Proxy pricing has traditionally been designed for human teams running predictable scraping operations. Monthly subscriptions, bandwidth-based billing, tiered plans with overage fees — all of these assume you know how much you’ll use before you start. AI agents break this assumption completely. Their usage is bursty, unpredictable, and driven by runtime decisions. Pay-per-request pricing is a fundamentally better model for this reality.

How Traditional Proxy Pricing Works

Most residential proxy providers offer some variation of these models:

Bandwidth-Based Plans

You buy a fixed amount of bandwidth (e.g., 5 GB, 20 GB, 100 GB) per month. Pricing per gigabyte typically decreases at higher tiers.

Problems for AI agents:

It’s nearly impossible to predict bandwidth usage when your agent decides at runtime which pages to visit and how many.
Large pages (image-heavy, JavaScript-heavy) consume bandwidth disproportionately, even if the useful data is a few kilobytes.
Unused bandwidth at the end of the month is wasted money.
Overages are often charged at punitive rates.

Subscription Tiers

Fixed monthly fees that include a set number of requests, IPs, or features. Higher tiers unlock geo-targeting, sticky sessions, or premium IPs.

Problems for AI agents:

Your agent might process 500 requests on Monday and 50,000 on Tuesday. A subscription sized for the peak wastes money on quiet days. One sized for the average fails on busy days.
Features locked behind higher tiers (geo-targeting, API access) are often essential, not premium.
Month-long commitments don’t align with project-based or experimental agent development.

Credit-Based Systems

You buy credits in advance and spend them on requests. This is closer to pay-per-request, but credits often expire, come in fixed bundles, and involve traditional payment methods that add friction.

The Pay-Per-Request Model

Pay-per-request pricing is exactly what it sounds like: you pay a fixed amount for each proxy request you make. No monthly commitment. No bandwidth tracking. No tier selection.

How It Works in Practice

Monthly Requests	Cost at $0.001/request	Equivalent Monthly Plan
1,000	$1.00	$30-75 (typical minimum plan)
10,000	$10.00	$75-150
100,000	$100.00	$200-400
1,000,000	$1,000.00	$800-2,000

At low volumes, pay-per-request saves dramatically compared to the minimum subscription most providers require. At high volumes, it’s competitive with mid-tier subscription plans. And critically, you never pay for requests you don’t make.

Why It Aligns with AI Agent Usage

AI agents have fundamentally different usage patterns than traditional scraping operations:

Bursty workloads. An agent might sit idle for hours, then fire off 10,000 requests when a user triggers a research task. Pay-per-request handles this naturally. You pay for the burst and nothing for the idle time.

Unpredictable volume. You can’t forecast how many requests your agent will make next month because it depends on user behavior, task complexity, and runtime decisions. With per-request billing, this unpredictability is not a problem.

Experimentation. When developing an agent, you might test hundreds of different approaches, each with different request patterns. A subscription penalizes experimentation by charging the same whether you use the capacity or not.

Multi-agent systems. If you’re running multiple agents, each with different proxy needs, per-request pricing lets you share a single balance across all of them without allocating separate subscription plans.

The USDC Advantage

Pay-per-request becomes even more powerful when the payment mechanism is designed for it. USDC (a dollar-pegged stablecoin) on Base L2 offers several advantages over traditional payment methods for proxy billing.

Micropayments That Actually Work

A $0.001 proxy request is a micropayment. Traditional payment systems struggle with this:

Credit cards charge $0.30 + 2.9% per transaction. You can’t economically charge $0.001 per request with credit card rails.
Invoicing requires minimum thresholds, payment terms, and accounting overhead.
Prepaid credits work but require up-front commitment and often expire.

USDC on Base L2 has transaction fees under $0.01, making it economically viable to settle payments as small as fractions of a cent. This means the pay-per-request model can be implemented as true pay-per-request, not “buy credits in advance and hope you use them.”

No Currency Risk

For international developers, USDC eliminates currency conversion. A developer in Brazil, Germany, or Japan pays the same $0.001 per request without exchange rate fluctuations eating into their budget.

Programmable Payments

AI agents with on-chain wallets can fund their proxy balance autonomously. The agent can:

Check its USDC balance.
Determine it needs more proxy credits.
Transfer USDC to the proxy provider.
Continue making requests.

No human needs to log into a billing dashboard, enter a credit card number, or approve an invoice. The entire funding flow is machine-to-machine.

Transparent Settlement

On-chain payments are auditable. You can verify exactly how much USDC was transferred, when, and to which address. For organizations running multiple agents, this transparency simplifies accounting and cost allocation.

Cost Analysis: Real Scenarios

Let’s walk through cost comparisons for three common AI agent scenarios.

Scenario 1: Research Agent (Low Volume)

An AI research assistant that scrapes 5-10 pages per user query, handling about 200 queries per day.

Daily requests: ~1,500
Monthly requests: ~45,000
Pay-per-request cost: $45/month
Typical subscription cost: $150-300/month (the minimum plan that includes enough requests)

Savings: 70-85%

Scenario 2: Price Monitoring Agent (Medium Volume)

An agent that monitors 5,000 product pages across 50 retailers, checking prices every 4 hours.

Daily requests: ~30,000
Monthly requests: ~900,000
Pay-per-request cost: $900/month
Typical subscription cost: $800-1,500/month

Comparable cost, but with no commitment and exact usage billing.

Scenario 3: Data Pipeline (Variable Volume)

A data enrichment pipeline that processes batches of varying size. Some days it processes 1,000 records, others 100,000.

Monthly requests: 50,000-2,000,000 (highly variable)
Pay-per-request cost: $50-2,000/month (scales with actual usage)
Subscription cost: $1,500-3,000/month (must be sized for peak)

Savings: 30-70% in most months. The subscription must be sized for the 2M-request peak, so you pay the peak price even in months with only 50,000 requests.

The Hidden Costs of Subscriptions

Beyond the direct pricing comparison, subscriptions carry hidden costs that impact AI agent developers:

Vendor Lock-In

Annual commitments and volume discounts create switching costs. If a better provider appears, you’re locked in. Pay-per-request means you can switch or split traffic across providers at any time.

Capacity Planning Overhead

Someone on your team has to monitor usage, predict future needs, and upgrade or downgrade plans. This is toil that adds no value to your product. With per-request pricing, capacity planning is eliminated.

Overage Penalties

Exceed your plan limit and you pay 2-5x the normal per-unit cost. For an AI agent with unpredictable usage spikes, this is a constant risk. Pay-per-request has no concept of overages — every request costs the same.

Wasted Prepayment

On average, subscription users leave 20-30% of their allocated capacity unused each month. That’s money paid for nothing. It’s a structural inefficiency of the model that’s invisible on any single invoice but adds up significantly over time.

Why Decentralized Marketplaces Change the Equation

Traditional proxy providers set prices based on their own cost structure, margin targets, and competitive positioning. A decentralized marketplace like RentaTube creates a different dynamic.

Host Economics

In the RentaTube marketplace, residential proxy hosts earn 90% of each request payment. This means:

At $0.001 per request, the host earns $0.0009.
The protocol takes a 10% fee for infrastructure and coordination.

This high host share incentivizes more participants to join the network, which increases the IP pool size, geographic coverage, and overall reliability.

Price Discovery

In a marketplace, pricing can adjust based on supply and demand. Countries with many available hosts might see lower per-request costs. Scarce regions might command a premium. This is more efficient than a single provider setting uniform global pricing.

No Middleman Markup

Traditional providers buy bandwidth from residential IP sources, add their infrastructure costs, add their margin, and sell to you. Each layer adds cost. A decentralized marketplace connects you more directly to the IP source, reducing the layers of markup.

Getting Started with Pay-Per-Request

If you’re building AI agents and want to try pay-per-request proxy pricing, the barrier to entry is intentionally low with RentaTube:

Register with an Ethereum wallet signature at https://api.rentatube.dev/api/v1 — no email, no KYC.
Fund your balance with any amount of USDC on Base L2. Even $1 gets you 1,000 requests.
Make requests through the REST API or the @rentatube/cli npm package.
Pay exactly $0.001 per request. No subscriptions, no minimums, no overages.

The pricing model should match how your agent works: use what you need, pay for what you use, and stop whenever you want. For AI agents operating in a world of variable workloads and tight iteration cycles, pay-per-request is not just a pricing option — it’s the pricing model that makes sense.