Microsoft is pulling the plug on Reserved Instance purchases for a swath of aging virtual machine series. Starting July 1, 2026, you won’t be able to buy new one- or three-year Azure Reserved VM Instances for at least 18 different hardware generations—effectively forcing a migration to newer silicon before 2028 for anyone chasing steep compute discounts. The cut affects 14 older series that will lose one-year reservation options, plus an additional set where both one- and three-year terms disappear. For Windows shops banking on predictable, discounted pricing, the clock is now ticking.

This isn’t a sudden power-down of running workloads. Existing reservations will be honored for their full term. But the end of new purchases and renewals puts cloud architects on a deadline: re-platform to supported instance families or accept on-demand rates—which can be up to 72 percent higher than Reserved Instance pricing.

The reservation retirement breakdown

Microsoft’s Azure portal and official pricing documentation have been updated with a precise list of affected VM series. While the company hasn’t published a single blog post enumerating every retired family, the policy change shows up in the reservation purchase UX and the Azure Retail Prices API.

For one-year reservations, 14 series are flagged as “no longer available for new purchase” as of July 1, 2026. These include the early A-series (Standard_A0–A7), Basic A-series (Basic_A0–A4), classic D-series v1, Dv2, and DSv2, the original F-series and Fs-series, plus several older memory-optimized and compute-optimized lines like the Ev3 and Esv3 families. The exact list is long, but the common thread is that each runs on Intel Haswell, Broadwell, or Skylake-era processors—chips that lack the performance-per-watt and security features of current Cascade Lake, Ice Lake, or AMD EPYC Milan instances.

An additional group of four “legacy” families loses both one-year and three-year reservation support. Here you’ll find the NV-series (NVIDIA Tesla M60 GPUs), the NC-series (Tesla K80), the ND-series, and the older H-series. These GPU and HPC instances are being pushed toward newer NVv3, NCv3, NDv2, and HB-series offerings, which provide dramatically higher throughput and TCO.

The table below summarizes the reservation cutoff categories:

Reservation Type Affected Series Last Purchase Date
1-year only A-series, Basic A, D1–5 v1, D1–5 v2, DS1–5 v2, F-series, Fs-series, Ev3, Esv3, and others (14 total) July 1, 2026
1-year and 3-year NV-series, NC-series, ND-series, H-series July 1, 2026

What Reserved Instances do—and what this change means for your Windows bills

Azure Reserved VM Instances are a commitment mechanism: you pledge to pay for one or three years of compute capacity, and Microsoft gives you a deep discount relative to pay-as-you-go rates. For many enterprises running steady-state Windows Server workloads, this can cut costs by 40–72 percent. The savings are especially critical for SQL Server, SharePoint, or domain controller farms that rarely scale down.

Because these reservations are tied to specific instance families in a specific region, the retirement directly impacts cost models. After July 1, 2026, you can’t lock in discounted pricing on the 18 outgoing families. If you need to retain an older instance size—perhaps because a legacy application balks at new hardware—you’ll either pay list price or resort to Spot VMs, which are ephemeral. Neither is a palatable option for production Windows environments that demand uptime.

Microsoft is clearly signaling that these families are at the end of their economic life. Several of them already carry a “scheduled for retirement” tag for the underlying VM sizes themselves. The Dv2 and Ev3, for instance, have been flagged in Azure Advisor since 2024. Shutting off reservation purchases is a softer nudge than an outright VM retirement, but it’s the harbinger: when the financial incentive vanishes, migration becomes the only prudent path.

Why now? The hardware sunset cascade

Underneath Azure’s reservation policy is a hardware refresh cycle that operators don’t talk about loudly. The Intel Xeon E5-2673 v3 (Haswell) that powered the original D-series v1 entered production in 2014. The E5-2673 v4 (Broadwell) behind many Dv2 and F-series followed in 2016. These processors miss modern speculative execution mitigations, have lower memory bandwidth, and deliver roughly half the performance-per-core of a 3rd Gen Xeon Scalable.

Microsoft’s capacity planning favors denser, newer nodes that can host more VMs per rack. By removing reservation incentives, the company can gracefully reduce demand for slots on old clusters, eventually reclaiming that floor space and power envelope. The timing aligns with the general availability of Azure Boost—a purpose-built hardware offloading system that requires newer host infrastructure.

The Windows migration imperative

For Windows-centric organizations, this isn’t just a compute refresh. It’s a chance to right-size, modernize operating system versions, and adopt ARM-based options where appropriate. Here’s how to tackle it.

Step 1: Audit your reservation portfolio

Use the Azure portal’s Reservations blade to export all active reservations, filtering by scope and billing subscription. Azure Cost Management can chart your effective rates. Mark any reservation tied to a retiring series. If a reservation extends beyond July 1, 2026, you’re safe for that term—but you won’t be able to renew. If you need capacity past that date, you must migrate before the reservation expires.

Step 2: Map to supported successors

Microsoft publishes a migration guide for each retiring family. Common mappings include:

  • Dv2 → Dv4 or Dv5: The Dv4 introduces Hyper-Threaded Intel Xeon Platinum 8272CL (Cascade Lake), while Dv5 moves to Ice Lake. Both offer better networking and ephemeral disk options.
  • Ev3 → Ev4 or Ev5: Memory-optimized successors. For SQL Server, the constrained-core variants (e.g., E8-2ds_v4) let you license fewer cores while keeping RAM, trimming license costs.
  • NV/NC → NVv3/NCv3: Modern GPUs (Tesla M60 vs. Tesla V100 or A100) deliver orders-of-magnitude improvement in AI and graphics workloads.
  • H-series → HBv3: For HPC, the HBv3 with AMD EPYC Milan-X often outperforms older Intel H-series at lower cost per node-hour.

Microsoft’s TCO calculator and the Azure Migrate tool can automate cost comparisons and right-sizing recommendations. Many workloads on Dv2 can move to smaller Dv5 sizes because of IPC gains, yielding double-digit savings even before applying a new reservation.

Step 3: Time the Windows Server upgrade

If you’re still running Windows Server 2012 R2, which exits extended support in October 2026, this hardware migration aligns perfectly with an OS refresh. Windows Server 2022 (and the upcoming 2025 LTSC) license mobility for Azure Hybrid Benefit allows you to carry on-premises licenses to the newer VMs, potentially stacking reservation savings with BYOL discounts.

Cost scenarios: old vs. new reservations

Consider a US East D8s_v2 3-year reservation taken in 2024. With Azure Hybrid Benefit, the effective monthly cost hovers around $435. The same workload on a D8s_v5 (running Windows Server 2022 Datacenter with BYOL) comes in at roughly $410/month under a 3-year RI—even before factoring in better performance. For a D4s_v2 to D4s_v5 migration, the monthly difference widens from $218 to $205, all while gaining GHz and security improvements.

If an organization ignores the cutoff and leans on pay-as-you-go post-2026, a D8s_v2 jumps to about $770/month (Windows license included), a 77% increase. The penalty is stark.

Reservation exchange and cancellation rules

Existing reservation holders aren’t entirely locked in. Azure’s reservation exchange feature lets you swap your current commitment for a new reservation on a supported family, mid-term, without penalty—provided the combined computation matches. This is the Swiss Army knife for this migration. If you hold a Dv2 reservation running through 2027, you can exchange it for a Dv5 reservation of equivalent compute today, lock in the new rate, and avoid worrying about the mid-2026 cutoff.

Exchanges are proportional: if you exchange halfway through a term, you’ll get a prorated refund applied toward the new reservation. There’s no fee, but you can’t reduce total monetary commitment below the original amount—you can only shift to equivalent or higher spend.

Governance and automation

Cloud governance teams should add a policy that blocks creation of VMs on the retiring families in all non-production subscriptions immediately. A custom Azure Policy can deny requests for sizes like Standard_D2_v2 or Standard_NV6. This prevents developers from accidentally spinning up short-lived workloads that will later become financial liabilities.

Azure Advisor’s cost recommendations will surface VMs on retiring families with warnings to migrate. Use these alerts in tandem with Azure Automation runbooks to snapshot, deallocate, and resize VMs during off-hours. For stateful workloads, Azure Site Recovery can orchestrate failover to the newer hardware with minimal downtime.

The 2028 event horizon

While the reservation cutoff is July 1, 2026, the real deadline for many will be 2028. Three-year reservations on the affected families bought in 2025 will run through mid-2028. If you’re currently in the middle of a three-year term, you’re insulated—but renewals are impossible. The moment that term expires, you move to on-demand rates. So the effective migration window for organizations that purchase a final three-year RI on a Dv2 in 2025 is between July 1, 2026 (last chance to buy) and mid-2028 (term end).

Cloud financial operations managers should model this now. Run a cost projection comparing a new three-year reservation on a successor instance versus riding out an old reservation and then paying on-demand. In nearly every case, starting a new reservation on modern hardware delivers lower total cost by 2029, even if you have sunk cost in an existing term.

What Microsoft hasn’t said

Microsoft’s documentation doesn’t promise that the VMs themselves will be retired by a specific date—only that reservations won’t be sold. Historically, when Azure stops selling reservations, VM availability still persists for several years, but capacity can become constrained. This raises the risk of allocation failures during scale-out events. If your autoscaling rules rely on Standard_D2_v2 instances, you may find insufficient capacity in 2027, leading to application brownouts.

The safest posture is to treat this reservation cutoff as a de facto VM retirement. Most of the affected series have been out of active marketing for years. By migrating now, you avoid the last-minute scramble and gain performance, security, and cost efficiency.

Next steps for Windows IT leaders

  1. Pull a reservation report today: Use the Azure Portal or PowerShell (Get-AzReservation) to inventory all active RIs.
  2. Identify migration candidates: Flag every VM running on a retiring series, particularly those where the Azure Hybrid Benefit is also in play.
  3. Test on successors: Spin up workloads on Dv5, Ev5, and equivalent GPU families in dev/test subscriptions. Validate application compatibility, especially for legacy Windows Server 2012 R2 or custom ISV software that may have hardware dependencies.
  4. Exchange early: If a financial model shows net savings, exchange existing reservations now. Interest rates and compute prices trend upward, so locking in a three-year price on current-gen hardware is a hedge.
  5. Update runbooks and infrastructure as code: Remove retiring SKUs from Terraform, Bicep, or Azure DevOps pipelines. Add safeguards to prevent fallback.

For most Windows-focused cloud teams, the next 18 months will define the platform’s efficiency for the second half of the decade. The reservation cutoff is a forcing function for modernization—and those who act early will capture the biggest gains.