Microsoft veri merkezi, arşiv fotoğrafı. Used with permission from Microsoft.
Used with permission from Microsoft.

Windows Server 2028 enters preview with hardened security and local AI plans

Microsoft has opened Windows Server 2028 to public preview. Announced on 17 November, it combines infrastructure improvements with a longer-term push toward on-premises AI. It is not a generally available stable release.

Different features, different stages

Foundry, Windows ML, vLLM and Llama.cpp integrations are under development. Hybrid AI capabilities have limited early access and hardware requirements. Intune server management is in preview, while Windows Admin Center Virtualization Mode is generally available. Storage additions include native NVMe, NVMe over TCP/RDMA and booting from ReFS. Trusted Launch, certificate management and post-quantum cryptography feature in the security roadmap.

Hardened Server Core

Starting with build 29706, Server Core installs with hundreds of security settings enabled. Microsoft warns that older protocols, authentication methods and management scripts may need changes. Administrators should test application compatibility before deployment.

Hardening remains incomplete: NTLM blocking, some hardware protections and printing/RPC safeguards are planned for later previews. A score below 100% in the built-in evaluation tool does not necessarily mean every reported setting is missing; measurement coverage is still being refined.

Preview limitations

Build 29706 offers Desktop Experience and Server Core options for Standard and Datacenter. Azure Edition is for VM evaluation only. Upgrades from previews older than 29531 are unsupported; Microsoft calls for a clean installation. The preview expires on 15 October 2027.

Known issues include an LSASS crash during TLS hybrid key exchange and AppCompat FOD installation problems after certain Server Core upgrades. These limitations underline why the preview should not be treated as production-ready software.

AMD acceleration

AMD announced Instinct MI350P support for Windows Server 2025 and 2028. Microsoft places its preview in the first half of 2027. AMD Inference Microservices targets simpler model deployment, while GPU partitioning allocates capacity among workloads. Local inference can give organizations control over where data is processed; identical cost savings or performance are not guaranteed across deployments.

Image: Microsoft datacenter, archive photograph. Used with permission from Microsoft.

Sources: Microsoft Windows Server 2028; Hardened Server Core; Build 29706; AMD.

Image

Subscribe for Updates