ServerBudget

How Old Is the OS Image on a Cheap VPS? Release at Signup, EOL Ownership, and What a Reinstall Deletes

By Marcus Reid · Published Nov 11, 2026 · Last updated Nov 11, 2026

Short answer: On the pages I could open, cheap-tier image freshness is not stated at all. One budget plan page names three distribution families and prints no release number [S17]. A large provider publishes exact release slugs and numeric image IDs, but the only date on that page is a documentation revision stamp of 2026-07-13, not a build date [S11]. A third provider says a version stays selectable for at least 3 months after it is marked deprecated [S15]. There is a real upper bound on the age, though: that provider's deprecation answer uses Debian 11 as its own example of a pre-made image [S15], and Debian's published table gives Bullseye's end of LTS as 2026-08-31 [S3]. After the clock stops, the distribution stops shipping fixes [S5], the hosting contract puts loss risk and backup copies on the customer [S20], and the documented route back to a clean image erases the disk [S13][S15]. Nobody in that chain promises to carry you to a supported major release [S5][S20].

1. What the order form actually names

Start with the narrowest page in the set. The budget KVM plan I read says it supports CentOS, Debian and Ubuntu, and that a machine can be running in two minutes or loaded from your own ISO through a control panel [S17]. No version. No point release. No kernel. No package-set state. No date on which the template was assembled.

That bare label is more ambiguous than it looks. CentOS Linux 8 reached its end of life on 2021-12-31, with the project stating plainly that it would not be automatically migrating anyone to the next version and that there would be no more updates to that content after the deadline [S5]. CentOS Stream is a different product: a new major version every three years, each maintained for five years, and maintained by the RHEL team [S6]. A plan page that prints only the word CentOS does not tell a buyer which of those two clocks applies [S17]. That is an interpretation, not a finding about what is installed: the reasoning chain is that two distinct published lifecycles share one catalogue label, and the catalogue does not disambiguate them.

The same vendor publishes a version somewhere else. Its own introductory guide names two Ubuntu LTS releases, 22.04 and 24.04, as the friendliest starting point for a newcomer and lists a first-boot package-update command among its setup steps [S18]. So the release number appears in a tutorial, after the purchase decision, while the plan page that the buyer actually orders from carries no number at all [S17][S18]. A guide and a catalogue are two different documents with two different jobs.

The comparison editorial pages sit in the same narrow lane. One roundup asks whether a buyer can choose a Linux distro on a cheap plan and lists Ubuntu, Debian and CentOS as the usual range [C1]. Another guide frames the operating system as a choice to be made on compatibility, ease of use and support [S19]. Neither raises the age of the template, the release that will be installed, or what happens when that release expires. A third provider's setup guide was not retrievable today [U5]. Coverage ends at selection; the clock starts after selection, and that is exactly the part nobody writes down.

Two larger providers do publish exact versions, which makes the omission sharper rather than filling it. A published image table lists Ubuntu 22.04, 24.04 and 26.04 LTS, Debian 13, Fedora 43 and 44, AlmaLinux 8, 9 and 10, Rocky Linux 8, 9 and 10, and CentOS Stream 9 and 10, each with an API slug and a numeric image ID [S11]. Its metadata carries a last-updated stamp of 2026-07-13 [S11]. A separate standard-image table lists Debian 11 Bullseye, 12 Bookworm and 13 Trixie, Ubuntu 22.04, 24.04 and 26.04 LTS, CentOS Stream 9 and 10, AlmaLinux 8, 9 and 10, Rocky Linux 8, 9 and 10, openSUSE Leap 15.6, and Arch at latest, and says support teams update the images regularly [S14]. That page was last changed on 2026-06-10 [S14].

Neither table contains a build date, a point release, a kernel version, or the patch level of the packages inside [S11][S14]. A release label is a name, not an age. "Updated regularly" gives no interval and no anchor [S14]. A documentation revision stamp tells you when a web page was edited, not when a disk image was cut [S11]. The computable quantity before paying is therefore not "how old is this image" but "how much of this release's published support window remains" — and that number comes from the distribution's own calendar, never from the vendor's order form [S1][S3][S11][S14][S17].

2. Two clocks, and who owns the second one

The distribution's clock is written down in unusual detail. Ubuntu LTS releases arrive every two years and receive five years of standard security maintenance for packages in the Main repository, while interim releases get nine months [S1]. Expanded Security Maintenance extends Main coverage to ten years and adds ten years for Universe; a paid Legacy add-on takes an LTS release to fifteen years [S1]. The two streams are named separately: esm-infra covers Main after the standard five years ends, and esm-apps covers Universe for ten years from the LTS release [S2]. Canonical puts roughly 2,300 packages in Main per release and over 36,000 in Universe as of 24.04 LTS [S2]. Ubuntu 20.04 LTS standard maintenance ended in May 2025, 22.04 LTS ends in May 2027, 24.04 LTS ends in May 2029, and the 26.10 interim release runs to July 2027 [S1].

That five-year figure is not a property of the whole software stack. It describes Main [S1][S2]. A cheap machine running a language runtime, a database driver or a web framework from Universe sits under a different boundary, and the free standard term does not cover that repository [S2]. Buyers who read "five years of support" as "five years for everything I installed" have taken a repository-scoped statement and applied it to a machine. That reading is mine; the chain is that the publisher defines the commitment by repository, the two repositories have different sizes and different streams, and the standard term names only one of them [S1][S2].

Debian publishes a comparable but differently shaped calendar. Its life cycle runs five years: three of full support followed by two of Long Term Support [S3]. The release index gives Debian 11 Bullseye an end-of-life date of 2024-08-14 and an end of LTS of 2026-08-31; Debian 12 Bookworm an end-of-life of 2026-07-11 and an end of LTS of 2028-06-30; and Debian 13 Trixie a release date of 2025-08-09, an end-of-life of 2028-08-09 and an end of LTS of 2030-06-30 [S3]. The current stable release is version 13, first issued as 13.0 on 2025-08-09 with a 13.7 point update on 2026-09-12 [S3].

The interesting part is who holds the pen after the first handover. Debian's LTS wiki says LTS is handled by a group of volunteers and companies, and explicitly says it is not handled by the Debian Security and Release teams [S4]. The LTS team takes over security maintenance once the security team stops, and the number of properly supported packages depends directly on the level of support the LTS team receives [S4]. So responsibility moves: a release team, then a volunteer-and-company team with a support-dependent package count, then a third-party paid extended tier for older releases [S3][S4]. At no point in that sequence does a VPS provider appear.

Other publishers set different lengths again. AlmaLinux says each minor version reaches end of life when the following minor version is released, while shipping stable updates and security patches for 8.x until 2029 and 9.x until 2032 [S8]. Its release notes give version 9 active support until 31 May 2027 and security support until 31 May 2032, and version 10 active support until 31 May 2030 and security support until 31 May 2035 [S7]. Rocky Linux advertises a ten-year support lifecycle [S9]. Fedora gives approximately 13 months of support per release, with the N-2 release reaching end of life four weeks after release N appears [S10]. AlmaLinux also points CentOS 7 users to ELevate, a community-maintained tool for in-place migration [S8].

That last pair matters for "image age" in a way a shopping list does not capture. A point release can expire while its major version is still supported [S7][S8]. Fedora's window is measured in months, not years [S10]. Asking whether the distribution is "still supported" is therefore too coarse; the question is which exact release string came with the machine, and where that string sits on its own publisher's table [S1][S3][S7][S8][S10].

Now the second clock: the contract. The budget provider's terms assign the customer all exposure to loss or damage affecting the machine and everything stored on it, and require the customer to hold archival and backup copies independently [S20]. The same clause says those servers do not function as an archive and rules out liability when content disappears [S20]. All content is permanently deleted on termination [S20]. The only provider-side update action I found in that contract is scoped to a full managed VPS: in the presence of vulnerabilities, the provider may reboot the machine up to two times a month without notice to install updates [S20].

Read that scope carefully. It is a reboot to install updates on a managed product, not a major-version migration after end of life, and it is not written for the unmanaged cheap tier [S20]. The vendor's own introductory guide makes the split explicit: unmanaged means you handle updates, security, software installation and troubleshooting, while managed includes OS updates, security patches and monitoring from the provider [S18]. Neither text describes anyone carrying an unmanaged machine to a new major release when the old one expires [S18][S20]. Canonical's "no forced upgrades" language describes staying securely on the same LTS without a major upgrade during standard maintenance; it is not a promise of transport to the next release [S2]. CentOS states the opposite case outright: no automatic migration, and no more updates after the deadline [S5].

The two clocks are therefore not connected by any document I could read. One is a publisher's maintenance calendar with a defined end [S1][S3][S5]. The other is a hosting contract that allocates risk and, on the cheap tier, allocates the work of updating to the customer [S18][S20]. The buyer stands at the join between them.

3. The exit button is also the erase button

Suppose the version on the machine has run out of road. The obvious remedy is to put a fresh image on it. Every reset path I could open describes that operation as destructive.

One provider's rebuild documentation says rebuilding wipes the disk and replaces it with an image you select, that the process is irreversible, and that without backups, snapshots or local copies the data is completely irretrievable by the provider [S13]. The same page recommends rebuilding when you want to keep an IP address, because rebuilding retains the address while destroying a machine releases it back into the datacenter pool [S13]. It also restricts rebuilding to an image from the same operating system family as the original [S13]. Its creation guide adds that weekly backups are retained for four weeks and daily backups for seven days, and that backups can be used to rebuild an existing machine [S12].

A second provider's FAQ says that to reset a server to its delivery state you select Rebuild, restore the desired OS, and that all saved data will be lost [S15]. The same FAQ keeps deprecated images available for at least another 3 months after the deprecation date, after which they may become unavailable at any time [S15]. A third provider's image documentation says plainly that images are not intended to serve as a full backup solution [S16]. The budget provider's acceptable-use policy says copies of new or changed data are produced weekly and attaches no assurance about whether those copies are intact [S21]. Its terms make independent archival copies the customer's job [S20].

Read those clauses together and the constraint becomes obvious. The operation that fixes an obsolete image is the operation that erases the data [S13][S15][S20]. The safety net is time-boxed in one case [S12], explicitly disclaimed as a backup solution in another [S16], and explicitly not guaranteed in a third [S21]. The contract has already told the customer to keep a copy somewhere else [S20]. An image is not a copy of your data; it is a copy of an operating system [S16]. Confusing those two is the mistake that turns a routine refresh into a loss.

There is a sharper edge. A rebuild limited to the same operating system family cannot, by its own wording, carry a machine from one family to another [S13]. A provider whose reset says "select and restore the desired OS" describes a different, broader operation [S15]. So the phrase "just rebuild onto a supported release" is not one action across providers; it is at least two, and on the narrower one it may require destroying and recreating the machine instead, which releases the address [S13][S15]. That is an inference with a reasoning chain: one manual constrains the target image by family, another does not, and the first manual's own destroy path is the only remaining route when the target falls outside the family. I did not test either operation.

The three sub-questions meet here. You cannot read the image's age off the order form [S11][S14][S17]. When its publisher's window closes, no contract I read assigns the migration to the provider [S5][S18][S20]. And the reset that would install a newer release also removes the files, with a narrow and explicitly qualified set of backups standing behind it [S12][S13][S15][S16][S20][S21]. The oldest component of a cheap VPS is often the thing nobody dated, and the cost of replacing it is paid in data rather than in dollars.

The agreements, and the four splits

The agreement is about shape, not numbers. Every publisher I read publishes a fixed support window rather than an open-ended one: five years for Ubuntu LTS [S1], five years for a Debian release [S3], ten years for Rocky Linux [S9], approximately 13 months for a Fedora release [S10], and dated end points for AlmaLinux 8.x, 9.x and 10 [S7][S8]. Every accessible reset path says the disk is cleared [S13][S15][S20]. None of the three catalogues I read publishes a build date or a package patch level [S11][S14][S17].

The splits are more useful. Catalogues disagree on disclosure itself: one prints slugs and numeric IDs [S11], one prints full distribution names with an "updated regularly" note [S14], and one prints bare family names [S17]. Publishers disagree on what happens after the free window: Canonical sells a paid extension with two named streams [S1][S2], Debian hands maintenance to a separate volunteer-and-company team whose package coverage depends on the support it receives [S4], and CentOS simply stops [S5]. Providers disagree on what a safety net is: time-boxed retained backups [S12], an explicit statement that images are not backups [S16], and weekly copies with no integrity guarantee [S21]. They also disagree on whether a reset may change the operating system family at all [S13][S15].

One more divergence is worth keeping. A distribution can remain listed after its own support dates have passed. The standard-image table still lists Debian 11 Bullseye [S14], and the FAQ's deprecation answer uses Debian 11 as its illustration of a pre-made image [S15]; Debian's own index gives Bullseye an end of LTS of 2026-08-31 [S3]. Being selectable is evidence of being in the catalogue, not of being current [S14][S15]. A deprecated image is described as still available and only removed later [S15]. That distinction is the whole gap in one sentence, and it is an interpretation built by joining three separate pages rather than a claim any one of them makes [S3][S14][S15].

The figures here cover OS image release dates; for the part that follows, the VPS outbound identity page lists the identity checks you can run on a trial instance.

If you came here for OS image release dates, the promo renewal cost page puts the promo rate and the renewal rate on one line.

Reading about OS image release dates raises a separate question, and the real cheap VPS complaints page collects the failures buyers report after the first invoice.

A pre-order checklist with four lines

What the checklist cannot prove is the package state of the image you will actually receive, because no catalogue read here publishes it [S11][S14][S17]. Nor can it promise that the release offered today is the release offered next month: a deprecated image stays selectable for at least 3 months and may disappear after that [S15]. No measurement of an image was taken for this page, and no estimate of how often any provider refreshes a template appears in it.

Limits: what could not be read

Several rebuild and reinstall documents were unavailable on the day this was written. A cloud-server rebuild guide returned HTTP 404 [U1]. A compute-instance rebuild guide returned HTTP 404 [U2]. Two reinstall and operating-system index paths on another provider returned HTTP 404 [U3][U6]. A legacy rebuild guide returned HTTP 403 [U4]. The third comparison site's setup article returned HTTP 403 [U5]. None of them is quoted, and none is used to infer a policy. Earlier notes about those pages were not used to fill the gaps.

The two retrieval observations for the comparison sites also differ. The Linux roundup and both editorial guides from the other comparison site opened normally [C1][S18][S19], while the setup article on the third site did not [U5]. Only the pages that opened are used for coverage statements; the one that did not is left as unreadable, and no wording from it appears here.

The bottom line, stated narrowly

Age is the wrong noun for what the order form withholds. A release label can be current while the disk carrying it is months out of date, and the publisher's calendar keeps running whether or not anyone touches the machine. What a buyer can actually establish before paying is narrower and less satisfying: which release, how much of its published window remains, who patches after the last date, and whether the reset that would be needed leaves a copy of the data somewhere else. Those four answers live in four different documents, held by four different parties. No page in this set joins them up. The cheapest tier is where that missing join is easiest to ignore, because the only number on the plan card is the price.

Frequently Asked Questions

Can I find out how old the image is before I pay?

Not from the pages read here. One budget plan prints no release number at all, and the two providers that do print exact release slugs or full distribution names publish no build date, no point release and no package patch level [S11][S14][S17]. What you can establish instead is the remaining support window of the named release, taken from the distribution's own table [S1][S3]. A documentation revision stamp is not an image build date [S11].

Who has to upgrade the OS after its end of life?

On the unmanaged cheap tier, the contract puts the work and the risk on the customer: independent archival copies are the customer's responsibility, and the provider's servers are described as not being an archive [S20]. The vendor's own guide says unmanaged means you handle updates and security, while managed includes OS updates and security patches from the provider [S18]. Neither document describes an automatic move to a new major release after expiry [S18][S20]. CentOS says outright that it will not automatically migrate anyone [S5], and Debian hands post-EOL maintenance to a separate volunteer-and-company team [S4].

Will reinstalling or rebuilding keep my files?

On every reset path I could read, no. One rebuild guide says the disk is wiped and the process is irreversible, and that data without backups, snapshots or local copies is completely irretrievable by the provider [S13]. A second provider's FAQ says all saved data will be lost on a reset to delivery state [S15]. A budget contract says content is permanently deleted on termination and makes independent copies the customer's job [S20]. Backups are time-boxed in one case, described as not a full backup solution in another, and offered without an integrity guarantee in a third [S12][S16][S21].

Sources

The three comparison sites were opened only to see which questions their pages ask. Their phrasing is not carried into this text, and nothing here comes from a transcript. Fixed names and short labels inside clauses are kept as printed, while continuous prose is rewritten throughout. Every cross-page reading above is labelled as an interpretation or an inference and carries its reasoning chain. No image was measured, no template was probed, and no estimate of how often any provider refreshes an image is offered.

Marcus Reid
Systems Architect & VPS Infrastructure Analyst

Marcus has been working with virtualized server infrastructure since 2015, deploying and benchmarking VPS and cloud instances across DigitalOcean, Vultr, Hetzner, OVHcloud, Contabo, RackNerd, Hostinger, and the major hyperscalers. His work focuses on the gap between what providers advertise and what actually happens when you deploy a real workload. He verifies every price and specification against official provider pages before publication.