Argo Smart Routing · Path, Latency, and Bottlenecks

Three views of the same performance story. Pick a source region above to see the path change. Scroll down for the latency math and the architectural reason behind the bottlenecks.

The point: traffic on the public internet takes whatever path BGP gives it · often longer, often congested. Argo Smart Routing puts traffic on Cloudflare's private backbone and picks the fastest path in real time. Same source. Same destination. Different middle. Faster end-to-end.

Source region

Destination

Argo Smart Routing

Path

Public Internet

via BGP

Hops

intermediate networks

Round-trip latency

340

milliseconds

Packet loss

2.4%

avg measured

What's happening: The user in Shanghai is reaching a US East datacenter over the public internet. Traffic crosses several transit ISPs along the way · every hop is a chance for congestion, peering issues, or a less-than-optimal route choice. This is what most internet traffic looks like.

Anycast IPs

"One address, many doors"

Cloudflare's 330+ data centers share the same IPs. BGP automatically routes the user to the closest one. No client config. No DNS tricks. It's the internet's own routing protocol picking the nearest door.

Private Backbone

"Our highway, not theirs"

Once at the edge, traffic doesn't go back onto the public internet. It rides Cloudflare's private fiber backbone between PoPs. No ISP handoffs. No peering disputes. One operator, end to end.

Smart Routing

"GPS that updates every few seconds"

Argo measures every path across the backbone for congestion, loss, and latency. Picks the fastest in real time. If a path degrades, traffic re-routes automatically. The user never knows anything changed.

Anycast picks the closest door. The backbone is the highway between the doors. Argo is the GPS that keeps re-checking. Three layers of routing intelligence the public internet doesn't have.

Argo OFF · Public Internet path

Path typeBGP / best-effort

BackbonePublic transit

Route stabilityVariable

Congestion handlingNone

Typical hops12-15

Argo ON · Cloudflare backbone

Path typeSmart-routed

BackboneCloudflare private

Route stabilityOptimized real-time

Congestion handlingRe-routes automatically

Typical hops3-5

Do the math · step by step, side by side

Same scenario from the world map above, broken into five chapters. Change the source region in the dropdown to see your specific numbers. Each row is the same conceptual step on both architectures, so you can compare them directly.

VPN path

Total: 490ms

Cloudflare path

Total: 212ms

A note on physics Singapore to Virginia is roughly 15,000 km. The speed of light in fiber caps the one-way latency at around 75ms, no vendor breaks that. About 150ms of the Cloudflare total is just the cross-ocean physical round-trip, and Cloudflare can't make it disappear. What Cloudflare eliminates is the avoidable hops: tunnel handshakes, backhaul to a central concentrator, and stateful inspection. That is the 278ms saved on every request, every user.

Why this is a bottleneck, not just a slow link

The latency numbers above are a symptom. Here is what is actually happening underneath, and what changes.

What VPN costs you

ArchitectureEvery user, every packet, funneled through one box

Capacity at peakMonday 9am, the concentrator runs out of room

Cloud egressAWS traffic hairpins back through your data center

SaaS trafficM365, Salesforce, web · still routed through your DC

Failure modeSingle concentrator down means everyone is out

Cost shapeConcentrator + firewall + cert store + MFA appliance. Replace every 3-5 years.

The patternYou cannot fix one problem without making another worse

What Cloudflare delivers

ArchitectureNo central box · users hit the nearest of 330+ PoPs

Capacity at peakLoad spreads across our global network automatically

Cloud egressDirect to your AWS / Azure / GCP via tunnel or CNI

SaaS trafficDirect, with CASB control inline at the edge

Failure modePoP-level redundancy · anycast routes around failures

Cost shapeOne subscription, one bill, no hardware refresh cycle

The patternFix one problem, the others get better too

The combined story Argo Smart Routing fixes the path your traffic takes. The architectural change above fixes the shape of the network it rides on. The latency math is what happens when you change both.

When is the right time (or use-case) to use Argo?

Argo is a paid add-on. Here's when it earns its keep — and when it doesn't.

Most of "the Cloudflare benefit" is free the moment you point DNS at us: anycast to 330+ PoPsFreeDDoS mitigationFreeedge cachingFreeTLS + HTTP/3Free. Argo Smart Routing is the paid layer on top that optimizes the path between our edge and your actual server — the VM, cluster, or cloud service where your app runs. Below are the concrete customer shapes where it pays for itself, and the ones where it's the wrong lever.

Quick term: when we say “your server” we mean wherever your application actually lives — an EC2 instance, a Kubernetes cluster, an on-prem box in your data center, an AWS ALB with 20 nodes behind it, a Google Cloud Run service, whatever answers when Cloudflare has to ask a “real” (non-cached) question. If you run entirely on Cloudflare Workers, you don’t have one — Cloudflare itself is where your code runs.

Use-cases where Argo pays for itself

Good fit

Global SaaS on a single-region origin

You're a US company on AWS us-east-1. You just signed a customer in Sydney, Singapore, and São Paulo. Their login page takes 2 seconds.

Every API call, every dashboard load, every dynamic page has to cross an ocean and back. Cache doesn't help — the response is personalized. Argo takes the routing decision away from the public internet and puts it on Cloudflare's backbone, avoiding congested peering points along the way.

Typical improvement: 20-40% TTFB reduction on cache-miss traffic

Good fit

API-heavy or GraphQL backend

Your app is a React frontend hitting 40+ API endpoints per page. Fintech, healthtech, dev tools, any B2B SaaS with a real dashboard.

Static assets cache fine at the edge, but the API calls are the bottleneck and none of them are cacheable. Every /api/* request round-trips to the origin. Argo optimizes exactly that path, on every request, for every user.

Where it shows up: p95 API latency, dashboard load time

Good fit

E-commerce checkout and cart

Retail, D2C, marketplaces. Product pages cache well. But add-to-cart, checkout, and payment are dynamic and every millisecond costs conversions.

Amazon showed that 100ms of latency costs 1% of sales. That number is old but the physics haven't changed. The checkout path is where Argo earns its keep — dynamic, high-value, hits the origin every time.

Business KPI: cart abandonment rate, conversion rate

Good fit

Mobile-first or emerging-market users

Your users are in India, Southeast Asia, Latin America, Africa. Or they're on mobile carriers with lossy last-mile connections anywhere.

The public internet from a mobile carrier in Jakarta to us-east-1 crosses 12-18 hops on random peering agreements, with 2-5% packet loss on the way. Argo puts that traffic on Cloudflare's own backbone as early as possible, where packet loss is under 0.1%.

Also relevant: gaming, real-time collab, video conferencing signaling

Good fit

Ad tech, real-time bidding, quote engines

You're serving programmatic ads, running RTB auctions, or generating insurance/travel quotes on the fly. Your SLA to your partners is in tens of milliseconds.

There is no cache. Every request is unique, priced, and time-boxed. Argo's percentage improvement lands directly on your win rate because you're competing against other bidders on latency.

Direct revenue impact: bid win rate, quote-to-bind ratio

Good fit

You've already tuned the free levers

Your team has aggressive caching, HTTP/3, Brotli, tiered cache, image resizing, everything Cloudflare gives you for free. TTFB is still slow on the dynamic paths.

This is the honest "we've done our homework" case. Argo is the next lever after the free ones are pulled, not a shortcut to skip caching hygiene.

Sanity check: your cache hit ratio should already be > 80% before you buy Argo

Skip it

Marketing site or static content

Your traffic is a Next.js marketing site, a documentation portal, a blog, or an image-heavy catalog. Most responses come from cache.

The edge cache is already doing the work. Argo optimizes cache-miss traffic. If your miss rate is 5%, Argo speeds up 5% of your traffic. Fix cache first, revisit Argo later.

Try instead: aggressive Cache Rules + Tiered Cache (both free/included)

Skip it

Origin and users in the same region

App runs in a US data center, users are 100% US. Or app in Frankfurt, users all in Europe. Domestic B2B is a common shape.

The public-internet path is already short and reliable inside a single continent. Argo's improvement is measurable but small — often single-digit ms. Not worth the per-GB bill for domestic-only traffic.

Re-evaluate when: you open a second geography or acquire an international company

Skip it

Low traffic volume

Internal tool, early-stage startup, hobby project. Traffic is under ~10 GB per month.

Argo is $5/month plus $0.10/GB. Under 10 GB the fixed fee dominates and you're paying for capacity you don't use. Get to product-market fit first.

Break-even is around: ~50 GB / month of dynamic traffic

Skip it

Your app runs entirely on Cloudflare

You've gone Workers + KV + R2 + D1 + Durable Objects. There is no external origin server.

Argo optimizes the path between Cloudflare's edge and your origin. If your code runs at the edge, there is no path to optimize. Save the money.

Exception: if you still have a legacy origin for a subset of routes, Argo helps those specific paths

Pricing (2026): Argo Smart Routing is ~$5/month + $0.10 per GB of traffic. Argo Tiered Cache and Argo Tunnel (now just "Cloudflare Tunnel" / cloudflared) are separate products — Tiered Cache is included with several plans, Tunnel is free. Don't conflate them with Smart Routing.