Choosing a jurisdiction: forty locations, four continents

Rank forty sites by latency, retention law and legal process, not a flag. One SKU price, Iceland and Norway +20%, Switzerland +35%. VPS from $3.00 annual.

Published 2026-08-30 · Updated 2026-10-14 · FYRHost Editorial · 16 min read

Pick the threat model before the flag

The flag is the least informative label on an order form, and most buyers read it first. What actually separates 40 sites on one map is who must reach your server quickly, which legal process you expect to see, how much data the facility can be made to keep, and whether your workload attracts complaints. Rank those four before you look at a single country name, because a decision that starts from the flag instead of from the four questions produces a server that is patriotic on paper and wrong in practice.

The mechanism is boring and decisive: latency answers to physics, retention answers to statute, and process answers to the courts that hold jurisdiction over the rack. Amsterdam answers milliseconds, Reykjavik answers speech, Zurich answers judicial process. Order your criteria before you order the server — who must reach it fast, which legal process you expect, how much retention posture matters, and whether the workload draws complaints. The country that survives all four questions is usually obvious before you ever open the price table.

Commerce is the part that does not move with the flag, which is why you should settle it first. Signup takes an inbox or a token plus a 12-character password, checkout runs the same 19 coins from BTC to the seven USDT networks at every city on the map, and abuse claims are read under the facility's own law. Identity demands, coin lists and logging rules are platform facts — identical in Reykjavik and Johannesburg — so the jurisdictional question never has to carry them.

Where it breaks: buyers rank countries by some forum's most-private-flags list, order in a city their users have never pinged, and discover the mistake in the first latency graph. The server serves milliseconds, not a brochure. If two sites tie on law and retention, let a stopwatch break the tie — run ping -c 20 from a machine where your users actually sit and let the answer pick the winner. That one command prevents the most common wrong purchase on this map.

Latency is physics, law is paperwork

No statute has ever shortened a round trip, and no court filing ever improved a packet's route. Peering density does that work: a site sitting on a big exchange hands your users near-instant responses, while a beautiful jurisdiction on a long transit hop adds a tax to every request. Treat the two as separate shopping lists, and never let either one vote on the other's question — a court cannot fix your time-to-first-byte, and a fast route cannot fix a subpoena.

Amsterdam is the latency anchor of this catalog because AMS-IX is one of the densest peering fabrics anywhere; western European users see the lowest round trips on the whole map, which is why SaaS frontends, VPN egress for the EU and anything users refresh constantly belong there. Measure before you believe: ping from the target region gives you the floor, and ss -ti on a live connection shows the round trip the kernel actually negotiated. Both tools ship with every distribution, and both are documented in the primary manuals rather than in folklore.

The flip side is the long hop. Reykjavik and Zurich both buy their posture with extra transit, and that is a fair price only when the jurisdiction is the point of the purchase. A publishing platform that trades milliseconds for speech protection is making a good deal; a session-based dashboard making the same trade while its users wait is not. Put the number on every candidate city before the shortlist exists, because after the shortlist exists you will defend it instead of measuring it. Entry lines run unmetered at 1 Gbps, so latency — not bandwidth — is the number that separates the cities.

The typical mistake is geography by inheritance: a team in Frankfurt orders wherever their previous provider had a rack, inherits the flag, and then wonders why the EU checkout page feels slow. Where it breaks is visible in the first week of metrics — response times climb on exactly the routes nobody measured. Measure from the user's vantage, rank by that number, and only then let law and retention prune the list. The stopwatch is the cheapest consultant you will ever hire.

Retention law decides what the disks keep

The question a jurisdiction answers about your disks is narrow and brutal: what can the facility be compelled to retain, and for how long? Everything else — the scenery, the climate pledge, the brochure adjectives — is decoration. The map is genuinely uneven on this axis, which is why retention deserves its own line in your ranking instead of being blended into a general feeling about a country. Privacy frameworks describe the considerations in the abstract; the statute books decide them in the particular.

Iceland sits at one end: no mandatory data retention, a speech-protective climate — IMMI, the International Modern Media Initiative, has its home there — and disks that legally sit outside the EU. Reykjavik is therefore the default rack for publishing platforms, Tor workloads and anyone whose threat model ends with the disk staying outside the Union. The +20% location premium is published on the price page rather than negotiated by sales, and the geothermal power under the floor is a bonus rather than the argument.

The EU end of the map runs on opposite machinery. Amsterdam is fast because it is central, and it is central under GDPR and Dutch takedown culture, which is not the ignore-the-letter kind. Bucharest is the practical middle: EU transit and competitive power under a Constitutional Court that struck mandatory retention down in 2014, the same year the CJEU gutted the EU-wide retention directive in Digital Rights Ireland. Foreign copyright templates arriving in Bucharest get treated as mail, not as orders, while ordinary criminal law keeps working normally.

Two symmetric mistakes ruin this section. The first is reading EU as one retention regime — the court cases above cut against the directive, and member states implemented wildly different national laws afterwards. The second is reading offshore as immunity: criminal process works at every site on the map, and the AUP travels with every SKU. Where the choice breaks is when a buyer wants Reykjavik's retention posture and Amsterdam's latency for a workload that draws complaints — no flag grants both, and pretending otherwise is how the wrong invoice gets paid.

The contract and the rack are separate layers

You sign with one company, and the RAM you rent sits in a different legal system than the one that company lives in. Both layers matter and neither cancels the other, which is the single most misunderstood fact about this catalog. Domicile governs billing and corporate process; the facility country governs what can happen to the disks. Confusing the two layers produces either false comfort or false alarm, and both are expensive.

Mechanically, the invoice comes from a Saint Kitts and Nevis entity — WESTFALL AND CO — no matter which of the 40 cities holds the hardware. Corporate process, billing disputes and refund questions follow the domicile. Search orders, seizure attempts and abuse claims follow the facility, because that is where the disks physically are. When you evaluate the jurisdiction, you are evaluating the second layer only; the first layer never changes with the flag and should never be priced into the decision.

What never changes at either layer is the acceptance policy: it travels with every SKU, at every site, in every premium region. Rent a Scout at $3.00 on the annual term in whichever city survives the four-question test above — the SKU, the ECC memory, the NVMe and the discount do not move with the flag. If a provider ever offered a jurisdiction that waives the AUP, that would be the reason to close the account, not to celebrate it.

The break happens when buyers treat corporate domicile as a content exemption. It is not: abuse is handled under the facility's jurisdiction, local criminal process works at every site, and the platform's own logging posture — no payload or traffic logs, connection metadata kept no more than 24 hours — is a factsheet, not a shield. Where people get burned is simple: they publish from a safe-looking flag something the AUP bans everywhere, and the account ends identically in every country on the map.

Premiums are published, not negotiated

One list price covers every city on the map, and the exceptions are printed rather than hidden. Five premiums exist in total: Iceland and Norway at +20%, Switzerland at +35%, and Luxembourg, Cyprus and Panama at +15%. Everything else, from Amsterdam to Johannesburg, bills at base list with the same annual discount applied. That is the entire pricing geography — there is nothing under the table, and nothing that changes after the invoice exists.

What each premium buys is law, not hardware. The Reykjavik and Oslo surcharge purchases non-EU disks and, in Iceland's case, a legislature that spent years pulling media jurisdiction onto its own soil. Zurich at +35% buys FADP — the Swiss Federal Act on Data Protection — which wraps data requests in a judicial overlay instead of an administrative one, plus non-EU status. Luxembourg, Cyprus and Panama price their own legal postures the same way: you are paying for process, and the invoice line says so plainly.

The premium never changes the machine. The same five VPS tiers, the same five dedicated tiers, the same IPMI, ECC and NVMe ship to the +35% rack as to base list, and the checkout math is transparent: the surcharge applies to the plan price, so a Scout that lands at $3.00 annual elsewhere simply scales with the percentage. If a compliance officer in your chain of command demands Swiss process, you will see exactly what that demand costs per year, in advance, on one page.

Where it breaks: buyers assume a premium region is better in some general sense and pay +35% for workloads that needed +0%. The premium purchases a specific legal posture; if your four-question test did not surface that posture as a requirement, the surcharge is pure waste. The symmetric mistake is avoiding premiums on principle when the workload — a mail platform, a Matrix homeserver, a client contract — genuinely names Swiss process as a condition. Price the requirement, never the vibes.

Complaints travel with the workload

Some workloads collect abuse tickets the way a seaport collects barnacles, and for those the jurisdiction question inverts: instead of asking what the flag protects you from, ask which facility's police will read the claims. Abuse handling here runs under the facility's own jurisdiction, so the country you pick is the country whose process you will experience. All 40 of them enforce the same short AUP — the choice controls who adjudicates, never whether a claim can arrive.

The AUP is short and identical everywhere: CSAM, phishing, spam and attacks are banned, and those categories are what facilities forward to their local law enforcement. Everything quieter — a chain node, a legal seedbox, a price-sensitive EU workload — is exactly what Bucharest was placed on this map for: EU transit, cheap power, and a court history since 2014 that treats foreign copyright templates as mail rather than orders. Heavy public-facing services should weigh Dutch takedown culture honestly; it is efficient, and efficiency cuts both ways depending on which side of a claim you stand.

Complaint volume is also a DDoS question, and that part is settled: L3/L4 filtering ships included at every site, so a denial-of-service flood is a platform feature rather than a regional one. What varies by flag is who adjudicates the complaints that survive the filter. A publishing platform in Reykjavik and a marketplace in Amsterdam may see identical attack traffic and wildly different letters afterwards — and that asymmetry is the actual product you are choosing when you pick the flag.

The mistake is jurisdictional arbitrage against your own workload: launching a mail-heavy or scanner-heavy project behind whichever flag scored best on privacy, then discovering that no flag makes spam lawful. Where it breaks is always the same hinge — the AUP is global, so the only variable your choice controls is which legal system processes the complaint, not whether one arrives. Pick the facility whose process you could actually navigate, in a language your paperwork can survive.

Failover across borders is a configuration, not a promise

Nothing in the account forces a single region, and no sales tier gates multi-country setups. Failover across jurisdictions is two orders and one DNS decision: put a server in two cities, let a record or a load balancer choose who answers, and you have survived a country-level outage without asking anyone's permission. The design question is not whether you may — it is how much consistency your application actually needs.

The mechanism is plain DNS plus discipline. Publish A and AAAA records for both endpoints, keep both boxes on one no-KYC account so billing never splits, and decide in advance whether the second site serves live traffic or waits cold. The DNS machinery has been described for decades in RFC 1034 and has not changed since; what the RFC cannot decide for you is consistency — replication between the two racks is your application's job, and it is the half people forget until the first failover test.

Do it concretely: put a Ranger at $10.00 on the annual term in Amsterdam and a second Ranger in Bucharest, then let DNS pick the survivor. The 2.5 Gbps uplink on that tier exists precisely so the second site can absorb the first one's traffic without renegotiating the ladder. Both invoices clear on the same 19 coins, both boxes answer on the same account, and neither city knows the other exists unless your application makes them talk.

Where it breaks: buyers assume the provider replicates anything. It does not — there is no hidden storage mirror between cities, and a failover whose database lives on one disk is a single point of failure wearing a costume. The second classic error is failing over to a box that was never warmed: test the second site quarterly by actually pointing traffic at it. A failover you have never served through is a hypothesis, not an architecture.

Metal does not tilt the decision — the flag does

The dedicated ladder is identical in all forty cities, which quietly removes the most common false tiebreaker. If IPMI, ECC memory and RAID ship on every tier at every site, then the metal cannot decide between Reykjavik and Amsterdam — the law can. Order the jurisdiction first and let the hardware follow, because the hardware menu does not change across the border anyway, and pretending it does only delays the real decision.

The five dedicated tiers travel as a set. The Shield at $30.80 on the annual term carries an E-2388G with 8 cores, 32 GB of memory and twin 500 GB drives behind IPMI; above it sit the EPYC-based Vault and Bastion, the Gold 6430 Fortress, and the dual-9474F Citadel with 25G uplinks and L3/L7 filtering. Every tier, every city, same ladder — so the question of which city gets better metal has no answer here, and the question itself is the trap.

What the flag does change is who can reach the disks. A Swiss mail server or a Matrix homeserver on a dedicated box with IPMI is a common pairing precisely because the buyer wanted FADP's judicial overlay on the data and used the metal for the control plane. Publishing platforms pair with Reykjavik for the same reason. The hardware decision — VPS versus dedicated, and if dedicated, which rung — is about workload shape; the location decision is about law, and the two never share a ballot.

The mistake is buying metal for jurisdictional reasons that a $3.00 Scout would have satisfied, or the reverse — staying on a VPS for a database that has outgrown shared CPU while telling yourself the flag matters more than IOPS. Where it breaks is cost: metal bought for the wrong reason sits idle at many multiples of the entry price. Decide the law with the four questions, then size the machine against the workload, and keep the two decisions in exactly that order.

The jurisdiction choice, in the order to run it

Run the decision as a sequence and it stays cheap. First, write the four questions down: who must reach the server fast, which legal process you expect, how much retention posture matters, and whether the workload draws complaints. Answer them before touching the location table, because the table is persuasive and the questions are not. A criterion that cannot be written as a sentence does not belong in the ranking at all.

Second, measure. Run ping -c 20 against candidate cities from a vantage where your users actually live, and record the numbers next to each country name. Third, prune by law: keep the sites whose facility jurisdiction matches the process you can navigate, and drop the ones whose retention rules conflict with what the disks may hold. This is where Reykjavik, Amsterdam, Bucharest and Zurich usually separate from the pack for the common threat models.

Fourth, price the survivors honestly. Base list is the same everywhere; Iceland and Norway add +20%, Switzerland +35%, and Luxembourg, Cyprus and Panama +15%. Decide whether a premium purchases something your test actually demanded. Fifth, check the workload against the AUP — the same short ban list at all forty sites — and against the complaint-handling posture of the facility's country, since that is the process you will live in if a claim ever arrives.

Sixth, order. Signup wants an inbox or a token plus a 12-character password, the invoice clears in one of 19 coins, and a VPS images in about a minute after the credit lands. Open the deploy page, pick the city that survived the sequence, and verify the region on the panel before you build anything on top. Dedicated metal follows the same path with a two-to-four-hour racking wait instead of the minute.

Seventh, and only if the first six steps earned it, add the second country: another server in a second jurisdiction, records for both, and a quarterly test that actually serves traffic through the backup. Revisit the whole sequence once a year or whenever your workload changes shape, because the correct answer to which jurisdiction is a property of the workload, not of the map — and workloads move even when borders do not.