Incident sheet · /infrastructure
- System
- Infrastructure & Cloud
- Filed
- Read
- 3 min · 661 words
- Severity
- Routine
Windows Server 2028 Preview Targets On-Prem AI and Self-Healing Infrastructure
Microsoft's first Windows Server 2028 preview adds ReFS boot protection, hardened Server Core defaults, Intune server management, NVMe over Fabrics and post-quantum ML-KEM encryption for enterprises running AI in-house.

Event log
- Microsoft released the first public preview of Windows Server 2028 at Ignite, targeting on-prem AI workloads.
- Preview support for AMD Instinct MI350P accelerators is scheduled for the first half of 2027.
- Intune can now manage Windows Server fleets, applying patches and security policies, in preview.
- The release adds ML-KEM, NIST's primary post-quantum key-encapsulation standard.
- ReFS now covers the OS boot volume, with continuous checksum validation and background corruption scans.
Microsoft shipped the first public preview of Windows Server 2028 this week, built on the premise that enterprises will run a growing share of their AI workloads on their own hardware rather than in the cloud.
The timing is deliberate. At the opening keynote of Microsoft's Ignite conference, CEO Satya Nadella told business leaders to keep control of their intellectual property. "It becomes an architectural imperative to separate the supply of intelligence from the authority over it," he said. Doing at least some AI modeling in-house, Nadella argued, lets enterprises protect their IP and cut token costs.
What does the release change for resilience?
The preview arrives days after an extended Microsoft 365 outage took down the cloud versions of Outlook, Teams, Copilot, SharePoint and OneDrive — what HyperFrame Research analyst Don Gentile called an "awkward week" with "the cloud hanging over everything." Nadella acknowledged the disruption, calling it "a reminder of how critical digital infrastructure is as the dial tone for the world."
The takeaway for enterprises running operations in-house: resilience has to be handled on-premises too, Gentile said.
Neil Hinnant, director of product management for Windows Server, outlined the new mechanisms in a blog post:
- ReFS on the boot volume. The Resilient File System now extends to the OS boot volume, continuously validating checksums during reads and writes and running periodic background scans to detect data corruption.
- Quick machine recovery. Servers that fail to boot after a critical error attempt automated restoration, with administrator-configurable retry intervals and wait times between boots.
- Hardened Server Core defaults. Server Core installations now boot with more restrictive settings for networking, authentication and remote access, limiting what intruders can do once inside.
Gentile offered a caveat: ReFS boot protects the system volume, but file-system integrity and application-level recovery remain separate requirements.
Fleet management also changes. For the first time, administrators can manage Windows Server through Intune, which until now primarily handled Windows desktops and mobile devices. Intune can automatically patch servers, check configuration compliance and manage security policies. The server capability is in preview.
How does the OS support on-prem AI?
Microsoft is positioning Windows Server 2028 as a platform for in-house AI work:
- Native accelerator support for large inferencing workloads, with preview support for AMD's Instinct MI350P accelerator scheduled for the first half of 2027.
- NVMe storage enabled out of the box, letting users move data to and from SSDs via NVMe over Fabrics instead of SCSI/iSCSI or Fibre Channel.
- Deeper Foundry integration planned, with out-of-the-box runtime support for open-weight and custom models through Foundry Local, Windows ML, vLLM and llama.cpp.
Gentile noted that no Foundry integration exists today. Foundry-based AI services will carry security and resilience measures, but customers running AI on-premises remain responsible for backup and security, which will require third-party services, he said.
The hybrid AI features sit behind a separate gated access program. Enterprises wanting earlier access to features still in development can apply to the Windows Server Technology Adoption Program.
What about VMs, containers and encryption?
Microsoft assigns VMs and containers different security roles. Confidential VMs, built on hardware-based Trusted Execution Environments to prevent tampering, protect sensitive runtimes — though the work is early-stage. "Confidential VM support begins a journey to support integrity-protected runtimes on supported CVM hardware," Hinnant wrote. Microsoft will work with AMD and Intel on silicon-level isolation.
Microsoft Execution Containers, introduced last month, tackle a different risk: AI agents running unchecked across enterprise resources. MXC sandboxes each agent's actions with external resources.
For data in transit, Windows Server adds ML-KEM, NIST's primary post-quantum cryptography standard for establishing shared encryption keys over public networks. The rationale is already familiar: attackers could be harvesting encrypted data now, to decrypt later once quantum computers arrive.
One preview, one message: the data center, not just the cloud, is where Microsoft expects AI to live.
via techtarget.com (Source)
More from Amara Osei
Show full bio
Correspondent covering regulation, policy and their impact on enterprise information technology and the leaders who run at CIO Monitor.
203 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- Microsoft Ignite 2026: Copilot Becomes the CRM Front Door
- Microsoft's Global Outage: A $7 Billion Lesson in Cloud Concentration Risk
- 24 Hours Down, Zero Defections: Why Microsoft Outages Don't Cost It Customers
- Microsoft 365 Back After Multi-Day Outage Hit Firms in 70+ Countries
- Windows Server 2028 previews enterprise AI inference on local servers