When residential proxies first appeared as a commercial product around 2012, the dominant pricing model was a “port” or “channel”. You bought a plan that gave you, say, 100 ports. Each port was a single concurrent connection. If you needed more concurrency, you bought a bigger plan.
That model fit the market for a while. It mapped to how customers thought about scale at the time. It made invoices predictable. It worked well enough that most of the industry kept it through the mid-2010s.
It also stopped making sense, gradually, then suddenly.
Why ports made sense at first
The early use cases for residential proxies were narrow. Sneaker bots, ticketing bots, niche scrapers. Workloads where you ran a known number of parallel sessions and you needed each session to look distinct. A port-based plan was a direct mapping to that mental model, one port, one session, predictable behavior.
The proxy networks themselves were small. Pools of a few hundred thousand IPs, mostly concentrated in a handful of countries. Operators couldn’t realistically support unlimited concurrency, so they exposed concurrency as a billable axis.
And ports were defensible economically. A customer with 100 ports was a meaningfully different size from a customer with 10. Pricing tracked the customer’s apparent scale.
Why ports broke
Three things happened, roughly in parallel, that made port-based plans the wrong primitive:
Pools got bigger by an order of magnitude. The residential proxy market consolidated and scaled. The leading providers moved from a few hundred thousand IPs to over 100 million. Concurrency stopped being a meaningful constraint at the network level. A customer asking for 1,000 simultaneous connections wasn’t asking for anything the network couldn’t trivially deliver.
Workloads diverged from the port mental model. Modern data collection isn’t 100 parallel scrapers running steadily for a month. It’s bursty fan-out, a thousand requests at 9 AM Monday for a price refresh, then near-zero traffic until next Monday. Or it’s an event-triggered scrape, a webhook fires, you need 50 concurrent requests for 90 seconds, then back to zero. Static port allocations were a poor fit for either shape. You either over-provisioned (paid for idle capacity) or got throttled.
Customers stopped wanting to think about it. The actual question buyers care about is “how much data can I move through this?” Bandwidth is the natural unit. It scales with the workload. It maps to invoices that look like AWS, predictable per-unit cost, total cost determined by actual usage.
The transition
The shift to bandwidth pricing took most of the residential proxy market about five years (2019–2024). Not because anyone was hiding it. Pure inertia, large customers had multi-year contracts on port plans, migration plans were painful to write, and most providers wanted to keep both models alive long enough to give legacy buyers an off-ramp.
By 2024, every serious residential network in the industry offered bandwidth-based pricing as the default. Shifter included. We kept the legacy port plans available for customers who had built their stack against them, that legacy commitment still stands, but everything new was bandwidth-priced.
What bandwidth pricing actually changes
A few things become easier and one thing becomes harder.
Easier, buying. “I need to scrape ~200 GB of pages per month” is a sentence a customer can write before they’ve built anything. “I need 80 ports” is a sentence that requires running the workload first to find out.
Easier, forecasting. Bandwidth scales linearly with the data you’re moving. Doubling your scrape frequency doubles your bandwidth. Triple it, you triple it. No threshold cliffs, no concurrency bottlenecks at peak.
Easier, billing. One number on the invoice. No “you used 110% of your concurrency but only 40% of your traffic, here’s the math.” Customers know what they’re paying for.
Harder, discipline. Bandwidth is an open faucet. A misconfigured scraper that downloads full page assets when it only needed the HTML body can quadruple bandwidth without producing any extra useful data. With ports, that scraper would have hit the concurrency wall and the operator would have noticed. With bandwidth, it just runs up the bill until someone audits the egress.
The mitigation is to be deliberate about request shapes. Don’t fetch images when you only want product data. Use Accept-Encoding: gzip (most clients do this by default, but worth checking). Strip query parameters that trigger asset bundles. Pull HEAD before GET if you only need response codes.
Where the industry goes from here
Bandwidth pricing has won the residential market. The next axis of differentiation isn’t pricing model anymore, it’s what’s bundled, what features come for free, and what counts against the bandwidth meter.
A few examples of where providers are differentiating in 2026:
Sticky session pricing. Some providers count sticky-session bandwidth at a premium. We don’t, sticky sessions are billed at the same rate as per-request rotation.
Geo targeting as a feature. Some providers price city-level or ASN-level targeting as an upgrade. We don’t, all geo precision is included on every plan.
Concurrency caps. Some providers still impose concurrency limits on top of bandwidth quotas. We don’t, unlimited concurrent connections on every plan.
API and dashboard. Now that bandwidth is the meter, customers want visibility. How many GB per endpoint, per geo, per session, at hour granularity. That’s table stakes now, not a premium feature.
For buyers, the lesson is the same it’s been for any infrastructure category: read the fine print on what’s billed and what’s bundled. The headline rate per GB is meaningless if half the features you’ll actually use trigger surcharges.
Looking ahead
Bandwidth pricing isn’t the final form. As workloads continue to shift toward AI agents that make a small number of high-value requests rather than scrapers that fan out across millions of low-value ones, request-level or success-rate-level pricing models will probably re-emerge for specific verticals.
For now, though, bandwidth has settled as the right unit. The per-port era served its purpose, and it’s over.