ESXi 9.1 installer reboots in a loop on a Skylake host (CPU_SUPPORT OVERRIDEWARNING)

Skylake-SP is supported in Deprecated Mode for VCF 9.1.x, but an unattended install still loops without --forceunsupportedinstall. Discontinued from 9.2.

Add --forceunsupportedinstall to the install line in your kickstart. On a Skylake-SP host the ESXi 9.1 installer prints CPU_SUPPORT OVERRIDEWARNING, waits for someone to acknowledge it, and reboots when nobody does. An unattended install cycles back into the installer and does it again, forever. Interactive installs do not hit this, because a human presses a key.

Skylake is not dead in 9.1, and that is worth knowing before you plan a rebuild

A lot of writing from early 2026 says Skylake is a hard stop in ESX 9.x, blocked at install with no way through. That was true, and Broadcom then revised it.

As of the current KB, Intel Skylake-SP is supported in Deprecated Mode for VCF 9.1.x. Deprecated Mode means installation is permitted and the CPU stays supported for that release’s lifetime, with a warning that it will not continue. It is not the same as Discontinued, where the installer genuinely blocks you.

The runway ends at 9.1.x. Skylake-SP is Discontinued starting from 9.2, and Broadcom is explicit that support “will not be extended to VCF releases beyond VCF 9.1.x.” So this buys you one release, not a reprieve.

Affected parts are the Xeon Platinum 8100, Gold 6100/5100, Silver 4100, and Bronze 3100 series.

Why the unattended install loops anyway

Deprecated Mode still emits an override warning at install time, and a warning wants acknowledging.

Interactively that is a keypress. In a kickstart there is nobody to press anything, so the installer waits, gives up, and reboots. The host comes back up, boots the installer again, and repeats. What you see is a machine cycling endlessly, which reads as a bad ISO, a broken PXE config, or a dead boot device long before it reads as a CPU warning.

The tell is the boot disk: zero bytes written. If the installer had cleared the CPU check and failed later, something would be on that disk. Nothing is.

The exact string, if you catch it before the reboot:

CPU_SUPPORT OVERRIDEWARNING: The CPUs on this host are
not supported by ESXi 9.1.0

Note OVERRIDEWARNING. That is a warning with an override available, and it is a different thing from CPU_SUPPORT ERROR, which is the hard block you get on a genuinely discontinued CPU.

The fix

In your kickstart:

install --firstdisk --overwritevmfs --novmfsondisk --forceunsupportedinstall

That is the whole fix.

For an interactive install, or for a CPU that throws the harder CPU_SUPPORT ERROR, the equivalent is a boot option instead: press Shift+O within the five second window and append allowLegacyCPU=true.

The part that will make you think the fix did not work

The warning still displays with the flag set. It appears exactly as before and clears itself exactly as before.

What changes is whether the install then proceeds. So do not check whether the warning is gone, because it will not be. Check whether bytes are appearing on the boot disk. That is the only observable difference, and it is the one that matters.

Worth saying plainly, because the natural reaction to seeing the same warning again is to assume the flag was ignored and go hunting for a different cause.

Verified, and what is inference

Verified here: installing ESXi 9.1.0.0200 into nested VMs on a physical Xeon Platinum 8168 (Skylake-SP) produced that warning and the reboot loop, with nothing written to the boot disk. Adding --forceunsupportedinstall to the kickstart resolved it, and the hosts built and verified afterwards.

From Broadcom’s KB, not tested here: the Deprecated Mode status for 9.1.x, the discontinuation at 9.2, and the affected CPU list.

Inference: that the loop is specifically caused by an unattended install having nobody to acknowledge the warning. It fits the observed behaviour exactly and it is why the interactive path behaves differently, but I did not instrument the installer to prove it.


The wider story of building this lab, including how an agent drove most of it, is in I Let Claude Code Deploy My VCF 9.1 Home Lab, and This Is What Happened.

Sources: CPU Support Deprecation and Discontinuation In VCF Releases and Deprecated CPU Systems / Servers in ESX 9.0.