How Much Does Browser Automation Infrastructure Cost?

The sticker price of "run Playwright yourself" is zero — it's open source. The actual cost shows up in three places most teams underestimate until they've been running it for a few months: proxy fees, engineering time, and downtime when detection changes break the stack.

The self-hosted stack, itemized

• Compute: Running headless Chrome instances at any real concurrency needs meaningfully-sized VMs — Chrome is not light. Budget for this scaling roughly linearly with concurrent sessions.

• Residential proxies: Datacenter proxies are cheap but get flagged quickly on any site with real defenses. Residential proxy providers typically charge per GB of bandwidth, and browser sessions with images and scripts loading are bandwidth-heavy — this line item grows faster than teams expect.

• CAPTCHA solving: A per-solve or subscription cost, layered on top, for the fraction of requests that still get challenged despite the proxy and stealth patches.

• Engineering time: This is the largest hidden cost. Detection vendors update signals continuously; someone needs to notice when the stack starts failing, diagnose why, and patch it. For teams without a dedicated infra engineer, this tends to land on whoever built the agent in the first place — time not spent on the actual product.

None of these line items have a stable industry-wide price — compute, proxy bandwidth, and CAPTCHA-solving rates all vary by provider and volume, and shift often enough that a specific number here would be stale within months. The point isn't the exact figure, it's that all four are real, ongoing costs that "just use Playwright, it's free" glosses over.

What managed, real-device infrastructure changes

Paying for managed infrastructure doesn't eliminate the underlying costs above — it bundles and shifts them. dat.ai's pricing is $99/month for 20 concurrent browsers on the Core plan, with a custom plan for teams scaling to 1M+ concurrent sessions. What that price is actually buying:

• No separate proxy line item — residential IPs across 100+ countries are the default, not an add-on.

• No CAPTCHA-solving subscription — real devices with genuine fingerprints trigger far fewer challenges in the first place.

• No dedicated engineer chasing detection updates — that maintenance burden sits with the infrastructure provider, not the team building the agent.

The actual comparison

The honest framing isn't "self-hosted is free, managed costs money." It's: self-hosted has a lower sticker price and a higher, less predictable total cost once proxy bandwidth and engineering time are counted. Managed infrastructure has a clear, predictable price that's higher upfront but removes the variable costs that tend to blow self-hosted budgets.

Which one is right depends on scale and team size. A team running occasional, low-volume scraping against undefended sites may genuinely be fine self-hosting. A team building an agent product that needs reliable browsing against sites that actively fight automation is usually better off treating browser infrastructure as a bought capability, the same way most teams don't self-host their LLM inference either.