On July 13, Microsoft released Windows Server vNext Insider Preview Build 29621, giving IT teams a single build that bundles four technologies destined for the next Long-Term Servicing Channel edition. Cloud-assisted boot recovery, the ability to install the OS on ReFS, native NVMe-over-Fabrics connectivity, and Trusted Launch for Hyper-V virtual machines all came together in this LTSC Preview ISO. The build is firmly for test environments—it expires September 15, 2026, and comes with a list of known limitations—but it marks the first time administrators can evaluate all of these capabilities at once.

What Arrived in Build 29621

Microsoft’s announcement, covered first by Windows Report, details the four headline additions. Quick Machine Recovery (QMR) is the most operationally significant. Originally developed for Windows 11 under the Windows Resiliency Initiative, QMR uses the Windows Recovery Environment to fetch cloud-based fixes when a server can’t boot. After repeated boot failures, the system enters WinRE, connects to the network, checks Windows Update for a remediation, applies it, and tries to restart. In Build 29621, QMR is available only through a test mode. Administrators can simulate a recovery by running:

reagentc.exe /SetRecoveryTestmode
reagentc.exe /BootToRe

After a restart, the server rehearses a crash-and-repair sequence without an actual fault. A Group Policy setting to control QMR is promised for a future preview—critical for any organization that needs to manage the feature at scale.

ReFS boot support allows Windows Server to install onto, and start from, a Resilient File System volume. For years, ReFS has been the file system of choice for data volumes and virtualization workloads, while NTFS remained mandatory for the operating system partition. This build removes that constraint. The trade-off: a ReFS boot installation requires a dedicated Windows RE partition of at least 2 GB. Microsoft warns that the recovery environment may be disabled if the partition runs out of space, and that deleting the partition and extending the ReFS volume is irreversible without a clean install.

The NVMe-over-Fabrics initiator, first previewed in March 2026, is included again. It lets Windows Server reach remote NVMe storage over Ethernet (NVMe/TCP) or RDMA (NVMe/RDMA). NVMe/TCP runs on standard IP networks, while NVMe/RDMA targets low-latency deployments using RoCE or iWARP adapters. Microsoft describes the initiator as “basic and evaluation-focused,” so storage administrators should test persistence, path recovery, and disconnect handling rather than measuring raw throughput alone.

Trusted Launch for Hyper-V Generation 2 VMs introduces Secure Boot, a virtual TPM, and vTPM state encryption. The setup is PowerShell-only, requiring the IsolatedGuestVm optional feature and a TrustedLaunch guest-state isolation type. However, these VMs cannot be moved between hosts, placed in failover clusters, replicated, or managed in Windows Admin Center. Boot integrity verification isn’t yet supported. For now, Trusted Launch is suited to standalone hosts and disposable test workloads.

For Server Admins, Four Experiments Worth Running

For server administrators, Build 29621 is not a production build. It’s a preview sandbox where you can start answering practical questions: Does your backup software recognize a ReFS system volume? Can your deployment scripts create the larger recovery partition automatically? What happens when an NVMe-oF storage target disappears mid-transfer? How does your monitoring stack react to a VM that can’t live-migrate?

QMR’s real promise is reducing hands-on recovery work during widespread incidents—think bad driver updates or a corrupted cumulative update that floors hundreds of VMs. A cloud-delivered fix won’t solve every boot failure, but it could eliminate the need to attach an ISO or walk to the rack for each affected machine. The catch: QMR needs a functioning WinRE partition and a network path to Microsoft’s services. If your servers are on isolated management networks or air-gapped, the feature won’t help. You’ll also want that Group Policy switch before rolling out a plan around it.

ReFS system volumes force a rethink of disk layout. The 2 GB WinRE requirement isn’t a suggestion; it’s a deployment constraint. Check your unattended answer files, imaging tools, and any automated partitioning scripts. If you’re used to reclaiming leftover space after setup, resist the urge if you’ve chosen ReFS as the OS volume. The “no going back” warning from Microsoft means a single resize mistake could mean rebuilding the server from scratch.

Storage architects evaluating NVMe-oF should treat Build 29621 as a connectivity testbed, not a performance benchmark. Ensure the initiator reconnects after a host reboot, that multipathing works the way you expect, and that your monitoring picks up fabric disconnects. This isn’t a replacement for iSCSI or Fibre Channel yet—it’s an early look at how Windows Server will speak directly to remote NVMe arrays.

Trusted Launch’s missing mobility features make it a curiosity for now. If your security team demands enhanced VM boot protections on Hyper-V, plan to run these VMs on individual hosts without clustering. Watch for migration and management support in future builds before integrating Trusted Launch into any availability design.

How We Got Here: Four Streams Converge

Build 29621 doesn’t introduce any single feature that’s entirely new to Windows Insiders. What’s unusual is the simultaneous arrival of four distinct initiatives, each with its own history.

Quick Machine Recovery first appeared in Windows 11 builds in early 2026 as part of Microsoft’s effort to make Windows more resilient against boot failures. The Windows Resiliency Initiative was a direct response to incidents where flawed updates crippled large fleets. Porting QMR to Windows Server is a logical next step, but it also raises the stakes: server downtime has a different price tag than a laptop restart.

ReFS has been part of Windows Server since 2012, mainly for file shares and Hyper-V workloads. Microsoft gradually expanded its role—adding deduplication, larger volume sizes, and performance improvements—but a bootable ReFS volume is a foundational shift. It opens the door to server deployments that use integrity streams and block cloning from the OS up, but it also demands that every piece of the software ecosystem catch up.

NVMe over Fabrics is a Storage Networking Industry Association standard that Microsoft embraced first in Windows Server 2025 with native NVMe support for local drives. The March 2026 announcement of an in-box initiator for remote NVMe was a signal that Windows Server would go beyond Direct Storage. Build 29621 keeps that effort moving, but the initiator is still tagged as basic evaluation.

Trusted Launch for Hyper-V is part of a broader security push that also includes the cloud-oriented Trusted Launch for Azure. The Hyper-V implementation borrows the same concepts—Secure Boot, vTPM, runtime state encryption—but the on-premises version must contend with mobility, clustering, and management tooling that don’t exist yet in this preview.

The build also resets the preview baseline. Anyone running Windows Server vNext builds older than 29531 must perform a clean installation; upgrade issues prompted Microsoft to establish a new baseline with Build 29531, and flighting resumed from Build 29550 onward. Build 29621 is the latest post-baseline release, offered as an LTSC Preview ISO in 18 languages, an English VHDX, and an Azure Edition evaluation image.

A few known issues deserve attention. A TLS hybrid key-exchange race condition can crash LSASS when a server negotiates certain elliptic-curve groups. The workaround is to disable the X25519 and ML-KEM hybrid groups using TLS cmdlets or Group Policy. Additionally, Windows Update may label the 29621 flight as “Windows 11” even though it installs Windows Server vNext—a cosmetic bug that should be fixed later.

What to Do Right Now with Build 29621

If you’re spinning up Build 29621 in a lab, here’s where to focus:

  • Simulate QMR using the test mode commands. Walk through the reboot cycle and confirm WinRE connects and completes the fake remediation. Then test what happens when the network is unreachable—how does the system fall back?
  • Install Windows Server on an ReFS partition. Use the default setup and verify the WinRE partition is at least 2 GB. After installation, check that your standard management agents, backup clients, and endpoint protection tools install and run without errors. Try an online servicing operation (like installing a quality update) to see if the OS partition behaves.
  • Test NVMe-oF connections over both TCP and RDMA. Connect to a target, write data, reboot the server, and confirm the storage reconnects automatically. Induce a target failure and monitor the error recovery behavior.
  • Create a Trusted Launch VM with PowerShell, note its immobility, and then delete it. The current feature set isn’t broad enough for realistic testing beyond confirming the setup syntax and vTPM state encryption.
  • If you encounter an LSASS crash, apply the TLS workaround. This isn’t specific to any one feature; it’s a build-wide bug that could trip up early testers.
  • Do not attempt an in-place upgrade from a previous vNext build unless you’re ready for failure. The community has reported problems, and Microsoft has only committed to fixing upgrade paths over time.

Above all, treat Build 29621 as a configuration and integration test. The individual features may work in isolation, but the real insight comes from seeing how they interact with your existing toolchain and operational practices.

What to Watch Next

Microsoft hasn’t announced a release date for the next Windows Server LTSC version, but Build 29621 hints at a feature set that could land in preview channels later this year or early next. The most important forthcoming milestones are QMR’s Group Policy controls, Trusted Launch support for clustering and live migration, and a more robust NVMe-oF initiator with management cmdlets.

For now, the build gives server teams a head start on planning. You can test disk layouts, validate recovery flows, and begin conversations with storage and security vendors about ReFS boot readiness and NVMe-oF compatibility. The lab work you do today will pay off when these capabilities eventually reach general availability.