ISP Speed Report: Real-World Internet Speeds Across 199 ISPs [March 2026 Data]

Your ISP says you're getting 500 Mbps. Their speed test confirms it. Case closed, right? Not so fast. We analyzed 646 real-world speed tests from Pong.com users across 199 ISPs in 40+ countries — and the data tells a very different story from the numbers on your bill.

Unlike ISP-sponsored reports that use optimized test servers sitting inside ISP networks, every test in this report ran through the real public internet via Cloudflare's global edge network. That means these numbers reflect what users actually experience when browsing, streaming, gaming, and video calling — not what a test server 2 hops away says.

The Headlines: 5 Things We Found

Overall Speed: What the Average User Actually Gets

Across all 646 tests, here's what real-world internet performance looks like in early 2026:

Notice the gap between average and median — especially for download (364 vs 276 Mbps) and upload (143 vs 92 Mbps). This tells us the distribution is heavily skewed by a small number of users with gigabit+ connections pulling the average up. The median is a more honest number — half of all users get less than 276 Mbps download and 92 Mbps upload on the real public internet.

For context, the average ISP-advertised plan speed in the US is around 200-500 Mbps for cable plans and 300-1000 Mbps for fiber. If the median real-world download is 276 Mbps, many users are getting noticeably less than what they're paying for when traffic crosses the real internet.

Internet Speed by Country: Who's Fastest?

We had enough test volume from 15 countries to draw meaningful comparisons. The results have some surprises:

Key Country Takeaways

• Canada leads downloads at 708 Mbps — likely driven by widespread fiber deployment from Bell and Rogers
• Australia surprises at 680 Mbps with the best ping (22 ms) — the NBN fiber rollout is paying off
• Switzerland has the fastest uploads at 346 Mbps — thanks to symmetric fiber from providers like Init7 and Swisscom
• Germany lags at 275 Mbps download, 53 Mbps upload — despite being Europe's largest economy, Germany's DSL-heavy infrastructure holds it back
• India brings up the rear at 191 Mbps download, 77 ms ping — mobile-first internet access means higher latency and lower throughput
• Ukraine's upload (219 Mbps) nearly matches its download (239 Mbps) — impressively symmetric, likely from fiber connections

ISP Report Card: Who Delivers and Who Doesn't

This is what your ISP doesn't want you to see: actual performance data from real users on the real internet. We identified 199 unique ISPs in our dataset. Here are the top 20 by test volume:

The Winners

Init7 (Swiss fiber provider) absolutely dominates with 859 Mbps download, 468 Mbps upload, and 12 ms ping. This is what symmetric gigabit fiber looks like when an ISP doesn't throttle it. Neptune Internet is close behind at 846 Mbps download with an incredible 21 ms ping. DIGI Romania rounds out the top 3 with 668 Mbps down and 360 Mbps up — Eastern European fiber is seriously underrated.

Among US ISPs, AT&T (581 Mbps) and Verizon (582 Mbps) lead, both likely driven by their fiber (AT&T Fiber / Fios) customers in our sample. Comcast comes in at 441 Mbps download but a much weaker 77 Mbps upload.

The Losers

Starlink averages just 177 Mbps download, 11 Mbps upload, and 83 ms ping across 30 tests. While satellite internet in rural areas is a game-changer for access, the latency and upload speeds make it poor for gaming, video calls, and any upload-heavy work. The 11 Mbps average upload is particularly concerning — that's barely adequate for a single HD video call.

Charter / Spectrum shows the most extreme asymmetry of any major US ISP: 270 Mbps download but only 17.4 Mbps upload. If you work from home on Spectrum and do video calls, cloud backups, or upload large files, this 15:1 download-to-upload ratio is a real bottleneck.

Reliance Jio (India) has the highest average ping in our entire dataset at 147 ms — nearly 5x the global average. Mobile-first internet in a developing market explains much of this, but it means online gaming and real-time applications are severely compromised.

The Upload Speed Crisis Nobody Talks About

The overall average upload speed is 143 Mbps — but the median is only 92 Mbps, meaning half of users get less than 92 Mbps upload. In a world where remote work, video calls, cloud backups, live streaming, and content creation are mainstream, upload speed matters more than ever. Yet cable ISPs in particular continue to deliver massively asymmetric service.

Compare this to the upload champions: Init7 (468 Mbps), DIGI Romania (360 Mbps), AT&T (278 Mbps), and Croatia's ISPs (266 Mbps). These are mostly fiber providers offering symmetric or near-symmetric service. The message is clear: if upload speed matters to you, fiber is the only answer. Cable and satellite ISPs are not keeping up with the demands of modern internet usage.

Bufferbloat: The Hidden Lag Problem in Your Router

This is the data point that no other speed test report can give you, because most speed tests don't measure it. Bufferbloat is when your router's buffer causes latency to spike under load — it's the reason your Zoom call freezes when someone starts a Netflix stream, even though you have plenty of bandwidth.

Download Bufferbloat: Mostly Fine

Good news: 77.7% of tests scored A or B on download bufferbloat. This means most users' routers handle download traffic reasonably well. Only 14% scored D or F. ISPs and router manufacturers have made progress here, partly because download-heavy traffic (streaming, browsing) has been the focus of optimization for years.

Upload Bufferbloat: A Disaster

The upload story is dramatically worse: only 42.3% scored A or B, while 41.6% scored D or F. Nearly 1 in 5 users has a failing upload bufferbloat grade. This directly explains why video calls and live streams stutter and break up — when you're on a Zoom call and someone else starts uploading to the cloud, the upload bufferbloat causes your call's latency to spike from 30ms to 300ms+.

Desktop vs Mobile vs Tablet: The Device Gap

We see a consistent and significant performance gap between device types. This matters because 60% of our tests came from mobile devices — and they're getting measurably worse performance.

Desktop users get 38% faster downloads than mobile users. This is expected — desktops are more likely to be on wired Ethernet connections, Wi-Fi 6E/7 adapters, or at least closer to the router. Mobile devices contend with cellular networks, Wi-Fi distance, and older wireless standards. Tablets perform worst overall, possibly because they're often used further from the router (couch, bed) on 2.4GHz bands.

Speed by Operating System: Linux Users Win

The operating system breakdown reveals some interesting patterns about user behavior and connection quality:

Linux users dominate with 629 Mbps average download — but this isn't because Linux makes your internet faster. It's a selection bias: Linux users tend to be power users, network engineers, and developers who are more likely to have fiber connections and wired Ethernet. Windows has the lowest jitter at 1.3 ms, suggesting stable wired connections. Android has the highest jitter (6.0 ms) and ping (53 ms), consistent with cellular/Wi-Fi variability.

Streaming Capability: Most Users Can Handle 8K

Pong.com estimates the maximum streaming quality each connection can sustain based on download speed and connection stability. The results are encouraging:

83.6% of connections can sustain 8K streaming, and 97.1% can handle 4K or better. Raw bandwidth is not the problem for most users. The issues arise from bufferbloat and jitter — your connection has the throughput for 4K Netflix, but the latency spikes from bufferbloat cause buffering events even though there's "enough" bandwidth. Speed tells you the pipe size; bufferbloat tells you if the pipe is clogged.

Packet Loss: The Good News

Of the 405 advanced tests that measured packet loss, 99.5% showed zero packet loss. Average packet loss across all tests was just 0.22% (pulled up by a single extreme outlier at 70%). This is genuinely good news — it means the infrastructure between users and Cloudflare's edge network is reliable, and dropped packets are not a common issue for most users.

What This Data Means For You

Here are the actionable takeaways from our analysis:

1. Your ISP's speed test lies by omission. It measures peak throughput on an optimized path. Run a test on Pong.com to see your real-world speed across the actual public internet.
2. Upload speed is being neglected. If you work from home, make video calls, or stream content, demand symmetric upload from your ISP — or switch to fiber.
3. Upload bufferbloat is an epidemic. 41.6% of users have D or F upload bufferbloat grades. Enable SQM on your router or get one that supports it.
4. Desktop + Ethernet still wins. Mobile and Wi-Fi add latency and reduce throughput. For critical tasks (gaming, important calls), plug in a cable.
5. Fiber is transformative. The fastest ISPs in our data (Init7, DIGI, AT&T Fiber) all deliver symmetric or near-symmetric speeds. No cable or satellite ISP comes close.
6. Starlink isn't there yet for latency-sensitive use. Great for rural access, but 83ms ping and 11 Mbps upload make it unsuitable for competitive gaming or high-quality video calls.

Methodology

For full transparency, here is exactly how this data was collected and analyzed:

• Test period: March 1-2, 2026 (approximately 38 hours)
• Total tests: 646 completed speed tests (416 Advanced, 230 Basic)
• Test path: All traffic routed through Cloudflare's global edge network via the real public internet
• Metrics collected: Download speed, upload speed, ping, jitter, bufferbloat grade (download + upload), packet loss, max streaming quality estimate
• User identification: ISP, country, city, device type, OS, and browser were collected via standard headers and IP geolocation. No personally identifiable information was stored.
• Unique ISPs: 199 distinct ISPs identified
• Countries represented: 40+ countries
• No filtering: All completed tests are included in the analysis. No results were excluded or manipulated.

Frequently Asked Questions