Buying Fuel in Bulk, Ahead of Time

Opening Scene

A shipping company that reliably runs the same route year-round can negotiate a bulk fuel supply contract in advance, locking in a meaningfully lower rate in exchange for committing to a predictable, sustained volume. This only works because the usage is genuinely predictable — negotiating a bulk contract for an unpredictable, occasional voyage would be a poor trade. Reserved instances and savings plans in the cloud follow this exact same logic.

In Plain English

Reserved instances and savings plans let organizations commit to a minimum, sustained level of cloud usage over a fixed term — typically one or three years — in exchange for a substantially discounted rate compared to on-demand pricing. The discount can be significant, often 30-70% depending on the commitment length and cloud provider, but the commitment only pays off genuinely when actual usage is reliably sustained.

The Old Way

Before reserved capacity purchasing was a well-established, data-driven practice, organizations often approached this decision less rigorously:

Organizations sometimes purchased reserved capacity based on rough estimates of future need, rather than genuine analysis of sustained, historical usage patterns.
There wasn’t yet a well-established practice of continuously monitoring reserved capacity utilization to ensure commitments genuinely matched actual usage.
Reserved capacity purchases were sometimes made once and left unreviewed, even as actual workload patterns shifted meaningfully over time.

Estimate-based, one-time reserved capacity decisions, without continuous utilization monitoring, is what disciplined committed-use purchasing practice replaced.

What’s Changing (and Why AI Is the Reason)

Organizations increasingly use historical usage data and forecasting tools to size reserved capacity commitments precisely, rather than relying on rough estimates.
This connects directly to the on-demand versus committed-use tradeoff covered in Article 3, applying that framework specifically to sustained, predictable workloads.
As AI training workloads often run predictably for extended, sustained periods, reserved capacity and savings plans have become an increasingly significant, material savings opportunity specifically for AI infrastructure spend.

The Metaphor, Fully Extended

The Ship’s Engineer	Cloud FinOps Concept
Negotiating a bulk fuel contract for a reliable, year-round route	Committing to reserved instances or savings plans for predictable, sustained usage
A meaningfully lower rate in exchange for a committed volume	A substantially discounted rate in exchange for a committed usage term
Only a good trade when the route’s usage is genuinely predictable	Only a good trade when workload usage is genuinely sustained and predictable
Reviewing the contract as the route’s actual needs shift over time	Continuously monitoring commitment utilization as workload patterns shift

For Beginners: What to Actually Do

Practice reviewing your cloud provider’s documentation on reserved instances and savings plans to understand the basic discount structures available.
Learn to distinguish workloads that run predictably and sustained from those that are genuinely variable or occasional.
Get comfortable with the idea that committed-use discounts are a genuine tradeoff, not a universally correct default choice.

For Practitioners and Leaders: The Deeper Layer

Use historical usage data and forecasting to size reserved capacity commitments precisely, connecting directly to the cost allocation practice covered in Article 4.
Build continuous monitoring of reserved capacity utilization into standard FinOps practice, adjusting commitments as workload patterns shift.
Prioritize reserved capacity evaluation specifically for sustained AI training infrastructure, where the savings opportunity is often especially significant.

Quick Recap

Reserved instances and savings plans trade a committed usage term for a substantially discounted rate.
This trade only pays off genuinely when actual usage is reliably sustained and predictable.
Precise, data-driven sizing and continuous monitoring are essential, not one-time estimate-based decisions.
Sustained AI training workloads present a particularly significant reserved capacity savings opportunity.

Where This Fits in the Series

Article 7 covered committing to capacity in advance for a discount. Article 8 turns to the opposite trade: fuel that’s meaningfully cheaper, but can run out without warning.