The VPS-vs-dedicated debate usually gets framed as "performance versus price". For privacy-critical workloads that's the wrong frame. The real question is what you're willing to share — and with whom — at the silicon level. A virtualised server is by definition a multi-tenant box: your kernel runs on top of a hypervisor that simultaneously runs strangers' kernels. For most workloads in 2026 that's a fine, well-defended boundary. For some, it's a structural liability that no amount of OS hardening can fix.
This guide draws the line. We'll cover the threat model that actually changes the answer, the hypervisor-escape CVE history you need to know about, the practical economics, and a decision matrix for which workloads belong where.
The threat model that decides the answer
Before comparing specs, write down a one-sentence threat model. The right server type falls out of it almost mechanically.
Single-tenancy: what it actually buys you
On a dedicated server — a physical box assigned exclusively to you — you control every layer below the OS that the host's contract permits: BIOS settings, secure-boot configuration, full-disk encryption with a passphrase the host literally cannot read, IPMI exposure, and which kernel modules load. There is no hypervisor between you and the silicon. There are no neighbours sharing the L1/L2 cache. There is no shared memory bus where a side-channel attack could observe your AES rounds.
On a VPS — a virtualised slice of a physical box — you control the guest OS and that's it. The host controls the hypervisor, the disk encryption keys (in most realistic configurations), and the physical machine.
Three threat categories
For privacy purposes, threats split into three buckets:
- Network adversary. Someone tapping or subpoenaing the wire. Defended by transport encryption (TLS, WireGuard, SSH) and jurisdiction. Server type is irrelevant.
- Host adversary. The hosting provider itself, or anyone who can compel them. Defended primarily by jurisdiction (covered in our jurisdiction guide) and secondarily by full-disk encryption with a passphrase the host doesn't have. Dedicated wins here, modestly.
- Co-tenant adversary. Someone who has rented a different VPS on the same physical box, or compromised one via a different vector, and is trying to escape their slice. Dedicated eliminates this category entirely; VPS does not.
If category 3 is in your threat model, the conversation ends — you need a dedicated server. If it isn't, a well-configured VPS in the right jurisdiction is fine for the overwhelming majority of privacy-sensitive workloads.
Hypervisor escape: how often does it happen?
The shortest honest answer: rarely, and with patches usually available within days. But "rarely" is not "never", and the historical record is worth knowing.
The big public escapes
- Xen XSA-226 (2017) — a memory corruption bug in the page-table handling that allowed a guest to escalate to host. Patched within a month; major cloud providers ran emergency reboots.
- VENOM (CVE-2015-3456) — a buffer overflow in the QEMU virtual floppy controller, affecting KVM and Xen. Old but instructive: the attack surface was a feature nobody was actively using.
- L1TF / Foreshadow (2018) — Intel speculative-execution side channel that could leak memory across hypervisor boundaries. Mitigated by microcode plus scheduling changes; performance hit on disabled hyperthreading was significant.
- KVM MDS variants (rolling, latest 2024) — Microarchitectural Data Sampling attacks. Each new chip generation produces a new variant; mitigations carry a measurable performance cost.
Public escapes that reach a CVE are the visible portion. Private exploits exist; nation-state-grade escapes have been demonstrated at Pwn2Own most years. For a workload where hypervisor escape is even on the list of plausible threats, you don't want to be on a hypervisor.
The IPMI / out-of-band channel
Both VPS and dedicated boxes typically expose IPMI (Intelligent Platform Management Interface) for the host's operations team. On a VPS, IPMI exposure is the host's problem and has nothing to do with you. On a dedicated server you can usually ask for IPMI to be on a private VLAN, behind VPN, or disabled entirely between maintenance windows. We default to "IPMI off, on request" on dedicated boxes — read the dedicated server page for the operational details.
Full-disk encryption: practical realities
Both server types support encryption at rest, but the trust model is different.
VPS encryption
You can run LUKS inside your VPS, which encrypts at the guest-filesystem level. This protects against a thief who steals the underlying disk after your VPS is shut down. It does not protect against a live memory snapshot taken by the hypervisor — your encryption keys are in RAM that the hypervisor can read. For most realistic threats this is fine; for a credible host adversary it is theatre.
Dedicated encryption
On a dedicated server, full-disk encryption with a remotely-typed passphrase (using dropbear-in-initramfs or similar) gives you a key the host literally cannot recover without your cooperation. The downside: a power cycle requires you to enter the passphrase, which is fine for personal infrastructure but awkward for autoscaling. The upside: a compelled host who seizes the box gets ciphertext.
VPS vs dedicated, head-to-head
| Dimension | VPS | Dedicated |
|---|---|---|
| Single-tenancy | No (shares CPU, RAM, hypervisor) | Yes (full physical isolation) |
| Co-tenant attack surface | Hypervisor + shared cache | None |
| FDE vs host adversary | Theatre (key in hypervisor-readable RAM) | Real (host gets ciphertext) |
| Spin-up time | Minutes | Hours to a few days |
| Hardware upgrade | Resize plan | Migrate to new box |
| Typical 2026 entry price (offshore) | $5–$15/month | $60–$200/month |
| IPMI exposure | Hidden (host problem) | Configurable (your problem) |
| Best for | VPN endpoints, build hosts, personal email, Tor relays, dev work | Mail servers at scale, key custody, CSAM-screening for image platforms, anything with category-3 threat |
Decision matrix by workload
Personal VPN, build host, dev box, Tor relay
VPS. The threat is network adversary plus jurisdiction; co-tenant escape is not a realistic concern relative to the operational cost. Browse VPS plans and pick the country that matches your jurisdiction guide outcome.
Personal mail server, small XMPP server, Matrix homeserver
VPS is fine for under ~50 users. Above that, performance starts to bite and you'll want dedicated for the IMAP/SMTP queue throughput regardless of the privacy axis.
Public-facing platform with user data (forum, image board, chat)
VPS for early growth, migrate to dedicated when the user count or the contents start drawing attention. Single-tenancy becomes valuable once you're a target rather than just incidental.
Crypto node holding meaningful funds
Dedicated. The category-3 threat is real — a co-tenant compromise that reads your seed via a side channel is not science fiction at this asset size. Dedicated server plans with full-disk encryption and IPMI off are the floor.
Whistleblower platform / leak host
Dedicated, in Iceland or Switzerland, with full-disk encryption and dropbear-initramfs. This is the workload where every layer matters. Pair with the jurisdiction guide for the legal layer.
Mass content distribution (video, large file hosting)
Dedicated, but for performance reasons more than privacy ones. A 1Gbps unmetered dedicated is cheaper than a VPS that bursts at the same rate.
The economics, honestly
VPS pricing in 2026 has compressed: 4GB / 2 vCPU / 80GB NVMe in an offshore jurisdiction is around $9 to $15 per month depending on country. Dedicated hardware in the same locations starts at $60 (low-end Atom or older Xeon) and runs to $200+ for current-generation EPYC or Xeon Scalable. The 4-to-10x multiplier is real, and for most workloads it isn't justified by privacy alone — it's justified by performance.
The honest split: about 80% of "I need privacy hosting" workloads are best served by a $10 VPS in the right country. The remaining 20% — high-asset crypto custody, journalism platforms, scaled content with attention from adversaries — need dedicated. Don't over-spend, and don't under-spend.
Operational checklist for whichever you pick
- Confirm the host's IPMI policy in writing before you order.
- Verify full-disk encryption support — for VPS, that LUKS is allowed; for dedicated, that dropbear-initramfs is permitted on first install.
- Check the AUP for explicit hypervisor escape disclosure clauses on VPS — a serious host commits to notifying customers within 24 hours of a confirmed escape.
- For dedicated, ask whether the box is new, lightly-used, or refurbished — and whether you can request a wipe before delivery.
- Read the rest of the privacy stack: VPN protocol choice, payment privacy, and the anonymous hosting use case.