AWS Cost Explorer lets you chart cost as Unblended, Blended or Amortized. What does each metric mean, and which one do you use to explain a month that contained a large upfront Savings Plan or Reserved Instance payment?

## Why there is more than one cost number A single hour of usage can be described honestly in several different ways, and AWS exposes those as selectable *cost metrics* in Cost Explorer. They do not disagree about facts; they disagree about **which account and which time period should carry a charge**. Knowing which one is on screen is the difference between a correct conclusion and a fake trend. ## Unblended cost Unblended is the default and the simplest definition: **the cost as charged, when charged, to the account that incurred it.** If an instance ran for an hour at the On-Demand rate, that is the unblended cost. If a reservation discount applied, the discounted rate is the unblended cost of that hour. If an all-upfront Savings Plan was purchased on the third of the month, the entire fee appears as unblended cost on the third. That last behaviour is what makes unblended both authoritative and misleading. It is authoritative because it sums to the invoice — for reconciliation, unblended is the number. It is misleading for analysis, because one month now carries twelve months of commitment and the following eleven months look artificially cheap. ## Blended cost Blended cost is a **consolidated billing** artifact, and it is the one most candidates get wrong. Inside an AWS Organization, usage across all member accounts is aggregated for volume tiers and for matching reservation discounts. That creates an attribution problem: if two accounts run identical instances and one reservation covers only one of them, which account gets the discount? Unblended answers by whichever line the discount actually matched. Blended answers by **averaging the rate across the family**, so both accounts show the same blended rate and the benefit is spread proportionally to usage. That is useful for internal fairness — no team gets an arbitrary windfall because the billing engine matched their instance first. It is a poor basis for optimization decisions, because a blended rate is an accounting average that no single resource actually paid. Blended cost also has no meaning at all for a standalone account with no organization: it equals unblended. ## Amortized cost Amortized cost redistributes **commitment fees across time**. An upfront Reserved Instance or Savings Plan payment is divided across the hours of the term and attributed to the usage that consumed the commitment in each hour. Recurring hourly commitment fees are treated the same way. The result is that every month shows the effective hourly cost of the workload as if the commitment were being paid smoothly. This is the metric for anything trend-shaped: - **Run rate and month-over-month trend** — unblended trends are unreadable across a purchase. - **Unit economics** — cost per customer, per request, per tenant. An upfront spike would make one month's unit cost nonsense. - **Showback to teams** — a team should carry a steady share of the commitment they benefit from, not a lump because the finance team bought in their month. There are also **net** variants — net unblended and net amortized — which apply negotiated or program discounts on top. "Net" answers the question of what you actually pay after your agreement, whereas the plain metrics show list-relative figures. ## Choosing, in practice A short decision rule: - Reconciling to the invoice, or explaining a specific charge to finance → **unblended**. - Trending, forecasting, unit cost, chargeback → **amortized**. - Comparing accounts inside one organization for fairness → **blended**, and only for that. - Anything where a negotiated discount matters → the **net** variant of the above. ## The failure this prevents The classic incident is a leadership escalation: "cost jumped 40% in March." Someone opens Cost Explorer on the default metric, sees the spike, and starts hunting a runaway workload. There was none — an all-upfront commitment was purchased in March. Switching the metric to amortized flattens the chart entirely and shows the run rate actually fell, because the commitment was doing its job. The reverse failure also happens: a team congratulates itself on a cheap quarter because an earlier upfront payment is not visible in unblended cost for those months. The habit to build is simple. Before drawing any conclusion from a Cost Explorer chart, look at which cost metric is selected, and ask whether the question you are answering is about *what was charged* or about *what the workload costs to run*.