Bright Data vs Oxylabs vs Decodo: which proxies?

Residential proxies are billed by the gigabyte, and almost everybody underestimates how many gigabytes their scraper will use. That single fact decides more about what a project costs than which provider you pick, so it is worth getting straight before comparing anybody's rate card.

A page's HTML is usually tens of kilobytes. The same page rendered in a headless browser — images, fonts, stylesheets, analytics, video thumbnails — is routinely two to five megabytes. At residential rates that is the difference between a job costing pennies and the same job costing tens of pounds, for identical results.

The three kinds, and when each is right

• Datacentre proxies. Addresses belonging to hosting providers, sold by everyone here including Bright Data and Oxylabs. Cheap, fast, and obvious to anyone looking — a site that cares can identify and block the whole range. Right for targets that do not block aggressively, which is more of them than the proxy industry's marketing implies.
• Residential proxies. Addresses belonging to real home connections, routed through consumer devices. Expensive, slower, and much harder to distinguish from an ordinary visitor. Right when datacentre addresses are being blocked and you have confirmed that is the actual problem.
• Mobile proxies. Addresses from carrier networks. The most expensive by a distance, and genuinely hard to block because carriers put thousands of real users behind one address. Reserve these for targets where nothing else works.

The order matters. Start at the top and move down only when you have evidence, because each step up roughly multiplies the cost.

Where the providers actually differ

The unglamorous conclusion is that for most jobs the mid-market providers do the same work for considerably less. The premium providers earn their price in two specific situations: targets with serious anti-bot defences where network size genuinely decides success rates, and organisations where somebody will eventually ask for documentation about where the addresses came from.

How residential networks get their addresses

This is the part worth understanding before you sign anything, because it is where the category's reputational problem lives.

Residential traffic is routed through real consumer devices. Those devices are enrolled either by paying people directly — the clearest case — or, far more often, through an SDK embedded in some other application: a free VPN, a mobile game, a browser extension. The user agreed to something when they installed it, and how clearly that something explained "your connection will be used to route other people's traffic" varies enormously between providers.

The larger providers now publish their sourcing practices and run compliance programmes, which is a genuine improvement on where this market was a few years ago. The cheapest end of the market mostly does not, and the price difference is partly explained by that.

If you are buying for a company, this is a due-diligence item rather than a philosophical one: ask how addresses are sourced, ask for the consent language shown to those users, and keep the answer. If a provider cannot answer, that is the answer.

Spending less by scraping better

Because billing is by the gigabyte, the cheapest optimisation is almost never switching provider. It is sending and receiving less.

1. Do not render unless you must. A headless browser downloads everything a real browser downloads. If the data is in the initial HTML — and surprisingly often it is, or it is in a JSON endpoint the page calls — fetching that directly costs a fraction of rendering the page.
2. Block images, fonts, media and analytics when you do render. One request interceptor rule typically cuts page weight by most of it, and nothing you wanted was in a web font.
3. Find the API. Many sites load their content from a JSON endpoint the page itself calls. Watching the network tab for ten minutes can replace a megabyte of rendered page with a few kilobytes of structured data, and it is more reliable to parse.
4. Cache aggressively and re-request nothing. Store what you fetched, deduplicate your queue, and resume rather than restart. A crawler that re-fetches 30% of its pages after a crash is paying 30% more than it needs to.
5. Route by difficulty. Send the easy targets through datacentre proxies and reserve residential for the ones that actually block you. Most crawls have a long tail of undefended pages paying premium rates for no reason.

Teams that do these five things routinely cut proxy spend by most of it, which is a larger saving than any provider's discount.

What signup is actually like

Not instant, and this surprises people. The reputable providers run know-your-customer checks — company details, intended use, sometimes a call — because the alternative is their network being used for credential stuffing. Budget a day or two before you can send traffic, and expect to describe your use case in writing.

That friction is a feature rather than a bug: a provider that lets anyone start scraping anonymously in ten minutes is a provider whose address pool is being used for things that will get it blocked, which becomes your problem in the form of worse success rates.

The legal line, stated plainly

We are not lawyers and this is not advice. What can be said usefully: the proxy is the least interesting part of the legal question. What matters is what you collect, what you do with it, and whose terms you agreed to. Public data, personal data and copyrighted data are three different conversations, and the jurisdiction you and the target are in changes all of them.

A proxy provider's compliance documentation covers how it sources its addresses. It does not transfer to you any permission regarding what you collect through them, and no provider claims otherwise in its own terms.

What to pick

• A small project, a target that does not block much. Datacentre proxies from any reputable provider. Do not pay residential rates until you have measured a block rate that justifies them.
• Mid-size work, your own money. A mid-market residential provider with a low minimum. The per-gigabyte saving against the big two is the largest single lever you have.
• A company where legal will ask. Bright Data or Oxylabs, for the documentation as much as the network.
• A target that defeats everything else. Mobile proxies, for that target only, routed separately so the rest of the crawl does not pay for it.
• You want a clean IP for yourself rather than scale. A proxy is the wrong product — see our VPN comparison, or self-host on a cheap server with our VPS setup guide.

Whichever you choose, the machine running the crawler matters less than people think — this is an I/O-bound job, not a CPU-bound one. A small server is fine, and our breakdown of what a cheap VPS actually costs covers the sums.