Mobile Proxy Farm vs Buying Mobile Proxies

Build a farm if all four of these are true: you need ports in one or two countries where you can legally buy and register SIMs in your own name, you need them continuously for at least twelve months, you have physical access to the site and someone whose job includes touching it, and your workload is stable rather than spiky. Buy ports if any one of those is false. Multi-country coverage is the most common dealbreaker, because a farm is geographically fixed and a second country means a second site, a second carrier relationship and a second set of hands.

Read the table as a veto list, not a score. One "Buy" in the bottom three rows usually settles it.

Four layers, and each one is a place things break.

The radio layer is a modem holding a SIM: a USB LTE dongle, an industrial M.2 module such as a Quectel EC25 or EG25-G on a carrier board, a 5G module like the RM520N-GL, or a rooted Android phone in USB tethering mode. The dongle is the cheapest entry and the least controllable. The module costs several times more and gives you a proper AT command interface, better thermals and a real antenna connector. The host layer is what those modems plug into, a mini PC or a Raspberry Pi with powered USB hubs, and USB 2.0 shares 480 Mbps across the whole bus, so ten modems on one controller share a ceiling rather than running ten independent pipes. The proxy layer, usually 3proxy or Squid bound per interface, turns each modem into an authenticated port with a rotation endpoint that issues the AT command to drop and re-establish the data session.

The exit layer is the one beginners forget. Your farm sits behind your own broadband, which usually means a dynamic IP and often carrier-grade NAT with no inbound ports at all, so you need a small public VPS and a WireGuard tunnel from the farm to it. That VPS is a single point of failure for the whole fleet, and its IP is the address your clients connect to, not the address your traffic exits from.

These are planning ranges from public retail listings, not a quote and not measured data. Price your own basket before you model anything, because modem prices move with chipset supply and SIM plans differ enormously by country.

Electricity is the one figure you can compute exactly instead of guessing. An LTE modem plus its share of the host and hub draws roughly 2.5 W continuously, which is 1.8 kWh a month: $0.27 at $0.15 per kWh, $0.63 at $0.35. Even a 5G module at 5 W only reaches 3.6 kWh. Power is never the reason a farm fails.

Labour is. Budget a floor of about two hours a month for a small farm no matter how few ports you run, because a fleet of five still needs the same firmware updates, the same tunnel babysitting and the same modem that has silently detached. Past roughly twenty ports the effort stops being per-port and becomes per-incident, which is why the labour line behaves like a step, not a slope.

For an LTE-class build, capex lands near $85 per port and recurring hardware and connectivity cost near $23 per port. Those two numbers carry the rest of this post.

SIM supply is the real constraint. Consumer unlimited plans are usually sold with terms that restrict modem, router or commercial use, and they carry tethering and fair-usage limits that a scraping workload hits in days. M2M and business data plans permit the use but are metered and priced accordingly. In a large number of countries, SIM registration requires identity documents per line, which caps how many you can buy in your own name and turns a fifty-SIM fleet into a corporate procurement exercise rather than a shopping trip. Check your jurisdiction and your carrier's acceptable use policy before you order hardware, not after. A farm running on plans that forbid the use is one audit away from being scrap.

Deprovisioning risk concentrates. Fifty SIMs bought from one carrier on one account behave like one customer. If that account is flagged, you do not lose one port, you lose the fleet, and the hardware has no resale value anywhere near what you paid.

Thermals and density. Modules throttle. A 5G module under sustained load will fall back to LTE once it gets hot, which quietly changes the network characteristics you paid for. Antenna isolation, not port count, caps how many radios you can pack onto one shelf.

Attrition. Dongles die. Plan a replacement budget from month one and validate your own rate in month three rather than trusting anyone's published figure, this post included.

Single-site fragility. One power cut, one broadband outage, one landlord, one country. Rented ports spread that risk across a vendor's infrastructure by default, and that is a large part of what the margin pays for.

Public list prices on mobile proxy vendors' own pricing pages, checked September 2026, cluster into two models. Confirm current numbers on each vendor's own site before building a model on them, because these move often.

The per-GB model is the one that surprises finance. Mobile bandwidth is the most expensive bandwidth in the proxy market by a wide margin, so a workload that renders JavaScript and pulls images can burn a month of budget in a week. Measure your bytes per successful record before you sign anything, then multiply by the vendor's rate. If that number is uncomfortable, the fix is usually to download less, not to change vendor.

Dedicated ports are the closer comparison to a farm, since fixed ports you control are exactly what a farm produces. That is the band used in the break-even below.

Assumptions, stated so you can replace them: $85 capex per port, $23 per port per month for SIM, VPS and power, labour at $50 an hour loaded, and a rented dedicated port at $60 a month, the midpoint of the published band above. The 50-port row adds $600 of racking, power distribution and cooling to capex and $150 a month of facility cost.

Three things worth noticing. Payback barely moves with fleet size, because both sides of the equation are close to linear, so "we will save more at scale" is not the argument people think it is. At five ports the saving is $85 a month, under two hours of engineering time, so one bad afternoon erases a month of gains. And the whole table collapses if your SIM cost lands at $35 instead of $20: build monthly at 20 ports becomes $960 against $1,200 rented, the saving drops to $240, and payback stretches past seven months.

Run the table with your own SIM quote first. If you cannot get a written SIM price that permits your use case, you do not have a build option yet, you have a hypothesis.

The cheapest mobile proxy farm is the one you never build, because the target never required a cellular IP. Mobile IPs earn their premium on a narrow set of targets: mobile-first social platforms, app APIs that check the network alongside the device, and sites that treat hosting ASNs as automation by default. Plenty of teams buy mobile because a forum post told them to, then pay 20 to 50 times the datacenter rate for a target that never checked.

SparkProxy does not sell mobile ports, which is exactly why this section is here. Our network is datacenter: over 1 million IPs across 80+ countries including 50,000+ US IPs, unlimited bandwidth on every plan, and thread-based pricing instead of per-GB metering.

Speed is a ceiling under the fair usage policy, not a guaranteed rate, and all four plans run 30 days with unlimited bandwidth. The gateway is gateway.sparkproxy.io: port 11000 for HTTP and HTTPS, 11002 for sticky sessions, 13000 for SOCKS5.

Before committing to either mobile path, measure your own block rate on your own target. Do not trust a published number, including anyone's benchmark post. This script runs the same URL set through whatever upstream you point it at and prints a success rate you can compare across pools:

import requests
from collections import Counter

UPSTREAM = "http://USER:PASS@gateway.sparkproxy.io:11000"   # swap for a mobile port to compare
TARGETS  = ["https://your-target.example/p/%d" % i for i in range(1, 61)]

def probe(url, upstream):
    try:
        r = requests.get(url, proxies={"http": upstream, "https": upstream},
                         timeout=20, headers={"User-Agent": "Mozilla/5.0"})
        return r.status_code
    except requests.RequestException:
        return 0

codes = Counter(probe(u, UPSTREAM) for u in TARGETS)
print(codes)
print("success rate: %.1f%%" % (100.0 * codes[200] / len(TARGETS)))

Run it twice, once through a datacenter gateway and once through a trial mobile port, on the same URLs inside the same hour. If datacenter clears your threshold, stop reading about farms. If it does not, you now have a real gap to price rather than an assumption. The trade-off is covered in more depth in datacenter vs mobile proxies and in the three-way comparison of residential, datacenter and mobile pools.

For rendering and anti-bot handling on top of whichever network wins, the SparkProxy Scraping API accepts your own upstream through own_proxy, so a rented mobile port or your own farm port stays on the path:

curl -X POST "https://scrape.sparkproxy.io/api/v1" \
  -H "X-API-Key: sk-xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://www.sparkproxy.io/",
    "own_proxy": "http://user:pass@203.0.113.10:8001",
    "device": "mobile",
    "render_js": true,
    "format": "json",
    "tag": "farm-test/port-03"
  }'

That API gives 1,000 free credits with no card, enough to run a genuine comparison. A plain fetch costs 1 credit, a JavaScript render 5, a screenshot or PDF 10. Paid tiers start at $49 for 250,000 credits a month with 50 concurrent requests, and go to $99 for 1,000,000 and $249 for 3,000,000.

If you land on buying, these are the questions that separate vendors once the marketing page is closed.

Verify the ASN yourself on day one instead of trusting the country label. A whois on the exit IP should return a mobile network operator, not a hosting company. If it returns a datacenter ASN, you are paying mobile prices for something else, and the IP reputation of that address is not what you were sold.

Farms make sense more often than buy-side marketing admits, in a few specific shapes.

You need one country, you can get compliant SIMs there, and you need many ports for a long time. Fifty ports for two years in a country where you already have an office is a straightforward capex decision, and the vendor margin you avoid is real money.

You need a specific carrier, because some targets behave differently per operator and a vendor pool will not let you pin that. You need session lifetimes measured in weeks with a hard guarantee nobody else has touched the IP, which a farm gives you by construction, though a rented dedicated port usually gives the same thing contractually.

Or you already run hardware. A team with its own racks absorbs a farm far more cheaply than a software-only team, because the failure modes are familiar and the person who fixes them already exists.

If your use case is account-heavy rather than scraping-heavy, the guide to mobile proxies for social automation and app testing covers the session and identity side that decides success there. And if you are weighing 5G hardware for the build, read 4G and 5G mobile proxies before paying the module premium, because on most networks the radio generation changes less about your exit IP than the price difference implies.

Two weeks and a few hundred dollars is enough to decide this on evidence instead of forum opinion.

Days 1 to 3: get a written SIM quote that explicitly permits modem and commercial use, for the number of lines you want, in the country you want. If that fails, the build option is closed and you have saved yourself the rest.

Days 4 to 7: rent two dedicated mobile ports and run the block-rate script above against your real targets, alongside a datacenter gateway on the same URLs. Log bytes per successful record too, since that sets your exposure if you ever move to per-GB pricing.

Days 8 to 11: build one port end to end, tunnel included, from one dongle, one SIM and one Raspberry Pi. The goal is not capacity. It is to learn how long the build takes and how it feels when the modem detaches at midnight, then multiply that friction by your fleet size honestly.

Days 12 to 14: refill the break-even table with your own numbers. If the monthly saving is under about four hours of your loaded engineering cost, buy. That threshold has nothing to do with proxies, it is just what maintenance costs.