Azure Reserved Instances commit the buyer to a specific compute capacity for one or three years in exchange for a discount that can reach seventy two percent against pay as you go on the same SKU. The savings are real. The wrong portfolio shape locks the buyer into capacity that the workload no longer needs by year two. Most enterprises run their RI portfolio as a spreadsheet of standing decisions rather than as a managed asset class. RIs are a portfolio management discipline first and a discount mechanic second.
An RI commits a defined compute resource at a defined capacity for one or three years. The discount applies against the pay as you go rate for matching consumption. Unmatched RI capacity earns no discount. The matching engine in Azure is sophisticated. The buyer who understands the matching engine extracts meaningfully more value than the buyer who treats the RI as a standalone purchase.
One year RIs produce a discount in the high thirties to mid forties against on demand on most compute SKUs. The shorter term is the right answer where the workload shape will change within twelve months or the platform team is still validating the long term sizing decision. The discount is meaningful and the lock in is bounded.
Three year RIs produce the headline discount of up to seventy two percent on standard compute SKUs. The right answer for genuinely stable workloads. The wrong answer for workloads with a credible chance of being retired, refactored, or migrated within the three year window. The discount is large and the commitment is real.
Azure RIs purchased with instance size flexibility match against any instance in the same series and region. The flexibility is the difference between an RI portfolio that stays matched at ninety five percent utilization and one that drifts to seventy. Most enterprises do not configure RIs for flexibility because the default purchase flow does not surface the option.
An RI on a D4s v5 matches a D2s v5, a D8s v5, or any size in the D v5 series at the appropriate ratio. The match is automatic. The buyer who purchases at the middle of the series captures matching against the workload up sizing and down sizing without forfeiting discount on either.
Shared scope RIs match across the entire billing account. Single subscription scope is the default. The shared scope produces meaningfully higher utilization because the matching engine sees more candidate consumption. The default is the wrong choice for most enterprises.
Until early 2024 Microsoft permitted free exchange of RIs into different SKUs through the term. The exchange right was removed for new purchases. Existing RIs purchased before the change retain their exchange rights through term. The change made the upfront sizing decision much higher stakes than it used to be.
An RI portfolio that produced eighty five percent utilization at purchase often produces sixty five percent utilization eighteen months later because the workload shape moved. The drift is invisible in the monthly bill because the on demand spend covers the gap. The drift is visible in the per unit economics across the portfolio.
The platform team upgrades to a newer compute series. The RI portfolio remains on the older series. The new consumption attracts no RI discount. The old RIs continue to bill at the committed rate against shrinking matching consumption. The portfolio utilization drops quietly across two quarters and the per unit economics get worse than pay as you go would have produced.
The defense is portfolio review on a quarterly cadence. RI utilization tracked by series. Recommendations modeled against the active consumption pattern. Exchanges executed where the existing rights permit. Cancellations modeled where the economics support the early exit. The portfolio stays matched to the actual workload.
Azure permits cancellation of RIs with a fifty thousand dollar annual cap on refunded amounts per billing account. The cap is small relative to most enterprise portfolios. The buyer that needs to exit a misaligned RI position in year two often finds the cancellation economics worse than continuing to pay against unused capacity.
The defense is sizing discipline at purchase. Three year decisions only on workloads that survive a three year refactor decision. One year RIs and Savings Plans cover the workloads where the shape is less certain. The exit is engineered into the entry decision.
A managed RI portfolio is three layers. A baseline of three year RIs against the workloads that genuinely persist. A middle layer of one year RIs against the workloads with a twelve month outlook. A coverage layer of Savings Plans across the elastic capacity. The mix produces a higher blended discount than any single layer at lower risk of stranded capacity.
The persistent baseline of the production estate. Always on infrastructure that has been stable for two refresh cycles and will be stable for two more. Database hosts, identity infrastructure, network function appliances. The three year RI captures the deepest discount on the most predictable capacity.
The baseline decision sits inside the broader EA renewal envelope because the baseline produces the floor on the MACC drawdown. The two negotiations interlock.
The middle layer covers workloads with a credible twelve month shape but less certainty beyond. New production workloads in their first stable year. Workloads under planned refactoring that will not complete inside twelve months. The one year RI captures meaningful discount without the three year lock.
The tactical layer rolls quarterly. The portfolio review identifies the workloads that earned a graduation to three year and the workloads that should drop to Savings Plan coverage instead.
The RI engagement is a portfolio diagnostic, a baseline and tactical sizing decision, a Savings Plan complement, an exchange and cancellation analysis on the existing portfolio, and ongoing quarterly portfolio management for the engagement period.
We pull the existing RI portfolio across all subscriptions. We model the utilization against the trailing ninety day consumption pattern. We identify the under utilized positions, the drift, the eligible exchanges, the cancellation candidates. The output is a baseline of the current portfolio and a remediation list for the underperforming positions.
We model the right three year baseline, the one year tactical layer, and the Savings Plan complement against the consumption forecast. The output is the portfolio that produces the highest blended discount at the lowest stranded capacity risk. We coordinate the purchase execution and stand up the quarterly portfolio review discipline for the engagement period.
The RI diagnostic surfaces drift, sizing errors, exchange opportunities, and the forward portfolio that produces the highest blended discount across the workload mix. The result is a meaningfully better effective unit price on the compute layer.