Cost Optimization Glossary

The cost to run trained models in production — API calls, GPU compute, and hosted endpoints — distinct from one-off training spend.

Discounted cloud rates in exchange for spend or capacity commitments — reserved instances, savings plans, and committed use discounts.

On-demand access to compute, storage, and managed services over the internet — billed by consumption rather than upfront capital purchase.

Attributing cloud spend to teams, products, or environments using tags, accounts, and policies — the basis for showback and chargeback.

The discipline of governing cloud spend — visibility, allocation, commitments, optimization, and accountability across engineering and finance.

Metadata applied to cloud resources (env, owner, service) that drives accurate showback, chargeback, and budget alerts.

Monitoring that flags unexpected spend spikes in near real time — before they land in the month-end close.

Projecting future technology spend from usage trends, commitments, and planned workload — not from last month's bill alone.

How far along an organization is in cloud financial management — from crawl to walk to run across inform, optimize, and operate.

AI systems that create text, code, images, or other content from prompts — typically priced via APIs, seats, or bundled SaaS features.

Up-front technology investments capitalized over time — datacenter builds, major license purchases, and transformation programs.

Assigning total IT spend to business units, products, or services — so owners see what they consume and finance can govern fairly.

Governing total technology spend as a financial discipline — planning, allocation, optimization, and reporting across cloud, SaaS, AI, and vendors.

Run-rate technology spend charged to the P&L — subscriptions, cloud consumption, support, and managed services.

AI models trained on large text corpora that power chat, coding, and search assistants — usually billed per token or via enterprise seat licenses.

Reviewing SaaS seat allocation and tiering to reclaim unused, duplicate, or over-provisioned licenses at renewal.

A commitment-based discount on cloud compute in exchange for a 1- or 3-year term — typically 30–60% off on-demand pricing.

Matching compute capacity to actual workload demand to eliminate over-provisioning.

Software delivered as a subscription over the internet — licensed per seat, usage, or enterprise agreement rather than as perpetual on-premise installs.

Uncontrolled growth of SaaS subscriptions across teams — duplicate tools, unused seats, and renewals without central visibility.

A flexible commitment discount covering a steady compute spend rate (USD/hr) across instance families, regions, and sometimes services.

SaaS tools purchased outside IT and procurement visibility — team cards, expensed subscriptions, and trials that auto-convert.

Attributing cloud and SaaS cost back to the teams, products, or cost centers that consume it — showback informs, chargeback bills.

Tracking entitlements, seat utilization, and renewals across software — reclaiming unused capacity and right-sizing tiers before vendors set terms.

Discounted cloud compute that can be reclaimed at short notice — ideal for fault-tolerant, interruptible workloads.

The full lifetime cost of a system — licenses, compute, storage, networking, people, migration, and exit — not just the cloud bill.

Per-token pricing charged by LLM providers for input and output — the primary cost driver for AI applications at scale.

Billing for LLM API usage by tokens processed — input and output text converted to billable units that scale with every request.

Cost per business unit — per order, per tenant, per active user — so efficiency becomes a trackable engineering outcome.