Time to first byte is the least glamorous performance metric and the one that most often explains a slow site. Before a browser can parse HTML, request CSS, or discover an image, it needs the first byte of the response. If that byte is late, everything downstream is late.
This archive covers a fixed 60-day window from 30 August to 28 October 2026. We bought five production plans at standard rates, deployed the same WordPress and WooCommerce workload, and sampled every host from Ashburn, Frankfurt, Singapore, Sydney, and São Paulo every five minutes. The result is a comparison of the tested plans under a fixed workload — not a universal ranking of the brands.
- Rocket.net recorded the lowest p75 TTFB at 96 ms, with the strongest regional consistency.
- Kinsta was the fastest managed platform at 118 ms, and its p95 tail was the tightest outside Rocket.net.
- Cloudways delivered 162 ms at the lowest hosting cost, but only after Redis was configured.
- Regional spread was larger than the gap between most hosts: São Paulo ran 64–82% slower than US East.
- Cache state changed the result more than any host-level tuning: cached pages were 3–5× faster than uncached ones.
The full ranking
The table below is the primary result. p75 is the headline because it describes a typical slow experience without being dominated by rare failures. p50 shows the median; p95 shows the tail that checkout and logged-in users may feel.
| Rank | Host | p50 | p75 | p95 | Best region |
|---|---|---|---|---|---|
| 01 | RNRocket.netCloudflare Enterprise edge | 71 ms | 96 ms | 168 ms | US East / 78 ms |
| 02 | KIKinstaGoogle Cloud + managed edge | 89 ms | 118 ms | 192 ms | US East / 92 ms |
| 03 | CWCloudways2 GB DO Premium + Redis | 121 ms | 162 ms | 268 ms | US East / 126 ms |
| 04 | WEWP EngineManaged platform + EverCache | 124 ms | 168 ms | 279 ms | US East / 131 ms |
| 05 | PRPressableManaged WordPress | 134 ms | 172 ms | 298 ms | US East / 137 ms |
Archive dataset: a fixed 60-day window from 30 August to 28 October 2026, measured from our own probe network. Values are the measured result for the tested plans; current host networks may differ. The global p75 uses the pooled request distribution, not a simple average of the five regional p75 figures.
p75 TTFB / all five hosts
Lower is better. Five-region aggregate.
The regional picture
A global average hides the only number that matters for a local business: the latency your customers actually experience. The table below shows p75 TTFB in each probe region. Read the column closest to your audience before you read the aggregate.
| Host | US East | Frankfurt | Singapore | Sydney | São Paulo |
|---|---|---|---|---|---|
| Rocket.net | 78 | 94 | 127 | 116 | 142 |
| Kinsta | 92 | 108 | 149 | 138 | 161 |
| Cloudways | 126 | 145 | 198 | 189 | 215 |
| WP Engine | 131 | 151 | 204 | 191 | 221 |
| Pressable | 137 | 158 | 211 | 199 | 225 |
Rocket.net and Kinsta use integrated edges; Cloudways was tested on its own cache stack. This is the product configuration a buyer would actually run, not a laboratory normalization. The methodology explains why we preserve that difference.
The regional penalty is not a host-specific flaw. It is physics plus footprint. A host with a point of presence in São Paulo can terminate TLS and serve a cached page locally; a host without one sends every request back to the origin. For a Brazilian store, the difference between 142 ms and 225 ms is a product decision, not a marginal tuning detail.
Cached versus uncached: the bigger variable
We probed two URL classes: anonymous pages eligible for full-page cache, and logged-in or cart pages that bypass it. The cache-state split explains more variance than the host ranking. Across the cohort, cached p75 was 68–104 ms; uncached p75 was 246–388 ms.
| State | p50 | p75 | p95 | What it represents |
|---|---|---|---|---|
| Cached anonymous page | 54 ms | 86 ms | 142 ms | Homepage, catalog, blog post with a warm edge cache |
| Uncached anonymous page | 188 ms | 274 ms | 462 ms | Cache miss, purge, or uncacheable template |
| Logged-in / cart | 241 ms | 318 ms | 584 ms | Session reads, Redis, dynamic PHP, database writes |
A store that brags about a 90 ms homepage but serves a 580 ms cart is not fast where it matters. This is why the WooCommerce load test exists and why we report checkout percentiles separately.
Method and configuration
All hosts ran WordPress 7.1 with a fixed theme, a fixed plugin set, and the same 12,000-product WooCommerce catalogue. We used PHP 8.4 where supported, enabled the host's recommended object cache, and configured full-page caching according to the host's documentation. The probes requested an anonymous product page and a logged-in cart page on a five-minute cadence.
- Probe locations: Ashburn (US-East), Frankfurt (EU), Singapore (AP-Southeast), Sydney (AP-South), São Paulo (SA-East).
- Measurement: synthetic HTTP requests from a fixed network, not browser-based lab tests. TTFB is measured to the first response byte.
- Aggregation: percentiles calculated over the full window, not averaged from daily averages.
- Exclusions: scheduled maintenance windows and requests that failed before receiving a byte are reported separately, not silently dropped.
What this report cannot tell you: how your site will perform. Theme weight, plugin count, database size, image strategy, and third-party scripts all sit on top of the host. The report compares infrastructure under a controlled workload; it does not predict your Core Web Vitals.
Five things that move TTFB more than switching hosts
- Cache hit ratio. A warm full-page cache can cut TTFB by 60–75%. If your cache is constantly purged, fix that before buying a faster server.
- PHP worker saturation. Queueing turns a 120 ms request into a 900 ms request. The PHP workers guide covers the math.
- Database and autoloaded options. A bloated options table slows every uncached request. Audit autoloaded data quarterly.
- Geography. A CDN helps cached pages; it cannot relocate your origin for cart requests. Choose a host with a region near your customers.
- Third-party calls in the critical path. A currency converter, review widget, or CRM lookup inside the template can add more latency than the host difference.
The short answer: Rocket.net is fastest on this workload.
Rocket.net is the fastest tested host in this window, and its edge is the main reason. Kinsta is the fastest managed platform and the most consistent across regions. Cloudways is the value option: third place overall at the lowest hosting cost, but only when the operator configures Redis and an edge cache.
The archive does not support a universal “best host” claim. It supports a narrower one: on a fixed WordPress workload, in these five regions, during this window, these were the measured differences. Use the regional table, the cache-state split, and your own operating model to make the decision.