Buy Azure Managed Disk reserved capacity

This is a pure savings lever, not a security or risk control. You are already paying for these Premium SSD disks at pay-as-you-go rates every hour; the reservation simply commits to the steady baseline and buys it at a lower one-year rate. Microsoft's own worked example reserves 100 P30 disks in a region for a one-year term at roughly 140,100 dollars total, payable either up front or as fixed monthly instalments of about 11,675 dollars, versus the higher running pay-as-you-go cost of the same 100 disks. The exact discount depends on SKU and region, so price your own estate with the calculate step before committing rather than assuming a headline percentage.

The decision is a coverage call, not a spend-more call. The right amount to reserve is the number of disks of each SKU you are confident you will run for the whole year. Reserve at or below that floor and the commitment is absorbed every hour, so the saving is close to risk free. Reserve above it and any hour where you run fewer matching disks than you reserved is a use-it-or-lose-it hour you have paid for and not used.

The payback is straightforward to model because disk usage is far steadier than compute. Take Advisor's recommended quantity, confirm it against the disks you know are permanent, run the calculate step to see the committed cost and the discount over pay-as-you-go, and compare the annual reserved cost against twelve months of the current on-demand run rate. For the steady baseline the reserved rate wins for the whole year, which is exactly why this recommendation is one of the cleaner cost wins in Azure.

Priya is the finance partner supporting the platform team. Advisor recommends a managed disk reservation, and the instinct is not to wave through whatever quantity the tool suggests; it is to ask how much of that usage is genuinely permanent. The team confirms that 30 of the 42 Premium SSD P30 disks sit under always-on production VMs and have done for over a year, while the other 12 are on burst and test machines that come and go. The recommendation is sound for the steady 30; the variable 12 should stay on pay-as-you-go so no reserved hour is ever wasted.

Priya then prices the commitment properly before signing anything. The team runs the reservation calculate step for 30 P30 disks in the region, which returns the committed one-year cost and the discount over pay-as-you-go. Following Microsoft's own worked example, a round 100 P30 disks for a year prices at roughly 140,100 dollars total or about 11,675 dollars a month, so 30 disks scales down from that. The finance pack reads: 'We reserve the 30 disks we run every hour of every year at a committed one-year rate, save the discount over on-demand on that baseline, keep the 12 variable disks flexible, and the only commitment is a single one-year term we can model exactly against the current run rate.' Because disk usage is far steadier than compute, the payback is unusually easy to stand behind.

The cost model here is unusually clean for a savings lever. You are already spending this money at pay-as-you-go rates; the reservation lowers the rate on the steady portion in exchange for a one-year commitment, and disk usage is steady enough that the baseline is easy to identify with confidence. Each qualifying P30-or-larger disk that runs all year and is not reserved is a known, recurring overspend equal to the discount you are declining.

The risk in the trade is bounded and self-inflicted: it is over-reservation. Reserve only the disks you are sure will run for the full year and the commitment is absorbed every hour, so the saving is close to guaranteed. Reserve above that floor and the idle reserved hours eat into the saving. The finance role is to anchor the reservation quantity to the demonstrated steady baseline rather than to a usage peak, to run the calculate step so the committed cost and discount are priced before signing, and to keep the variable usage on pay-as-you-go where it belongs.

Some of the storage bill is genuinely variable and some of it is fixed cost dressed up as pay-as-you-go. The Premium SSD disks under the always-on systems fall squarely in the second group: they run every hour of every day at the same SKU. Paying the on-demand rate for that baseline is the equivalent of renting permanent equipment by the hour.

A managed disk reservation converts that steady baseline to a committed one-year rate at a discount, and it does so without migrating, resizing or risking any workload, because nothing about the disks changes except the price. The trade is a known one-year commitment in exchange for a lower rate, and the only real risk is over-committing to disks you will not keep running.

The leadership question is not whether to approve a project; it is whether the team has the disk baseline disciplined enough to commit to it. Reserve the floor you are confident in, leave the variable part on pay-as-you-go, and review coverage as the estate grows. Done that way it is one of the lowest-risk savings decisions in the cloud bill.

After approving a wave of reserved VM instances the previous year, the CFO asked the obvious follow-up at the cost review: if we reserved the servers, why is the storage under them still on the hourly rate? The honest answer was that disk reservations are a separate purchase nobody had got to, so the steady Premium SSD baseline, the disks under the always-on fleet, was still being rented at full pay-as-you-go rates all year.

The team's response was to treat the disk baseline like the compute baseline. They counted the Premium SSD P30 disks that genuinely run every hour, separated them from the burst and test disks, priced a one-year reservation for the steady set, and bought it through a controlled, reviewed purchase rather than a portal click. Nothing about the disks or the workloads changed; the discount simply began applying automatically to the matching disks. The next time the CFO asked, the answer had moved from 'the storage is still on-demand' to 'the always-on disk baseline is reserved at a one-year rate, the variable usage stays flexible, and we review coverage each quarter.' That shift, from renting permanent storage by the hour to committing only to the floor you are sure of, is the confidence signal this recommendation delivers, and it costs nothing in workload risk to reach.

Almost every major cloud cost programme reaches the same realisation: a large slice of the bill is steady-state infrastructure being paid for at on-demand rates, and committing to that baseline at a reserved rate is found money. Reserved VM instances capture this for compute; managed disk reservations capture it for the Premium SSD storage that runs under the same always-on systems, and the two are meant to be bought together.

The cost of getting this wrong is small and recoverable: size the reservation to the steady baseline and the downside is essentially nil, while the only way to lose is to over-commit to disks you stop running. This is the rare class of decision where the saving is real and recurring, the workload risk is zero because nothing about the disks changes, and the only discipline required is honesty about which disks you will actually keep running for a year.