Bandwidth-Based Pricing: Examples & Companies

What is it

Bandwidth-Based Pricing is a billing unit where customers are charged per gigabyte of data transferred out of the platform.

It is the oldest variable-cost meter in cloud infrastructure, and it survives because bandwidth maps almost perfectly to the vendor’s own marginal cost: every byte served downstream costs the provider money in transit and peering. Wherever data leaves a platform at scale — a deployed website, a GPU instance pulling model weights, a vector database returning query results, a real-time voice stream, a proxy network relaying scraped pages — bandwidth becomes a material line on the bill. Ten companies in the corpus meter it, and they fall into two camps: platforms that bundle an allowance and charge per-GB overage, and web-data vendors that treat per-GB traffic as a primary product meter in its own right.

In the platform camp, Vercel bills outbound “Fast Data Transfer” at $0.15/GB beyond the 1 TB included on Pro, Upstash charges $0.03/GB on Redis and Vector once the 200 GB Pay-as-You-Go allowance is exhausted, Chroma meters network egress at $0.09/GiB as one of four usage dimensions, and LiveKit bills real-time data transfer at $0.10–$0.12/GB above each tier’s allotment. The GB here prices simple datacenter egress, so the rate stays low.

In the web-data camp, the same unit prices something very different: Bright Data and Oxylabs bill rotating residential proxies at $2.50–$8/GB, because that GB buys access to a rotating real-IP network that defeats anti-bot defenses, not raw transfer. ZenRows, Apify, Browserbase, and Vast.ai round out the set, each metering bandwidth alongside their headline unit (results, compute units, browser-hours, and GPU-hours respectively).

How it works

Bandwidth is metered by measuring bytes transferred — almost always egress (data leaving the platform), occasionally split into external versus internal transfer. The vendor sets an included quota per plan tier, then charges a per-GB (or per-TB) rate for everything above it. The two structural decisions that define a vendor’s bandwidth pricing are how generous the included quota is and how steep the overage rate is relative to the quota’s implied value.

Unit math: A Vercel Pro site serving 3 TB/month pays the $20 base plus (3,000 GB − 1,000 GB included) × $0.15/GB = $300 in bandwidth overage — fifteen times the base fee, which is why bandwidth dominates a real bill.

Unit math: A scraping job pulling 50 GB through Oxylabs residential proxies at the $6/GB Starter rate costs 50 × $6 = $300 — the same dollar figure as 3 TB of Vercel egress, off 60x less data, because the GB prices the proxy network, not transfer.

The bundled-allowance pattern (Vercel, Apify, LiveKit, Upstash, Chroma) keeps small workloads on a flat fee and only exposes per-GB pricing to heavy users. The primary-meter pattern (Bright Data, Oxylabs, ZenRows) exposes the per-GB rate from the first byte, then discounts it down a committed-spend ladder. Apify sits in between, drawing data transfer ($0.20/GB external) from the same prepaid balance as compute units and proxies. To see how a bundled allowance plus per-GB overage compounds on a real invoice, model it in the Vercel pricing calculator.

Companies using this

Ten companies in the corpus meter bandwidth in GB or TB — split between platforms that bill per-GB overage above an allowance (Vercel, Upstash, Chroma, LiveKit, Apify, Vast.ai) and web-data vendors that price per-GB proxy traffic as a primary product (Bright Data, Oxylabs, ZenRows, Browserbase).

Patterns observed

• The per-GB rate encodes what the byte buys, not how big it is. The ~250x spread across the corpus tracks one variable — IP type and anti-bot capability — not transfer volume: a datacenter-egress GB and a residential-IP GB are physically the same byte priced for entirely different jobs.

• The bundle-and-overage pattern dominates the platform camp. Six of the ten companies bury bandwidth inside a plan allowance, then meter overage: LiveKit alone steps its included transfer 50 GB → 250 GB → 3 TB across Build/Ship/Scale, joined by Vercel, Upstash, Chroma, Apify, and Vast.ai. The allowance is the marketing number; the overage rate is the real one.

• Web-data vendors run two value metrics side by side. Oxylabs bills residential and mobile proxies per GB ($6 and $7.50/GB) but ISP and datacenter proxies per dedicated IP; Bright Data does the same. The vendor matches the meter to the cost driver — traffic where bandwidth is the cost, IP count where dedicated capacity is.

• Committed spend buys a lower per-GB rate. Bright Data residential drops $8 → $7 → $6 → $5/GB across the $499/$999/$1,999 monthly tiers; Oxylabs residential falls to $2.50/GB at the $2,500/mo Corporate tier; even LiveKit shaves transfer from $0.12 to $0.10/GB between Ship and Scale. The discount ladder is how per-GB vendors reward volume without abandoning the unit.

• On platforms, bandwidth is the silent top line. Chroma flags network egress as “the one buyers underestimate” among its four meters, and Apify draws data transfer from the same prepaid balance as compute. Bundled allowances hide bandwidth until scale, then it dominates.

Counterexamples & variants

The cleanest variant is unit framing: Vast.ai meters bandwidth in $/TB for both upload and download rather than $/GB, with each marketplace host setting its own rate — so the “per-GB” model becomes a per-TB, host-variable, marketplace-floated charge layered on top of GPU $/hr and storage $/GB/hr. It is still bandwidth-based pricing, but the buyer cannot quote a single list rate.

The most instructive counterexample is ZenRows, which once published a standalone per-GB residential-bandwidth slider and then dropped it in favour of “Unlimited Bandwidth,” moving its meter to per-1,000-successful-results on the Scraper API. ZenRows now exposes a per-GB rate only on its Scraping Browser and residential proxies, where the effective rate is derived from each plan’s included GB rather than a published line item. It shows that even within web-data, per-GB is not inevitable: a vendor can re-anchor on outcomes (successful results) and treat bandwidth as a bundled input.

Chroma and LiveKit show a second kind of variant — bandwidth as one meter among several in an open-source-backed usage model. Chroma bills egress ($0.09/GiB) alongside writes ($2.50/GiB), storage ($0.33/GiB-mo), and queries ($0.0075/TiB), so egress is a genuine line item but never the headline; and because the full engine is Apache-2.0, “egressing your data is easy,” which implicitly caps how aggressive the egress rate can be. LiveKit likewise stacks data transfer ($0.10–$0.12/GB) beneath agent-session minutes, WebRTC media minutes, and inference credits — the transfer meter is real but is dwarfed by inference on a typical voice-AI bill. In both cases the open-source escape hatch disciplines the bandwidth rate downward.

Upstash is a variant in the other direction — its bandwidth meter is almost an afterthought ($0.03/GB on Redis/Vector after a 200 GB free allowance, $0.05/GB on QStash/Workflow after 50 GB), because requests and storage carry the pricing. Bandwidth here is a guardrail against abuse, not a revenue lever.

What this means for buyers vs vendors

For buyers

Model bandwidth at your expected scale, not the headline plan price — once you exceed the included TB it is routinely the largest line item on the invoice. Ask exactly what the per-GB rate prices: a residential-proxy GB bundles IP-network access and anti-bot bypass, so comparing it to a hosting or vector-DB egress GB is meaningless. Where a vendor runs per-GB and per-IP side by side, pick the meter that matches your cost driver, and push for the committed-tier rate if your volume justifies it. See the usage invoicing & billing-cycles guide for how per-GB overage lands on an invoice, and the entitlements & usage-grants guide for how included allowances are enforced.

For vendors

Bandwidth-based pricing fits whenever data egress is your real marginal cost — hosting, edge, GPU rental, vector databases, real-time media, and proxy networks all qualify. Decide deliberately between bundling an allowance (keeps small customers on a flat fee, exposes per-GB only at scale, as LiveKit and Vercel do) and metering from the first byte (right when the GB is your primary product, as with residential proxies). It requires accurate byte-level metering and a clear external-vs-internal transfer distinction, as Apify draws. If your value is really an outcome — successful scrapes, served requests — consider whether per-GB is the honest meter or whether ZenRows’ move to per-result better aligns your incentives with the buyer’s, as covered in the introduction to usage-based pricing.

Explore this theme in the knowledge graph →