrong number does not announce itself. It sits in the column looking like every other number.
What caught them was not a better first pass. It was a second pass with a specific question attached. Re-reading a pricing page to answer 'what is the lowest rate on this page, and what exactly unlocks it' is a different activity from reading it to fill a field, and it finds different things. Checking is cheap. Re-checking is the part that works.
The direction of the errors matters as much as the count. The IPBurger error made a competitor look more expensive than it is, which flatters us. The SOAX error made a competitor's pricing look less extreme than it is, which does not. Mistakes that scatter are ordinary. A set of errors running consistently in our own favour would be a much worse sign, and a reason to distrust the whole dataset rather than repair four rows of it.
The part that generalises
Every one of these figures passed a first reading because it was plausible in isolation. Plausibility is not verification, and a stored number stops being questioned the moment it is stored. If you keep a price sheet on your own suppliers, the cheapest improvement available is not a more careful first capture. It is reading the same page again a week later with one specific question in hand.
What changed in how the record is kept
Three changes, all of them dull, each aimed at one of these failures.
Every competitor price now carries the date it was captured and the exact condition that unlocks it, written out rather than implied. A figure with no condition attached is the state IPBurger's $8.17 was in, and a condition is the thing a second reader can go and test.
Promotional and standard rates are stored separately instead of collapsed into a single figure. Bright Data's promotional ladder is still recorded. It is recorded as a promotion.
Every product's prices are filed under its own product type, so a datacenter rate can no longer be recorded as a residential one. Where a vendor advertises across that line themselves, as Infatica does, the record says so rather than quietly picking a side.
Our own residential price sits in the same record on the same terms: $0.50/GB from the first gigabyte, falling to $0.44/GB above 2,000 GB+. We publish our own spread in the study for the same reason we publish everyone else's.
Sources
Every figure below was read off the vendor's own pricing page on 10 August 2026, one week after the original capture for the residential price study.
- IPBurger, pricing (the $59 / $149 / $269 residential ladder; no $8.17 rate and no $139 plan anywhere on the page)
- SOAX, pricing (geography tier against monthly volume; the $0.25/GB cell needs Tier 3, Enterprise at $3,000/month and 15+ TB)
- Bright Data, residential proxies pricing (list ladder $8.00 to $5.00; the RESIGB50 promotional ladder $4.00 to $2.50, reverting after three months)
- Infatica, residential proxies (residential ladder $4.00 to $2.60; the $0.30/GB figure belongs to Shared Datacenter IPs at 2,000 GB for $599, while the Products and Pricing dropdown labels it residential)
- DataImpulse, residential proxies (advertised and charged rate identical)
- PacketStream, packetstream.io (advertised and charged rate identical, no published volume discount)