What do unlimited bandwidth proxies actually mean?

Proxy pricing comes in two shapes. Metered plans charge for every gigabyte that passes through the proxy, so the bill follows your traffic. Flat plans charge for a slice of capacity, such as 40 simultaneous connections or 10 ports, and let you push as much data through it as that capacity can carry. “Unlimited bandwidth” is the marketing name for the second model.

People look for unlimited plans after a metered bill surprises them: a headless browser pulled in images, fonts and video, a crawler retried a failing page thousands of times, or a scraping job simply ran longer than planned. The trouble is that “unlimited” can hide its own limits, and per-GB pricing really is the better deal for some workloads.

This page explains how both models work, shows how to estimate your own monthly gigabytes from page size and page count, works out where the break-even sits, and lists the limits that still apply on a flat plan.

Proxy providers bill along one of four axes, and each one rewards a different kind of job.

• Per GB. You pay for traffic. Concurrency and IP count are often generous, because the provider is paid by volume. Good for small or bursty jobs, bad for heavy pages and long-running work.
• Per thread. You pay for the number of connections that can be open at the same moment. Traffic is not counted. Storm’s rotating proxies work this way: 40 threads for $39 a month whether those threads move 5 GB or 5 TB.
• Per port. You pay for gateway ports, and each port carries one exit IP at a time. Storm’s residential plans are sold like this, from 1 port for $19 to 100 ports for $550, with up to 50 threads per port.
• Per IP. You rent specific static addresses. Storm’s private dedicated proxies start at 5 for $10 a month, each with 100 concurrent threads.

The last three are all “unlimited bandwidth” plans. What they limit instead is how much you can do at the same time.

Less than people fear with a plain HTTP client, and far more than they expect with a browser. The HTTP Archive’s 2024 Web Almanac puts the median home page HTML at about 18 KB, while the median full page load is around 2.3 MB on mobile and 2.65 MB on desktop. Real targets vary a lot: product and search pages on large shops often return 100 to 500 KB of HTML because the data is embedded as JSON.

Some worked numbers, before retries:

• API or JSON endpoints, 20 KB each, 500,000 calls: about 10 GB a month.
• Raw HTML, 150 KB per page, 1 million pages: about 150 GB.
• Headless browser, full pages at 2.5 MB, 200,000 pages: about 500 GB.
• Browser with images blocked, roughly 600 KB per page, 200,000 pages: about 120 GB.

Two habits shrink metered bills: ask for compressed responses (Accept-Encoding: gzip) and block images, media and fonts in browser automation. Both are worth doing on a flat plan too, because smaller responses finish faster and free up threads sooner.

Flat pricing is the better deal when traffic is high relative to concurrency. That covers most continuous scraping, price monitoring that runs every hour, SEO rank tracking, crawling whole sites, and anything driven by Playwright, Puppeteer or Selenium, where every page drags in megabytes of assets. It also wins when you can’t predict traffic: an AI agent or a crawler with a bug can burn through gigabytes overnight, and on a flat plan that costs nothing extra.

Per-GB is the better deal when traffic is low relative to concurrency. If you run a one-off job of a few thousand JSON calls, or need hundreds of IPs in parallel for a few minutes a day while moving very little data, a metered plan can be cheaper than buying the threads. Per-GB plans also tend to offer things a flat datacenter plan may not, such as many countries or city-level targeting, so if you need those, cost isn’t the only factor.

A quick rule: run the break-even formula with the real rates you’re quoted. If your monthly estimate is below the break-even and stable, metered is fine. If it’s above, or you aren’t sure, flat is the safer bill.

No bandwidth meter doesn’t mean no rules. On any flat plan, check these:

• Concurrency. Threads, ports or proxies cap how much work runs at once. Going over the cap makes connections fail; it doesn’t buy you more speed. On Storm rotating plans, the thread count covers every machine and script using the plan together.
• IPs at a time. On residential, one port is one exit IP at a time, and that IP changes every 5 minutes. Ten different IPs at once means ten ports.
• Target rules. Search engines are sensitive: on Storm rotating plans, keep them on the Main gateway and to at most a quarter of your threads.
• Acceptable use. Reselling, hacking, torrents and illegal activity aren’t allowed, and outgoing email is blocked (SMTP ports 25 and 587). Torrent traffic is exactly the kind of use that makes some providers’ “unlimited” plans unsustainable, which is why it’s excluded.
• Authentication. Storm rotating and residential plans authorize your IP rather than a username and password, which matters if your job runs from cloud functions or changing IPs.

Because traffic is never billed, you size a Storm plan by parallel work, not data:

• Testing a scraper or a small daily job: the 10-thread rotating package ($14) is meant for testing. It’s too small for browser tools.
• HTTP scraping at a steady pace: 40 threads ($39) or 80 threads ($59). With 2 seconds per request, 40 threads finish roughly 20 requests a second.
• Browser automation: a single tab can hold about 10 connections open, so 4 tabs already fill a 40-thread plan. Look at 150 ($97) or 200 threads ($147).
• Sites that want home IPs: residential ports, one IP per port, up to 50 threads each.
• Logged-in accounts that need the same IP every day: dedicated proxies, one per account.

If you’re unsure, the smallest package of each proxy type has a 24-hour money-back guarantee on your first order, so you can check that your traffic pattern fits before scaling up. The rotating vs static guide covers which type suits which task.