Windows Server NIC teaming on 2025 comes in two flavours that look alike on paper and behave very differently once you pull a cable. The older LBFO team (the one in the NIC Teaming console) still works for a plain server, but the moment the Hyper-V role is installed it can no longer sit under a virtual switch, and Switch Embedded Teaming (SET) takes its place. We wanted to know what each one does when a link actually dies, so we built both on the same box and dropped links on purpose.
This guide covers a switch independent LBFO team with New-NetLbfoTeam, a SET vSwitch with New-VMSwitch -EnableEmbeddedTeaming, a link-pull test on each, and what LACP needs from the switch, including the “Faulted LACP negotiation” state you get when the switch side is missing. I ran through all of it on Windows Server 2025 Datacenter (build 26100) on a Proxmox homelab box in October 2026, and every timing and error below comes from that run.
LBFO or SET: which NIC teaming to use on Windows Server 2025
Use SET on any server that runs Hyper-V, and use LBFO only on servers that do not. Since Windows Server 2022, Microsoft has listed “Hyper-V Virtual Switch on LBFO” under features no longer in development. The vSwitch cannot be bound to an LBFO team any more, while LBFO stays supported for non-Hyper-V roles. The table sums up how the two differ, with the lab result for each row where we could test it.
| Item | LBFO (NIC Teaming) | SET (Switch Embedded Teaming) |
|---|---|---|
| Created with | New-NetLbfoTeam or the NIC Teaming console | New-VMSwitch -EnableEmbeddedTeaming $true |
| Needs the Hyper-V role | No | Yes |
| Teaming modes | Switch independent, Static, LACP | Switch independent only |
| Load balancing | Dynamic, TransportPorts, IPAddresses, MacAddresses, HyperVPort | Dynamic, HyperVPort |
| Can carry a Hyper-V vSwitch | No, New-VMSwitch refuses it (captured below) | Yes, it is the vSwitch |
| Managed with | NetLbfo cmdlets, NIC Teaming console | Hyper-V cmdlets (Get-NetLbfoTeam shows nothing) |
| Link pull on the Hyper-V host (lab) | 850 of 850 probes answered | 1,905 of 1,905 probes answered |
SET’s teaming-mode row is the one that catches people coming from older builds. SET has no LACP at all, so if your switch team insists on a port-channel for every server, the choice is LBFO on a host without Hyper-V, or a conversation with the network team about switch independent ports. Microsoft’s host network requirements for Azure Local also say SET supports up to eight adapters and expects them to be identical in make, model and speed.
What we tested on
The lab is one Windows Server 2025 VM on Proxmox with nested virtualization, so the same box could be a plain server first and a Hyper-V host later. If you are building something similar, our Windows Server template on Proxmox is the starting point we used, and the virtio adapters need the drivers covered in installing VirtIO drivers on Windows.
- Windows Server 2025 Datacenter, 24H2, build 26100.32230
- 4 vCPU with
cpu=host, 8 GB RAM - Two virtio NICs on the same Proxmox bridge, both 10 Gbps
- Link pulls done with
link_down=1on each virtual NIC - A PowerShell probe pinging the gateway every 100 ms
Every PowerShell command below runs in an elevated session. On a physical server you would pull a real cable instead of flipping link_down, and Windows sees the same thing either way, because the adapter reports media disconnected. Keep a console or iLO/iDRAC session open while you build the team, because the server’s IP moves from the physical adapter to the team interface and a remote session over that adapter will drop for a few seconds.
Step 1: Check the adapters you want to team
Both teaming types want adapters of the same speed, and SET wants the same make and model as well. List the adapters first so you have the exact names, because every later command takes them as strings:
Get-NetAdapter | Sort-Object Name | Format-Table Name, InterfaceDescription, Status, LinkSpeed, MacAddress -AutoSize
Our two virtio adapters came up as Ethernet and Ethernet 2, both at 10 Gbps:
Name InterfaceDescription Status LinkSpeed MacAddress
---- -------------------- ------ --------- ----------
Ethernet Red Hat VirtIO Ethernet Adapter Up 10 Gbps BC-24-11-00-00-01
Ethernet 2 Red Hat VirtIO Ethernet Adapter #2 Up 10 Gbps BC-24-11-00-00-02
Both rows say Up at the same speed, which is what you want. If one adapter shows a lower speed, fix the cabling or the switch port first, because the New-NetLbfoTeam reference states that teaming adapters of different speeds is not supported. The MAC of the first adapter matters later too: the LBFO team took over the MAC and the DHCP lease of Ethernet, and the SET host vNIC did the same.
Step 2: Create a switch independent NIC team with New-NetLbfoTeam
Switch independent is the safe default because the switch needs no configuration at all, and Dynamic load balancing spreads outbound flows across both members. This is the command we ran on the Hyper-V host before any vSwitch existed:
New-NetLbfoTeam -Name Team1 -TeamMembers 'Ethernet','Ethernet 2' -TeamingMode SwitchIndependent -LoadBalancingAlgorithm Dynamic -Confirm:$false
New-NetLbfoTeam returned after 1.63 s, but the team does not carry traffic that fast. It reported Up 32.8 s later, and the System log explains the wait with event 16948, “Connection Timer Started for Member Nic … It will be in use shortly”, for the second member. Expect the team to report Up only after about 30 s. Once it settles, check the team and its members:
Get-NetLbfoTeam -Name Team1 | Format-List Name, Members, TeamingMode, LoadBalancingAlgorithm, Status
Get-NetLbfoTeamMember -Team Team1 | Format-Table Name, AdministrativeMode, OperationalStatus, TransmitLinkSpeed, FailureReason -AutoSize
A healthy team shows Up with both members Active and NoFailure:
Name : Team1
Members : {Ethernet, Ethernet 2}
TeamingMode : SwitchIndependent
LoadBalancingAlgorithm : Dynamic
Status : Up
Name AdministrativeMode OperationalStatus TransmitLinkSpeed FailureReason
---- ------------------ ----------------- ----------------- -------------
Ethernet Active Active 10 Gbps NoFailure
Ethernet 2 Active Active 10 Gbps NoFailure
The new Team1 interface picked up a DHCP address 2 s after it came up, and Get-NetAdapter now lists it as “Microsoft Network Adapter Multiplexor Driver” at 20 Gbps. Static IPs do not move by themselves, so on a server with a fixed address set it on Team1 with New-NetIPAddress once the team is up.
Running the same command inside a guest VM is a different story. In a VM, LBFO refuses Dynamic and LACP outright, and the team we built that way did not survive losing its first member. Both errors and the failed test are in the last section. We only tested a Proxmox guest with both virtual NICs on one bridge, but the errors apply to any guest.
Step 3: Pull a link and watch the team fail over
A team you have never failed is a team you are only hoping works. We used a small probe that pings the gateway every 100 ms with a 500 ms timeout and writes each result with a timestamp, so a gap or a failure shows up to the tenth of a second. Create the script:
New-Item -ItemType Directory -Path C:\lab -Force
notepad C:\lab\pingloop.ps1
Paste in the probe exactly as we ran it:
param([string]$Target, [int]$Seconds = 120, [string]$Log = 'C:\lab\ping.log')
$p = New-Object System.Net.NetworkInformation.Ping
$end = (Get-Date).AddSeconds($Seconds)
"start $(Get-Date -Format 'HH:mm:ss.fff') target $Target" | Set-Content $Log
while ((Get-Date) -lt $end) {
$t = Get-Date -Format 'HH:mm:ss.fff'
try { $r = $p.Send($Target, 500); $s = "$($r.Status) $($r.RoundtripTime)" } catch { $s = "EXC $($_.Exception.InnerException.Message)" }
"$t $s" | Add-Content $Log
Start-Sleep -Milliseconds 100
}
"end $(Get-Date -Format 'HH:mm:ss.fff')" | Add-Content $Log
Start it against your gateway (ours is 10.0.1.1) for a few minutes, then pull one cable while it runs. Give the team at least 30 s after you plug the cable back in, since the returning member goes through the same connection timer as at creation:
powershell -ExecutionPolicy Bypass -File C:\lab\pingloop.ps1 -Target 10.0.1.1 -Seconds 170 -Log C:\lab\ping-lbfo.log
With the link down, the NIC Teaming console tells the story in one screen. Open it with lbfoadmin and look at three places:

- 1Team status
- 2Pulled link
- 3Surviving member
- Team status: Warning, so the team is degraded but still passing traffic
- Pulled link: Faulted, Media disconnected
- Surviving member: Active, carrying everything
Our probe log had 850 entries over 170 s and every one of them was a success, through the pull at 18:54:36 and the reconnect at 18:56:13. One caveat from a nested lab: the probe itself stalled for up to 16 s at times while the VM was busy, so read this as “no failed probe” rather than as a sub-second failover number. The System log showed only the pulled member disconnecting (event 16949), and the member rejoined 32 s after the link returned (event 16950).
Step 4: Install Hyper-V and see the vSwitch refuse the LBFO team
This is the step that sends most people searching. Install the role (it took 356.7 s on our VM and needs a reboot), and the NIC Teaming console starts reporting the server type as Physical, because Windows now runs as the Hyper-V parent partition. Our Hyper-V on Windows Server guide covers the role itself in more depth:
Install-WindowsFeature Hyper-V -IncludeManagementTools
The result comes back with RestartNeeded : Yes, so restart the server before going further. After the reboot, try to put an external vSwitch on the existing LBFO team:
New-VMSwitch -Name LbfoSwitch -NetAdapterName Team1 -AllowManagementOS $true
Windows Server 2025 rejects it with a clear message that points you straight at SET:
New-VMSwitch : Failed while adding virtual Ethernet switch connections.
An LBFO team may not be attached to the Hyper-V Virtual Switch. If teaming is required, use the 'New-VMSwitch' to create a Switch Embedded Team (SET). For more information, please see: https://go.microsoft.com/fwlink/?linkid=2185083.
Nothing is created and the LBFO team keeps working, so this failure is harmless. The fix is to remove the LBFO team and hand both adapters to a SET switch in the next step. Expect a short outage on the management address while the team goes away, so do it from the console. A static IP set on Team1 is removed with the team, and we only saw a DHCP lease come back on the new switch, so on a static server plan to set the address again with New-NetIPAddress on vEthernet (SETswitch) after Step 5:
Remove-NetLbfoTeam -Name Team1 -Confirm:$false
Both adapters were back as standalone NICs, Up at 10 Gbps in Get-NetAdapter, when we checked 10 s later, ready to be added to the switch.
Step 5: Create a SET vSwitch with New-VMSwitch -EnableEmbeddedTeaming
A SET team is not a separate object. It is a property of the vSwitch, so you create the team by creating an external switch with more than one adapter. -AllowManagementOS $true adds a host vNIC so the server itself keeps network access through the switch:
New-VMSwitch -Name SETswitch -NetAdapterName 'Ethernet','Ethernet 2' -EnableEmbeddedTeaming $true -AllowManagementOS $true
On our box that took 6.2 s, much faster than the LBFO team. Confirm the switch type and the team settings:
Get-VMSwitch | Format-List Name, SwitchType, EmbeddedTeamingEnabled, NetAdapterInterfaceDescriptions, AllowManagementOS
Get-VMSwitchTeam -Name SETswitch | Format-List
Look for EmbeddedTeamingEnabled : True and both adapters in the team:
Name : SETswitch
SwitchType : External
EmbeddedTeamingEnabled : True
NetAdapterInterfaceDescriptions : {Red Hat VirtIO Ethernet Adapter, Red Hat VirtIO Ethernet Adapter #2}
AllowManagementOS : True
Name : SETswitch
Id : 64e493c0-b701-4307-a0aa-da3725d3c9bf
NetAdapterInterfaceDescription : {Red Hat VirtIO Ethernet Adapter, Red Hat VirtIO Ethernet Adapter #2}
NetAdapterInterfaceGuid : {94cd7d51-765a-4a9e-a167-8bbb0b5fb9fd, ee73553b-15a8-4990-96d2-bd47669e770e}
TeamingMode : SwitchIndependent
LoadBalancingAlgorithm : HyperVPort
Note the load balancing value. The Set-VMSwitchTeam reference lists Dynamic as its default, yet New-VMSwitch built this team with HyperVPort on our build, so always read it back rather than assume. The host vNIC, vEthernet (SETswitch), kept the first adapter’s MAC and had its DHCP lease back 1.4 s after the switch was created. Get-NetLbfoTeam returns nothing once the switch exists, which is expected: SET lives in the Hyper-V switch, so the LBFO cmdlets and the NIC Teaming console do not manage it.
Step 6: Pull a link on the SET team
We repeated the probe for 260 s against the host vNIC and pulled each member in turn: Ethernet for 69 s, then Ethernet 2 for 68 s. The probe logged 1,905 results and all of them were successes. The largest stall in the probe was 10.15 s, and it happened before the first pull, so it was the nested VM being busy rather than the team.
SET does not write to the System log the way LBFO does. Its events are in the VmSwitch operational channel, which is the log to watch on a Hyper-V host:
Get-WinEvent -LogName 'Microsoft-Windows-Hyper-V-VmSwitch-Operational' -MaxEvents 30 | Sort-Object TimeCreated | Format-Table TimeCreated, Id, Message -AutoSize -Wrap
The command also returns older events, so the block below is trimmed to our pull window. Each pull and each return shows up as a member being dropped from load balancing, a connect timer, and the member being used again about 32 s later:
10/30/2026 7:01:38 PM 156 Stopping to use Member {94cd7d51-765a-4a9e-a167-8bbb0b5fb9fd} for load balancing
10/30/2026 7:02:47 PM 155 Starting ConnectTimer for Member {94cd7d51-765a-4a9e-a167-8bbb0b5fb9fd}
10/30/2026 7:03:20 PM 154 Using Member {94cd7d51-765a-4a9e-a167-8bbb0b5fb9fd} for load balancing
10/30/2026 7:03:30 PM 156 Stopping to use Member {ee73553b-15a8-4990-96d2-bd47669e770e} for load balancing
10/30/2026 7:04:38 PM 155 Starting ConnectTimer for Member {ee73553b-15a8-4990-96d2-bd47669e770e}
10/30/2026 7:05:10 PM 154 Using Member {ee73553b-15a8-4990-96d2-bd47669e770e} for load balancing
Each GUID matches an entry in the NetAdapterInterfaceGuid list from Step 5, so you can tell which physical port flapped. Those IDs (154, 155 and 156) are worth adding to whatever collects your event logs, because a member that keeps cycling through 156 and 154 is a bad cable or switch port, and the team will hide it from you otherwise.
Step 7: Change SET load balancing and team members
Microsoft recommends Hyper-V Port on adapters of 10 Gbps and above, and Dynamic is the only other choice. We switched to Dynamic here just to show the change, and it applied live with no reboot:
Set-VMSwitchTeam -Name SETswitch -LoadBalancingAlgorithm Dynamic
Get-VMSwitchTeam -Name SETswitch | Format-List Name, TeamingMode, LoadBalancingAlgorithm
The team reports the new algorithm straight away:
Name : SETswitch
TeamingMode : SwitchIndependent
LoadBalancingAlgorithm : Dynamic
Members can also be swapped while the switch is running, which is how you replace a failed NIC without rebuilding the vSwitch. We removed Ethernet 2 and added it back while the switch was in use (we did not run the probe during this part):
Remove-VMSwitchTeamMember -VMSwitchName SETswitch -NetAdapterName 'Ethernet 2'
Add-VMSwitchTeamMember -VMSwitchName SETswitch -NetAdapterName 'Ethernet 2'
Get-VMSwitchTeam -Name SETswitch | Format-List Name, NetAdapterInterfaceDescription
After the re-add, both adapters are back in the team:
Name : SETswitch
NetAdapterInterfaceDescription : {Red Hat VirtIO Ethernet Adapter, Red Hat VirtIO Ethernet Adapter #2}
A host built like this is also the starting point for a cluster, and our failover clustering guide picks up from there. Give every node its own SET switch built the same way, so the VMs see the same network wherever they run.
NIC teaming switch independent vs LACP: what needs the switch
Switch independent teaming needs nothing from the switch, which is why it is the default for both LBFO and SET. Static and LACP modes only work when the switch ports are bundled into one logical link on the switch side. LACP is the negotiated version: the server and the switch exchange LACP packets, and a member only joins the team once the switch answers. If the switch does not answer, the member never comes up.
We could not build a working LACP team in this lab because the Proxmox bridge is not an LACP partner, so the switch configuration below is explained, not tested. What we could test is the failure. With the server running as a Hyper-V host (so LBFO allowed LACP), we created an LACP team on two ports with no LACP on the other side:
New-NetLbfoTeam -Name Team1 -TeamMembers 'Ethernet','Ethernet 2' -TeamingMode Lacp -LacpTimer Fast -LoadBalancingAlgorithm Dynamic -Confirm:$false
Our LACP team never came up. PowerShell gave the same answer at 31 s, 63 s and 124 s after creation:
Name TeamingMode LacpTimer LoadBalancingAlgorithm Status
---- ----------- --------- ---------------------- ------
Team1 Lacp Fast Dynamic Down
Name AdministrativeMode OperationalStatus FailureReason
---- ------------------ ----------------- -------------
Ethernet Active Failed LacpNegotiationIssue
Ethernet 2 Active Failed LacpNegotiationIssue
The NIC Teaming console shows the same thing as “Faulted LACP negotiation”, which is the exact text people search for when a new LACP team will not come up:

- 1Team status
- 2Teaming mode
- 3Faulted LACP negotiation
- Team status: Fault, the whole team is down
- Teaming mode: LACP, waiting on the switch
- Faulted LACP negotiation on both members
When you see this on real hardware, check the switch side in this order. These are the usual causes, not something we reproduced on a physical switch:
- Both ports in the same port-channel or LAG
- The LAG set to LACP (active or passive), not static
- The same LACP timer on both ends
- Ports on the same switch or the same stack
Do not skip the timer item, because nothing on the server side warns you about it. Our team used -LacpTimer Fast, and a Datto knowledge base article written for Server 2012 R2 on resolving Faulted LACP negotiation traces the same console message to a fast/slow timer mismatch between Windows and a switch that defaults to slow. New-NetLbfoTeam takes -LacpTimer Slow as well, which is the server-side setting to use if your switch runs the slow timer. Remember also that none of this applies to SET, which has no LACP mode at all, as the last section shows.
What broke in the lab
Most of what went wrong happened before Hyper-V was installed, when the server was an ordinary VM. That is the situation of anyone trying to team two virtual NICs inside a guest, so each error is kept here word for word.
“The only valid LoadBalancingAlgorithms in a Virtual Machine are ‘TransportPorts’, ‘IPAddresses’, and ‘MacAddresses'”
LBFO detects that it is running in a VM and refuses Dynamic and HyperVPort. Our first New-NetLbfoTeam with -LoadBalancingAlgorithm Dynamic failed with this CimException after 1.22 s and created nothing. Rerunning it with -LoadBalancingAlgorithm TransportPorts worked, and the console labels that mode “Address Hash”.
“‘SwitchIndependent’ is the only TeamingMode value supported in a Virtual Machine”
LBFO’s VM detection blocks LACP and Static in a guest as well, so you cannot test LACP inside a VM at all. Only once the server became a Hyper-V host (Server Type “Physical” in the console) did -TeamingMode Lacp go through and produce the faulted team shown above.
In-guest team goes Down: VirtualSwitchLacksExternalConnectivity
Nothing else in the lab surprised us as much. With the guest team built on TransportPorts, we pulled the member that owned the team’s MAC address. Event 16949 logged both members as disconnected in the same second, the team went Down, and the member that was still plugged in reported a failure reason of its own:
Name TeamingMode LoadBalancingAlgorithm Status
---- ----------- ---------------------- ------
Team1 SwitchIndependent TransportPorts Down
Name Team AdministrativeMode OperationalStatus FailureReason
---- ---- ------------------ ----------------- -------------
Ethernet Team1 Active Failed PhysicalMediaDisconnected
Ethernet 2 Team1 Active Failed VirtualSwitchLacksExternalConnectivity
Traffic only came back when the pulled link did: 79 s of outage the first time and 83.7 s when we repeated it. Even then Ethernet 2 stayed failed with the same reason for almost six minutes, until Restart-NetAdapter -Name 'Ethernet 2' brought it back after 32.9 s. Pulling the other member only cost one timed-out probe. Both virtual NICs sat on the same Proxmox bridge, which is not the layout Microsoft supports for guest teaming. Its guidance for Hyper-V guests puts each member on a different external switch and turns on Set-VMNetworkAdapter -AllowTeaming On on the host, and we did not test that layout. Our take is simple. If you run Windows as a guest, do not team inside the VM. Team on the hypervisor instead (a Linux bond on Proxmox, as in our network bonding guide), and give the guest one NIC.
Set-VMSwitchTeam: Cannot convert value “Lacp” to type VMSwitchTeamingMode
Trying to turn a SET team into LACP fails at parameter binding, before anything touches the switch:
Cannot bind parameter 'TeamingMode'. Cannot convert value "Lacp" to type "Microsoft.HyperV.PowerShell.VMSwitchTeamingMode". Error: "Unable to match the identifier name Lacp to a valid enumerator name. Specify one of the following enumerator names and try again:
SwitchIndependent"
SwitchIndependent is the only value the enum holds, so there is no hidden switch for LACP on SET. If the network design needs LACP to a Hyper-V host, the honest options are LBFO without a vSwitch on top, or switch independent ports for the SET members. When you pick hardware for SET, buy identical adapters from the start; our 10GbE NIC roundup is a good place to compare them, and the Windows Server 2025 install guide covers the base system the steps above start from.