In the first 51 days of PakSpeed, Pakistanis ran 30,653 speed tests across 537 locations. The result is the most granular, independently verified view of Pakistan's internet ever published.
More than half (52%) of measured internet tests in Pakistan are under 10 Mbps. This is the infrastructure reality that aggregate ISP marketing numbers have obscured for a decade. PakSpeed measures it in real homes, in real cities, in real time.
Unlike Ookla or similar global platforms, PakSpeed is Pakistan-first, bilingual (Urdu default), and publishes every ISP name, city, and anomaly openly. 100% of data is community-contributed. 100% is public.
Pakistan is the world's 5th most populous country. Its 130+ million internet users outnumber the entire population of Japan. Yet until this report, no independent, community-owned measurement documented what those users actually receive from their providers. This is a condition shared across lower- and middle-income economies worldwide.
Global platforms like Ookla, M-Lab, and RIPE Atlas provide valuable cross-country baselines — but they are externally operated, rarely bilingual in local languages, and do not typically publish ISP-resolved findings to local policy audiences. In countries where ISP self-reporting is the default and regulators lack granular independent data, the result is a decade-long accountability gap. PakSpeed is a model for closing it: Urdu-first, open-source, locally governed, and designed to hand evidence to the people most affected by the infrastructure it measures.
Regulators and policymakers (PTA, provincial IT ministries, digital-transformation units); civil society organisations working on digital rights and inclusion; academic researchers studying connectivity and development; journalists covering telecommunications; international funders and multilateral bodies investing in digital equity; and the Pakistani citizens whose tests built this dataset in the first place.
Every speed test gets placed in one of six brackets. The distribution, visualised across all 30,653 tests, shows the true shape of Pakistan's broadband.
Of the 241 cities with measurable data, only 15 are tier-1 metros. The other 226 — secondary cities, tehsils, and rural districts — tell a very different story.
78% of all PakSpeed tests originate from 15 metros. The spread between the fastest and slowest of them is 3.7× — larger than most international comparisons.
| Rank | City | Tests | Avg Download | Avg Upload | Avg Ping | Tier |
|---|---|---|---|---|---|---|
| 1 | Lahore | 5,314 | 18.3 Mbps | 6.9 | 144 ms | Mid |
| 2 | Karachi | 4,299 | 16.1 Mbps | 7.9 | 131 ms | Mid |
| 3 | Islamabad | 2,787 | 20.0 Mbps | 9.0 | 135 ms | Good |
| 4 | Faisalabad | 1,934 | 21.3 Mbps | 7.1 | 135 ms | Good |
| 5 | Multan | 1,040 | 19.1 Mbps | 8.1 | 135 ms | Mid |
| 6 | Rawalpindi | 946 | 19.5 Mbps | 9.9 | 132 ms | Mid |
| 7 | Peshawar | 447 | 18.5 Mbps | — | 135 ms | Mid |
| 8 | Gujranwala | 360 | 19.6 Mbps | 6.6 | 144 ms | Mid |
| 9 | Hyderabad | 299 | 13.4 Mbps | 7.1 | 123 ms | Mid |
| 10 | Bahawalpur | 253 | 18.4 Mbps | 8.7 | 147 ms | Mid |
| 11 | Sialkot | 241 | 27.1 Mbps | 10.9 | 142 ms | Good |
| 12 | Quetta | 240 | 14.5 Mbps | 7.1 | 154 ms | Mid |
| 13 | Hafizabad | 238 | 12.5 Mbps | 3.4 | 149 ms | Mid |
| 14 | Swabi | 231 | 7.3 Mbps | 3.3 | 176 ms | Poor |
| 15 | Sargodha | 163 | 16.8 Mbps | 6.3 | 149 ms | Mid |
Swabi records 231 tests averaging only 7.3 Mbps — the lowest of any major city. This is not a sampling issue. It represents a persistent infrastructure gap in a district of 1.6 million people. Swabi is what rural KP looks like from the inside of a metro.
| City | Tests | Avg Download | Beat which metro? |
|---|---|---|---|
| Sādiqābād | 57 | 32.6 Mbps | Beats every top-15 metro |
| Khanpur | 54 | 30.1 Mbps | Beats every top-15 metro |
| Haripur | 49 | 26.5 Mbps | Beats Lahore, Karachi, Islamabad |
| Rahim Yar Khan | 154 | 23.8 Mbps | Beats Lahore, Karachi |
| Shahkot | 73 | 22.2 Mbps | Beats Lahore, Karachi |
| Muzaffargarh | 63 | 21.8 Mbps | Beats Lahore, Karachi |
| Sahiwal | 41 | 21.6 Mbps | Beats Lahore, Karachi |
PakSpeed observed 317 distinct ISPs. The top 10 account for the majority of traffic — and the most revealing gaps in service.
| ISP | Tests | Cities | Avg DL | Verdict |
|---|---|---|---|---|
| Zong | 4,208 | 165 | 16.0 Mbps | Broadest reach, mid speeds |
| PMCL LDI IP Transit | 4,275 | 151 | 21.9 Mbps | High-volume, good speeds |
| PTCL | 3,554 | 121 | 21.9 Mbps | Incumbent, solid |
| Telenor | 1,736 | 111 | 9.1 Mbps | Wide reach, weak speeds |
| Cyber Internet Services | 2,535 | 52 | 16.8 Mbps | Urban-focused |
| Connect | 732 | 31 | 12.2 Mbps | Niche, mid speeds |
| Nayatel | 585 | 19 | 32.8 Mbps | Fastest at scale |
| Trans World Enterprise Services | 516 | 22 | 20.4 Mbps | Good regional player |
| National WiMAXIMS | 407 | 24 | 11.2 Mbps | Legacy WiMAX |
| Special Communication Org | 271 | 13 | 10.6 Mbps | AJK/GB coverage |
Telenor serves 111 Pakistani cities with an average download of 9.1 Mbps — below Pakistan's own broadband threshold. Across 1,736 tests, this is not an outlier: it's a service level. Telenor is the single largest source of sub-broadband internet in rural Pakistan.
Zong has the widest rural footprint in Pakistan — 165 cities, more than any competitor. But its 16 Mbps average is 17% below PTCL and PMCL LDI. If Zong's reach were paired with PTCL's performance, rural Pakistan's average speed would rise overnight.
Nayatel serves only 19 cities but averages 32.8 Mbps — the fastest network at scale in the country. This proves Pakistani fiber can hit world-class speeds. The question is why this quality isn't scaling beyond Islamabad/Rawalpindi.
PakSpeed's AI layer (Cloudflare Workers AI) compares every test against a rolling ISP/city/hour baseline. Statistically significant drops are flagged as anomalies. Here are the most recent weeks' flags.
| Week | ISP | City | Type | Incidents |
|---|---|---|---|---|
| Apr 11–18 | PMCL LDI IP Transit | Lahore | Throttling | 1 |
| Apr 11–18 | PMCL LDI IP Transit | Multan | Throttling | 1 |
| Apr 11–18 | S. B Link Network | Sādiqābād | Throttling | 1 |
| Apr 11–18 | Z COM Networks | Lahore | Throttling | 1 |
| Apr 11–18 | Zong | Karachi | Throttling | 1 |
| Apr 4–11 | National WiMAXIMS | Lahore | Throttling | 3 |
| Apr 4–11 | PTCL | Rawalpindi | Throttling | 3 |
| Apr 4–11 | PTCL | Dera Ismail Khan | Throttling | 1 |
| Apr 4–11 | Cyber Internet Services | Lahore | Throttling | 1 |
Every ISP × city × hour combination has a rolling baseline (mean + std deviation) calculated nightly. A test that falls more than 2 standard deviations below its baseline — with sufficient historical data — is flagged as a throttling event. Confidence scores are assigned per event. The system is deliberately conservative: 5 anomalies this week out of 1,457 tests (0.3%) reflects high specificity, not low noise.
PakSpeed has been publishing weekly bilingual (English + Urdu) reports every Sunday since launch. Each report is AI-generated, archived permanently, and publicly accessible.
| Week | Date range | Tests | Cities | ISPs | Avg DL | Anomalies |
|---|---|---|---|---|---|---|
| W1 | Mar 9–16 | archived | — | — | — | — |
| W2 | Mar 14–21 | archived | — | — | — | — |
| W3 | Mar 21–28 | archived | — | — | — | — |
| W4 | Mar 28–Apr 4 | archived | — | — | — | — |
| W5 | Apr 4–11 | 1,902 | 116 | 106 | 18.0 Mbps | 9 |
| W6 (latest) | Apr 11–18 | 1,457 | 109 | 106 | 14.5 Mbps | 5 |
Average download dropped 19.4% between W5 and W6 (18.0 → 14.5 Mbps) despite test volume remaining comparable. This is a statistically meaningful degradation that merits follow-up in the W7 report.
This report is a first milestone, not an endpoint. Four expansion tracks are actively underway or scheduled for the next twelve months — all built on the open, community-owned foundation PakSpeed has established in its first 51 days.
Deploy community measurement partners across the 191 under-sampled rural districts — prioritising the 31 critical-bracket towns this report names. Closes the single largest sampling gap in the current dataset.
IPv6 family detection is now live on every speed test (deployed 21 April 2026). Forthcoming: the first independent, ISP-disaggregated IPv6 adoption dataset for Pakistan, updated continuously.
Peer-reviewable longitudinal reports designed for regulatory, policy, and academic use — tracking ISP performance, throttling patterns, and digital-divide trendlines over time.
Public API and interactive dashboard at pakspeed.com/data. Full dataset available to researchers, journalists, and civil society under CC-BY-4.0.
A training cohort of 20–30 Pakistani journalists, civil society researchers, and university analysts — building independent capability to interrogate the open dataset without PakSpeed as an intermediary.
Open-source architecture designed for replication. Interest welcomed from civil society technologists across South and Southeast Asia who want to stand up equivalent platforms in their own economies.
Funders investing in digital equity and internet-governance infrastructure. Universities and research institutions interested in longitudinal connectivity analysis. Journalists and policy organisations who can use the open dataset to hold providers accountable. Civil society technologists in neighbouring economies who want to replicate the model. Contact: PakSpeed contact page.
Every test is run through LibreSpeed on a PakSpeed-operated Oracle VPS. Tests are voluntary, anonymous, and stored in a single Cloudflare D1 database with no PII beyond ISP name (from ASN lookup), city (from IP geolocation), and bandwidth measurements. Raw IPs are never stored. Users can toggle between Urdu and English; both locales share the same backend.
Three AI-powered systems operate on top of the raw data: (1) a throttling detector that compares each test to a rolling ISP/city/hour baseline, (2) an Urdu/English review NLP pipeline that categorizes citizen feedback, and (3) a weekly report generator that produces bilingual accountability summaries every Sunday. All AI runs on Cloudflare Workers AI. All outputs are stored and auditable.
1. Sampling is opt-in. PakSpeed data over-represents users motivated to measure their own speeds (often: people facing problems). Absolute averages should be read as ceilings on typical performance, not population means. 2. ISP names come from ASN lookup. Corporate naming is inconsistent (e.g. PMCL LDI IP Transit = Jazz). We normalize known aliases but some fragmentation remains. 3. Test volume is uneven. Cities with fewer than 3 tests are excluded from rankings. Rural sample sizes are growing but small — a second milestone at 100K will resolve most remaining uncertainty.
Every chart, table, and number in this report is derived from SQL queries against a single Cloudflare D1 database (pakspeed-db). Raw aggregate exports are available to researchers on request. The full worker source code is open-source at github.com/urduaiorg/pakspeed.