eosl.ai
Vendor lifecycle data · recomputed Aug 2026

How Long Each Vendor Supports Hardware After End of Sale

Ten vendors, 2,775 vendor-published end-of-sale to end-of-service-life date pairs. One question: how long does each vendor keep supporting hardware after it stops selling it? Every figure is computed from the manufacturers' own end-of-life bulletins and recomputed on every database refresh.

62%
of published date pairs land within 3 months of exactly 60 months
60 mo
median support tail across every vendor-published pair
75/112
annual end-of-sale cohorts (2002–2027) whose median is the 60-month convention
2,775
vendor-published date pairs measured
What this page is — and is not. These are measurements, not verdicts. Every figure is counted from the vendors' own published end-of-life bulletins, and none of it is turned into a grade, score or rating: a grade would need thresholds EOSL.ai invented rather than anything a vendor published. EOSL.ai also has no data on product quality, reliability, support quality or value, and says nothing about them. Two columns in particular are not criticisms — "what this source carries" describes the feed EOSL.ai ingests, not everything a vendor publishes; and a low part-number figure means the vendor dates individual variants separately, which is more precise, not less honest.

By vendor

VendorTails on the 60-month conventionWhen they missWhat this source carriesOne date covers the familyPairs
Fortinet 98% negligible Both dates, per model 0% of 152 families 831
Cisco 95% negligible Both dates, per model 14% of 22 families 150
Brocade 93% runs long Both dates, per model 15
Palo Alto Networks 86% runs short Both dates, per model 29
Extreme Networks 84% both directions Both dates, per model 61
Arista 78% runs short Both dates, per model 18
Dell 70% both directions Both dates, per model 2% of 48 families 432
SonicWall 63% runs short Both dates, per model 19
Juniper 62% runs long Both dates, per model 13% of 46 families 208
Cambium Networks 47% runs short Both dates, per model 6% of 35 families 161
Cisco Meraki 45% both directions Both dates, per model 121
Calix 44% both directions Both dates, per model 18
IBM 15% runs long Both dates, per model 35% of 98 families 478
Zyxel 0% runs short Both dates, per model 195
Netgate 0% runs short Both dates, per model 32
HPE not enough data fewer than 15 published pairs End-of-sale only 7
A10 Networks not enough data fewer than 15 published pairs End-of-sale only 0% of 9 families

Tails on the 60-month convention — share of tracked end-of-sale → end-of-service-life pairs landing within ±3 months of 60. Shown only at 15+ pairs. When they miss — whether the misses run longer or shorter than the convention; running long means more support than expected, not less. What this source carries — which dates the specific source EOSL.ai ingests provides for that vendor. One date covers the family — share of families with more than one dated SKU group where a single end-of-service-life date covers all of them; shown at 8+ such families. A low figure means you must check the exact part number.

Where a support tail cannot be measured

These vendors are absent from the convention column above because the source EOSL.ai ingests for them does not carry both dates. This is a limitation of that source, not a finding about the vendor — and neither is judged on it.

Where the ingested source carries no per-model end-of-service-life date, EOSL.ai computes one from that vendor's published support-life policy and labels it computed — never presented as a date the vendor published. How the data is sourced →

The 60-month convention

Distribution of all 2,775 vendor-published support tails, in months from end of sale to last date of support. No standards body sets this number, and no vendor cites another vendor's policy — yet the industry has converged on five years.

≤36321 · 12%37–48156 · 6%49–5697 · 3%57–631,722 · 62%64–72201 · 7%73–96148 · 5%>96130 · 5%

57–63 months — the convention window, ±3 months of 60. Months on the left, pairs and share of the total on the right.

It has not moved

Grouping each vendor's pairs by end-of-sale year and taking the median per cohort (cohorts with at least 8 samples). For the vendors with the longest run of usable cohorts, the median support tail is the same today as it was at the start of the record.

VendorCohorts spanAnnual cohortsOn the convention
Dell2002–20262424 of 24 on 60 months
Fortinet2006–20262019 of 20 on 60 months
IBM2003–2022195 of 19 on 60 months

Method

A support tail is the number of months between a SKU group's vendor-published end-of-sale date and its vendor-published end-of-service-life date. Pairs are included only where the vendor published both dates for that group; policy-derived end-of-service-life dates are excluded, because they would all equal the policy length and manufacture a false consensus. Tails of zero or negative months, or of 240 months or more, are treated as data errors and dropped. Medians, not means — bulletin outliers are common. Below the sample thresholds a figure is withheld rather than estimated. The database tracks 2,443 hardware products; this page measures the 2,775 SKU groups within them that carry a complete vendor-published date pair.

Vendor lifecycle practices — questions

Is there a standard support period for enterprise hardware after end of sale?

There is no published industry standard, but there is a de-facto one: 62% of the 2,775 vendor-published end-of-sale to end-of-service-life date pairs in the EOSL.ai database fall within ±3 months of exactly 60 months — a 57 to 63 month window around five years. The overall median is 60 months. Measured across 15 vendors with at least 15 published date pairs each; policy-derived dates are excluded.

Has the hardware support period gotten shorter over time?

Not in this dataset. Grouping every vendor-published date pair by its end-of-sale year gives 112 annual cohorts with at least 8 samples, spanning 2002 to 2027; 75 of those 112 cohorts have a median support tail within ±3 months of 60 months. The 60-month convention has been stable across the period this database covers.

Which hardware vendor most consistently follows the 60-month support convention?

Fortinet, at 98% of its 831 published end-of-sale to end-of-service-life pairs landing within ±3 months of 60 months. This measures only how closely a vendor's published support tails cluster around the 60-month convention, which is what makes a refresh cycle plannable before the bulletin lands. It is not a rating, and not a measure of product or support quality.

Why can a support tail not be calculated for every vendor?

Because the machine-readable source EOSL.ai ingests for that vendor does not carry both dates. For HPE, the source EOSL.ai ingests carries an end-of-sale date but no per-model end-of-service-life date, for 95% of tracked groups — those EOSLs are computed from HPE's published support-life policy and labelled "computed" throughout the site, never shown as vendor-published. For A10 Networks, the source EOSL.ai ingests carries an end-of-sale date but no per-model end-of-service-life date, for 70% of tracked groups — those EOSLs are computed from A10 Networks's published support-life policy and labelled "computed" throughout the site, never shown as vendor-published. This describes the specific feed or document set this database ingests — it is not a finding about everything those vendors publish, and no vendor is scored on it.

Does EOSL.ai rate or rank hardware vendors?

No. This page reports measurements, not verdicts. Every figure is counted from the vendors' own published end-of-life bulletins, and none of them is converted into a grade, score or rating. A grade would require thresholds EOSL.ai invented rather than anything a vendor published, so the numbers are shown as they are and readers can judge them. Rows are ordered by the share of support tails landing on the 60-month convention, which is a fact about the dates, not a ranking of the companies.

Cite this page: “EOSL.ai Hardware Support-Tail Data by Vendor, Aug 2026” — https://eosl.ai/vendor-report-card/. Figures recompute on every database refresh. The underlying records are downloadable under CC BY 4.0 from the open dataset, so any figure here can be reproduced independently.