A 25 percent discount can leave you paying 50 percent more.
No hidden fee is required.
Consider a hypothetical GPU purchase. The on-demand rate is $4 per GPU-hour. A commitment reduces it to $3, provided you pay for all 100 GPU-hours in the agreed period.
You end up needing 50.
The committed purchase costs $300. Buying those same 50 GPU-hours on demand would have cost $200.
The discount is genuine. So is the extra hundred dollars.
Assume equivalent hardware, performance and service, available on-demand capacity, no additional charges, and the ability to terminate on-demand instances whenever they are no longer needed. This illustrates a fixed purchase obligation, not every product sold under the word “reserved.”
Under those conditions, the arithmetic is exact. The buyer secured a lower rate and incurred a higher bill.
The critical number was never the discount alone. It was how much of the commitment the buyer would use.
In this example, the purchase breaks even at 75 GPU-hours. Above that threshold, within the agreed 100 GPU-hours, the commitment saves money. Below it, the buyer would spend less purchasing only the capacity required.
The discount puts a price on the buyer’s confidence in future demand.
A related commercial mechanism is documented in AWS Savings Plans. These exchange an hourly spending commitment for discounted eligible usage. Each hour’s commitment must be used within that hour; unused commitment cannot be carried forward. This differs from the fixed GPU-hour purchase above, but illustrates the same exposure: demand can fall while the payment obligation remains. AWS Savings Plans documentation
Flexibility also has to be exercised. Lambda’s on-demand billing documentation states that instances are charged while running, whether actively used or not. Instance billing ends upon termination. Buying on demand does not itself prevent paying for idle time. Lambda billing documentation
Termination gives the buyer a way to stop future instance charges. The fixed commitment in our example remains payable.
That distinction deserves a place beside the hourly rate. Under a fixed commitment, the provider receives a payment obligation and the buyer bears the cost of the portion left unused. A lower rate can compensate the buyer for accepting that exposure. Whether the compensation is sufficient depends on what happens next.
For the hypothetical purchase above, knowing that 100 GPU-hours of demand will arrive makes the $3 rate attractive. Needing only 50 makes the same agreement expensive. The provider, hardware and contracted rate have not changed. The buyer’s requirement has.
This is why “What discount did you get?” can be the wrong first question.
A more revealing one is: “What must happen for this agreement to save money?”
How much capacity must be used? Over which period? Can unused entitlement move between workloads or accounts? What happens if the project is delayed? Can the buyer reduce the commitment, and at what cost?
Those answers describe the bargain more fully than the percentage beside it.
The comparison must also establish which CPU, memory, storage and networking resources the rate includes, together with any technical support, managed services or service guarantees specified in the agreement. Separately billed services belong in the calculation too: the break-even result above assumes equivalent purchases, and a GPU-hour alone does not establish that equivalence.
Sorso View’s recent observations offer a useful contrast. Fourteen providers each showed the same recorded public rate across Issues 007 through 011. That numerical persistence is a finding about the published offers we tracked. It cannot settle whether a particular commitment is economical for a particular buyer.
Even an unchanged rate can support different purchasing outcomes.
The question of what customers are actually billed still matters. But collecting invoices without understanding the obligations behind them would leave us vulnerable to another misleading comparison. A low contracted rate spread over unused capacity can tell a different story from the same rate applied to capacity fully used.
For Sorso View, this suggests a practical addition to the public-price record: document the purchase obligation and included services alongside the price, then calculate the conditions under which an offer becomes economical. Where the terms are unclear, leave the calculation unresolved.
There is no universal verdict against commitments here. Predictable demand can make them valuable. Capacity assurance, where separately provided, can also justify a cost that an hourly comparison misses.
The decision belongs in the relationship between the rate, the obligation, the service and the buyer’s expected use.
A discount percentage describes how far a price has fallen. It does not describe how much the buyer will spend to obtain it.
Before celebrating the rate you negotiated, count the hours you promised to buy.