Unlimited Bandwidth Proxies: What You Actually Get

Buy an unlimited bandwidth plan if your traffic is heavy, byte-hungry, and aimed at targets that accept datacenter IPs. Crawling product catalogs, pulling HTML-heavy pages, downloading images or PDFs, monitoring hundreds of sites on a schedule. In those jobs the byte count is unpredictable and large, and a per-GB meter turns every extra kilobyte of page weight into a line item.

Stay on per-GB pricing if your monthly volume is small, your target list is short, or your targets only accept residential and mobile IPs. Under roughly 30 to 50 GB a month, metered residential bandwidth is usually the cheaper purchase and the trust level is higher. The exact crossover depends on your per-GB rate, and the table below lets you calculate yours in ten seconds.

The uncomfortable middle case: you need residential trust and high volume at the same time. Nobody sells that cheaply. If a vendor offers it as unlimited, read the fair usage terms twice before you pay.

Every proxy product meters on at least one of three axes:

• Bytes. You pay per gigabyte transferred. Standard for residential and mobile pools.
• Requests. You pay per call or per credit. Standard for scraping APIs.
• Concurrency. You pay for how many simultaneous connections you may hold open. Standard for datacenter and ISP pools.

"Unlimited bandwidth" removes only the first axis. The other two stay, and they are what you are really buying. A plan advertised as unlimited is a concurrency plan with the byte meter switched off, and its price ladder climbs with thread count, not with traffic.

That is a different purchase, not a discount on the same one. Under per-GB pricing your bill scales with how fat your target's pages are, a variable you do not control. A retailer adding a 400 KB hero image to every product page raises your costs overnight. Under unlimited bandwidth, page weight costs you time, not money, and your bill only changes when you buy more threads. For background on how providers count transferred bytes, what bandwidth means in proxy services covers the accounting.

1. Thread count. The number of concurrent connections you may hold open through the gateway. This is the primary throughput lever on any unmetered plan, because requests per second is roughly threads divided by average round-trip time. It is also the number that sets your price. Concurrent connections in proxies explains how providers count them and where the accounting gets confusing.

2. Speed cap. The maximum aggregate throughput in Mbps. Sold as a ceiling, never a guaranteed rate. SparkProxy's fair usage tiers cap Starter at 25 Mbps, Core at 50, Boost at 100, Plus at 150, Pro at 200, and Pro+ at 250, with custom arrangements going up to 1 Gbps. Treat any published Mbps figure as the top of a range you may not reach on a slow target.

3. Fair usage policy. The document that defines what "unlimited" excludes. Any serious provider publishes one. It is the single most useful page on a proxy vendor's site and almost nobody reads it before paying.

Two secondary limits matter for setup rather than throughput: whitelist slots, which cap how many source IPs can authenticate without credentials, and plan validity, which on SparkProxy's four published plans is 30 days.

Skip the vendor comparison charts and do this on your own invoice. Break-even in gigabytes per month is the flat plan price divided by your current per-GB rate.

SparkProxy's four published unlimited plans are Starter at $75 a month with 100 threads and 5 whitelist slots, Core at $140 with 250 threads and 10 slots, Boost at $240 with 500 threads and 15 slots, and Plus at $440 with 1000 threads and 25 slots. All four are unlimited bandwidth with 30 days validity. Higher Pro and Pro+ tiers exist in the fair usage policy but carry no public price, so ask for a quote rather than assuming a number.

Here is the crossover volume at each plan, for a range of metered rates. Read down the column for the rate you currently pay:

Published residential list prices move constantly and vary by commitment tier, so use the rate on your last invoice rather than a number from a review site. The shape is the point: even at cheap metered rates, a few dozen gigabytes a month flips the decision, and any serious crawl clears that inside a week.

Now run it the other way, as effective cost per gigabyte once you are on the flat plan:

The interesting column is the third. Unlimited pricing is not slightly cheaper at volume, it is a different order of magnitude, which is why bandwidth-heavy crawlers end up on datacenter pools regardless of what the marketing says about residential trust. If you are weighing that trade-off directly, residential vs datacenter proxies sets out where each one actually holds up.

The speed cap converts straight into a volume ceiling. A 25 Mbps cap is 3.125 megabytes per second, about 8.1 TB over a 30 day month if you saturated it every second. No real workload does: targets are slow, retries eat capacity, and nothing runs flat out around the clock. Treat these as arithmetic ceilings, not forecasts:

Plan against a fraction of that. A crawler running eight hours a night at half the cap moves roughly 45 GB a night on Starter, about 1.3 TB a month, still far past the point where metered pricing would have cost more than $75.

The other ceiling is request rate, driven by threads and latency. Requests per second is approximately threads divided by average round-trip seconds:

Latency is the variable you can move without spending anything. Cutting average round-trip from 1.5 s to 0.9 s raises throughput by two thirds on the same plan, cheaper than upgrading. Connection reuse, tighter timeouts, and dropping subresources you never parse all pull in that direction.

This is the part most buyers miss, and it decides whether you should buy more threads or a faster tier. You have two independent ceilings and only one of them binds for your workload. Which one depends on your average response size. The crossover point is the speed cap in bytes per second, multiplied by average latency in seconds, divided by thread count:

Read it like this. Pulling small JSON endpoints averaging 8 KB, you are nowhere near the byte ceiling and every extra thread converts directly into throughput. Pulling full retail HTML pages well above 100 KB, the speed cap binds long before you use all your threads, and buying more threads changes nothing.

The fix is either a higher tier for the cap, or a smaller payload: request compressed responses, and pull the JSON API behind the page instead of the rendered page whenever one exists.

Measure your own average response size first. Log the byte count of every response for one full run, take the mean, and compare it against the row for your plan. That single number tells you which upgrade is worth buying. Most buyers guess, upgrade the wrong axis, and conclude that unlimited was a scam.

Four models are on the market. They price the same capability differently, and the right one is a function of your traffic shape rather than of which is objectively best.

The credit model deserves a second look if bytes are your problem, because a credit is charged per request regardless of payload size. On SparkProxy's Scraping API a plain fetch costs 1 credit, a JavaScript render 5, a screenshot or PDF 10. Starter at $49 buys 250,000 credits with 50 concurrent requests, Growth at $99 buys 1,000,000 with 100 concurrent (roughly $0.099 per thousand plain fetches), Pro at $249 buys 3,000,000 with 200 concurrent, and Scale at $599 buys 8,000,000 with 400 concurrent. There are 1,000 free credits with no card if you want to price your own workload first. The trade-off against running your own rotation is laid out in web scraping API vs self-managed proxies.

A useful way to hold all four in your head: unlimited bandwidth is byte-blind and concurrency-metered, credits are byte-blind and request-metered, per-GB is byte-metered, per-IP is identity-metered.

An unlimited plan without a published fair usage policy is one the vendor can reinterpret whenever your traffic becomes inconvenient. Read it before you pay and look for four things.

A named speed ceiling per tier. If the policy does not state Mbps by plan, the provider is reserving the right to shape your traffic to whatever they feel like on the day. A published ladder, the way SparkProxy publishes 25 through 250 Mbps across tiers with custom arrangements up to 1 Gbps, is a commitment you can hold them to.

Whether the cap is a ceiling or a rate. It is always a ceiling. Any vendor implying otherwise is describing a dedicated line, not a shared proxy pool.

What counts as abuse. Prohibited targets and protocols, resale restrictions, and behavior that degrades the pool for other customers. This is where "unlimited" gets its real boundaries, and the list is usually short and reasonable. Read it against your actual use case.

What happens when you cross a line. Throttling with notice is survivable. Immediate termination with no refund is a different risk profile, and worth knowing before you build a production pipeline on it.

1. No published speed cap anywhere. Unlimited bytes with an undisclosed rate limit is a blank cheque in the vendor's favor.
2. A request cap buried in the terms. Unlimited bandwidth beside a monthly request quota means you are on a request plan wearing a bandwidth plan's label.
3. One thread per port. Some cheap listings sell ports, not concurrency. At one connection per port, unlimited bytes are unreachable by construction.
4. Unlimited residential or mobile bandwidth at a datacenter price. The provider pays for that bandwidth upstream. If the arithmetic does not work at their price, ask how the pool is sourced and expect a specific answer.
5. A pool of a few thousand IPs. Unmetered traffic through a small pool burns IP reputation fast, and you feel it as rising block rates rather than as an invoice.
6. No fair usage document at all. See the previous section. This is the strongest single signal.
7. Unlimited for month one, metered after. Trial pricing dressed as a plan. Check what renewal costs before you migrate onto it.

One related check: whether the plan is shared or dedicated at the IP level, since a shared unmetered pool behaves very differently under load. Shared vs dedicated datacenter proxies covers what changes.

A price monitoring team tracks 40,000 retail product pages twice a day. That is 80,000 requests a day, and their own logs put the average response at 140 KB after compression.

Bytes. 80,000 times 140 KB is about 11.2 GB a day, roughly 336 GB a month. On metered residential at $4 a GB that is around $1,344 a month. Starter at $75 covers the same volume with enormous headroom.

Speed. 336 GB a month sits well under Starter's 8.1 TB arithmetic ceiling, so total volume is not the binding constraint.

Threads. At 140 KB average they are above Starter's 47 KB crossover, so the 25 Mbps cap binds before the 100 threads do. At 3.125 MB/s the ceiling is about 22 requests per second, so 80,000 requests need roughly one hour of saturated transfer. Add retries and a sane request pace and call it three to four hours per pass. Two passes a day fits comfortably.

Verdict. Starter is the right buy, and the upgrade trigger is not volume growth but window compression. If the team later needs both passes inside a single hour, they move to Core for the 50 Mbps cap, not for the extra threads.

That is the whole method: bytes to pick the model, speed cap to check the window, threads to check the request rate, then buy the smallest tier that clears all three. For the criteria around throughput, what to evaluate when selecting a proxy service and datacenter proxy pricing models cover the rest.

Connection details do not change with the tier. SparkProxy routes through gateway.sparkproxy.io on port 11000 for HTTP and HTTPS, port 11002 for sticky sessions, and port 13000 for SOCKS5, across more than 1,000,000 datacenter IPs in 80 or more countries, including over 50,000 in the United States.

• Your volume is genuinely small. Under about 30 GB a month at typical metered rates, you are paying for headroom you will never touch.
• Your targets reject datacenter IPs outright. Unmetered bytes through a pool blocked at the ASN check is a cheap way to collect error pages. Test against your real targets first.
• You need one durable identity per account. That is a per-IP purchase, static ISP or dedicated datacenter, priced on identities rather than throughput.
• You need deep coverage of one small country. Pool depth in that market matters more than the billing model.
• You do not want to own rotation, retries, and unblocking. The request-metered API fits better even though it looks pricier per call, because the engineering time it removes is the real cost you pay today.

The honest summary: unlimited bandwidth proxies are the cheapest way to move a lot of bytes through datacenter IPs, and nothing more. They do not buy trust, geo depth, or speed you did not pay for. Price the byte axis first, check that the concurrency and speed ceilings clear your busiest hour, and the decision makes itself.