Can I use this to collect training data for a commercial model?

Yes. What we sell is bandwidth through residential, ISP and mobile exits; what you may collect through it is governed by our acceptable use policy and by the terms and robots directives of the sites you fetch. Those obligations stay with you, and they do not change because a proxy is in the path.

Does unused bandwidth expire?

No. Gigabytes are bought once and stay on the account until they are used. There is no subscription, no monthly minimum and no expiry date, which is why a collection run that pauses for a quarter costs nothing while it waits.

How large an order can I place in one go?

The volume ladder runs to 1 TB in a single checkout, which is also where the per-GB price bottoms out. Larger than that, write to us and we will quote the run rather than have you place ten checkouts.

Which pool should a corpus run use?

Start on ARC Basic — rotating ISP addresses at our lowest price per GB — because most of a corpus is plain HTML from sources that are not heavily protected. Move individual sources up to ARC Fast when you see failures you cannot explain, and to ARC Ultra only for what still fails after that.

Can I split one balance across several crawlers?

Yes. The balance belongs to the account, not to a connection, so any number of workers, containers or machines can authenticate with the same credentials and draw from it. Generate a separate endpoint per worker if you want to keep their sessions apart.

Use case · AI training data

Proxies for AI and LLM training-data collection

Building a corpus means fetching millions of public pages over weeks, not thousands over an afternoon. Bandwidth here is bought once at a volume price and never expires, so a collection run that pauses for a month does not cost you the balance.

Get started See pricing

Pay per GB · No subscription · Prices in EUR, excl. VAT

Illustration. The URLs and values are fictional; the connection string is the real grammar.

€0.75
per GB on ARC Basic, excl. VAT: −21%
at 1 TB (€0.59 / GB): 185+
countries, city targeting: Never
expires — no subscription

The problem

What goes wrong without the right IPs

One IP, one rate limit

A crawl from a single datacentre address hits per-IP limits long before it hits your bandwidth ceiling. Spreading the same volume across a rotating pool turns the limit into a throughput question instead of a wall.

The corpus is regional without you asking

Catalogues, news fronts and help centres serve different text per country. Collect everything from one location and your dataset quietly becomes a dataset about that location.

Runs stop and start

Annotation, legal review and model schedules pause a collection for weeks. A monthly bandwidth subscription bills through the pause; a balance that never expires waits for you.

Budgets are approved per project, not per month

A one-time purchase at a volume price is a line item a finance team can approve once. There is no renewal to forecast and no seat to true up.

Mixed page weights

Plain HTML is a few kilobytes; a rendered page with assets is megabytes. Billing per gigabyte rather than per request means the cheap pages actually stay cheap.

Provenance has to be defensible

Collection that is public, logged-out and attributable to a commercial provider with an invoice is easier to document than a pool of unknown origin.

Setup

Recommended setup

Most corpus work is plain HTML at volume, so start on the cheapest pool and move only the sources that need it.

Plan	Role	Session	Targeting	Price
ARC Basic Rotating ISP	Default for bulk HTML The lowest price per GB we sell, and plain fetching of unprotected pages is most of a corpus run. Route the long tail of sources here first.	Rotating	Country, optional city	€0.75 / GB €0.59 at 1 TB (−21%)
ARC Fast Rotating residential	For sources that need a consumer IP Real residential exits for the sites that treat hosting ranges differently. The steepest volume ladder we offer, which matters at corpus scale.	Rotating, or sticky per source	Country + city	€2.00 / GB €1.12 at 1 TB (−44%)
ARC Ultra Premium residential	For the few that still fail Our largest premium pool. Route only the sources that ARC Fast cannot complete — it is the priciest rotating plan, so it should carry the smallest share.	Rotating	Country + city	€3.50 / GB €2.17 at 1 TB (−38%)

Prices are list prices in EUR, excl. VAT. Volume discounts apply automatically at checkout — full ladder on the pricing page.

Illustration

What a 2 million page collection run costs

An illustration with round numbers — your pages, payloads and retry rate will differ. The arithmetic is the real price ladder.

Assumptions

Pages to collect: 2,000,000
Average page weight: 120 KB of HTML, assets blocked
Retries and redirects: +15% overhead
Pool: ARC Basic, rotating
Bandwidth for the whole run: ≈ 276 GB

On ARC Basic

€179.40for the whole run, excl. VAT

276 GB bought in one purchase at €0.65 / GB (−13% volume rung, applied automatically).

The same 276 GB on ARC Fast: €342.24 at €1.24 / GB. Route only the jobs that need it through the pricier pool.

Bought as one purchase rather than month by month, the whole run lands on a single volume rung — and whatever is left over is still there for the next corpus.

Bandwidth never expires, so a bigger single purchase is only ever cheaper per GB. See the ladder.

Works with anything that takes a proxy URL

Scrapy
Python requests / httpx
Crawlee
Playwright
LangChain loaders
Docker

FAQ

Questions

Something not covered? Write to contact@arcproxies.io.

Get started

Buy a gigabyte and try it on your own targets.

No subscription, no sales call. Code WELCOME15 takes 15% off your first purchase, and whatever you do not use is still there next quarter.

Create your account Questions first? contact@arcproxies.io

Web scraping

A second provider that just works — Scrapy, requests, Crawlee or Playwright, self-serve, no sales call, bandwidth that never expires.

Alternative data

Turn public listings, pricing and hiring signals into a time series your models can trade on — collected on schedule, from the market you are modelling.

Real estate & jobs

Listings and postings are ranked by where the visitor is. Collect them city by city so your index reflects the market, not your server's location.

Price monitoring

See the price a shopper in Berlin, Lyon or Chicago sees. ISP bandwidth at our lowest price per GB for HTML checks, residential where a marketplace is strict.