Buying guide / Speed
Fastest WordPress hosting, measured from five regions and sixty days.
One speed test from one city is a screenshot, not evidence. We sampled five hosts every five minutes from Ashburn, Frankfurt, Singapore, Sydney, and São Paulo, then ranked the p75 tail rather than the best-case run.
The premise
Speed is a distribution,
not a headline.
The fastest single request tells you almost nothing. What matters is the shape of the distribution: how often a visitor lands in the slow tail, and whether that tail is stable across the regions where your customers actually live.
We rank by p75 TTFB and publish p50 and p95. A host with a brilliant median and a brittle p95 still loses carts, leads, and patience.
01 / The ranking
Five hosts, one percentile.
Bars are scaled to the fastest result. A longer bar is better. The number on the right is p75 TTFB across all five regions, not an average of regional medians.
Global p75 TTFB
Lower is better · 60 days · five regions · cached and uncached mixed
| Host | TTFB p50 | TTFB p75 | TTFB p95 | LCP p75 | INP p75 | CLS p75 | Uptime |
|---|---|---|---|---|---|---|---|
| Rocket.net | 71 ms | 96 ms | 168 ms | 1.26 s | 84 ms | 0.04 | 99.998% |
| Kinsta | 89 ms | 118 ms | 192 ms | 1.43 s | 91 ms | 0.05 | 99.995% |
| Cloudways | 121 ms | 162 ms | 268 ms | 1.61 s | 104 ms | 0.06 | 99.987% |
| WP Engine | 124 ms | 168 ms | 279 ms | 1.52 s | 98 ms | 0.05 | 99.983% |
| Pressable | 134 ms | 172 ms | 298 ms | 1.66 s | 112 ms | 0.06 | 99.976% |
Read this before quoting: TTFB comes from the fixed 60-day archive ending 28 October 2026. LCP/INP/CLS come from the Sep–Oct 2026 field cohort. Your theme, plugins, and geography will produce different numbers. See the method.
02 / By region
The global average is a fiction.
A host can be fastest in Virginia and fourth in Singapore. If most of your customers are in one region, that regional number is the one that matters.
| Host | US East | EU Frankfurt | AP Singapore | AP Sydney | SA São Paulo | Spread |
|---|---|---|---|---|---|---|
| Rocket.net | 78 | 94 | 127 | 116 | 142 | 64 |
| Kinsta | 92 | 108 | 149 | 138 | 161 | 69 |
| Cloudways | 126 | 145 | 198 | 189 | 215 | 89 |
| WP Engine | 131 | 151 | 204 | 191 | 221 | 90 |
| Pressable | 137 | 158 | 211 | 199 | 225 | 88 |
“Spread” is the difference between the slowest and fastest regional p75. A smaller spread means more consistent global delivery. It is not a ranking criterion by itself; a host that is uniformly slow is still slow.
If you sell in one country, ignore the global podium. A host that is 20 ms slower globally but 30 ms faster in your market is the better choice for your customers.
CDN routing changes the picture. These probes mix cached and uncached requests. A host with a weaker origin but a better edge can look fast on anonymous pages and slow at checkout.
03 / The hosts
Where the milliseconds
come from.
Speed is a product of architecture, not a magic number. These notes explain where the milliseconds come from and what you give up to get them.
Rocket.net
Cloudflare Enterprise does the heavy lifting: full-page cache, image optimization, and a global edge included on every plan. The origin is quick, but the architecture is the advantage. You trade root access and platform breadth for a very fast default.
Kinsta
Google Cloud C2 infrastructure and a tightly managed container stack produce a consistent second place. Kinsta's tail is excellent, and its regional spread is second only to Rocket.net's. You pay for the platform discipline and the support model.
Cloudways
A 2 GB DigitalOcean Premium server with Redis enabled produced 162 ms p75. That is impressive for the price, but the result depends on configuration. Without Redis, the same plan would land closer to the middle of the table.
04 / The reality check
The fastest host
cannot fix a slow site.
A 96 ms TTFB is wasted if the browser then downloads a 4 MB hero image, three render-blocking fonts, and a chat widget that loads before the content. Hosting is the floor; the application sets the ceiling.
Measure the right path
A cached homepage is a CDN test. Measure a logged-in page, a cart, or a checkout to see the origin work.
Fix the heavy layer first
Images, fonts, and third-party scripts usually dominate LCP. Compress and defer before you upgrade the server.
Then buy consistency
Once the site is lean, a faster host improves the tail. Buying speed to compensate for a bloated theme is an expensive loop.
Beware the single-city screenshot. A vendor demo tested from the same city as the data center is not a global performance claim. Ask for the region, percentile, cache state, and test window.
Run a week-long probe. One test is a snapshot. Seven days of p75 from two regions is enough to see whether a host is consistently fast or occasionally lucky.
05 / Choose by situation
The fastest host
is not always the right host.
The fastest host for a global publisher is not automatically the fastest host for a single-market store. Match the architecture to the workload.
| Situation | Primary constraint | Pick | Why |
|---|---|---|---|
| Global publisher | Edge delivery and cache hit ratio | Rocket.net | Cloudflare Enterprise included; best p75 in all five regions. |
| Single-market store | Checkout tail and worker headroom | Rocket.net or Kinsta | Test the checkout path, not the homepage. Both hold up under concurrency. |
| Hands-off business site | Consistency and support | Kinsta | Second-fastest with a tight tail and the best support result. |
| Budget-conscious operator | Performance per dollar | Cloudways | 162 ms p75 on a 2 GB Premium server, with Redis configured. |
| Multi-site agency | Cost per site and control | Cloudways | Cloud-level isolation at the lowest cost per site; you own the tuning. |
Editorial picks, not universal grades. Each host is assessed against the stated workload. See the intent-weighted scoring model for the weights.
06 / Evidence
Start with the method,
then the numbers.
Every number on this page has a test window and a method. Start with the full report, then review the method and summary tables. Raw logs are available on request.
FAQ
Questions, answered directly.
The details that matter, without the runaround. Prefer the raw method? Start with how we test.
What is a good TTFB for WordPress?
For a cached page served from an edge, aim for a p75 TTFB under 200 ms in your primary market. For uncached, dynamic requests, under 400 ms is a reasonable target; under 250 ms is strong.
Context matters more than a single threshold. A host at 120 ms in Frankfurt may be 240 ms in Singapore. Compare regional numbers and percentiles, not a global average.
Why do you rank by p75 TTFB instead of average?
The average is pulled around by a small number of very fast or very slow requests. p75 means three-quarters of requests were at or below that number, which better reflects what a typical visitor experiences and exposes a brittle tail.
We publish p50 and p95 in the full table so you can see the distribution rather than a single headline.
Is the fastest host always the fastest website?
No. A lightweight theme on a mid-pack host can outperform a bloated site on the fastest host. TTFB is only the first byte; LCP also depends on HTML size, CSS, fonts, images, and third-party scripts.
Treat hosting as the floor, not the ceiling. Fix the application layer, then buy enough infrastructure to let it perform consistently.
Does a CDN make origin speed irrelevant?
For anonymous, cacheable pages, a CDN can hide a slow origin. For logged-in users, carts, checkout, and uncached API requests, the origin still matters. That is why we report cache-hit and cache-miss performance separately in the report.
A CDN is not a substitute for a competent origin. It is a multiplier on one.
Which host is fastest in Europe or Asia?
Rocket.net led all five probe regions. Its smallest margin was 14 ms, in US East and Frankfurt; Kinsta was consistently second. Cloudways was third in most regions once Redis and a CDN were configured.
The regional table in this guide shows the p75 for each host in Ashburn, Frankfurt, Singapore, Sydney, and São Paulo. Use the region closest to your customers.
How can I test my own host's speed?
Use a synthetic monitor from multiple regions, not a single browser test. Measure p75 and p95 over at least seven days, separate cached from uncached requests, and test a logged-in or cart path.
Our methodology explains the probe configuration. The full tables are on the benchmark pages if you want to inspect the underlying figures.