Cost Predictability: Why Cloud Pay-As-You-Go Breaks the IT Forecast

Cloud was sold as a cost story. It is now a predictability story. Pay-as-you-go pricing was designed to track usage, which means it was also designed to be volatile. For a CIO defending a 12-month forecast to a CFO, that volatility is the binding constraint, not the absolute amount.

Five drivers that break the forecast

Compute spikes

Autoscaling responds to load, but load is non-linear: seasonal traffic, AI inference bursts, batch retries, multi-tenant noisy neighbours. Only 2% of CIOs report spending less on cloud than they projected; 89% of CFOs say rising cloud costs have eroded company profitability (State of FinOps 2026; CFOTech 2025).

RAM and capacity sizing

Reserved-versus-on-demand decisions are made under uncertainty. 84% of organisations name managing cloud spend their #1 cloud challenge; estimated wasted IaaS and PaaS spend is 27%; cloud budgets are exceeded by 17% on average (Flexera 2025 State of the Cloud Report).

Ingress and egress data transfer

Bytes moving out of the cloud and between regions or availability zones are billed per gigabyte. 95% of IT leaders have encountered unexpected storage and transfer charges; 55% name egress the single biggest barrier to switching providers. AWS, Azure and GCP charge approximately 0.085 to 0.09 USD per gigabyte outbound.

AI workload pricing

Training spikes, usage-driven inference, experimentation noise. Only 6% of IT organisations report clear measurable value from AI today (38% in pilots, 32% partial production) per the WNDRVR field survey 2026.

Month-to-month variance

Standard month-to-month variability of 5% to 10% is out of line with what CFOs expect from any other major expense. 44% of CFOs name improving cloud cost forecast accuracy as their top 2026 focus.

The 37signals case study

37signals (Basecamp, HEY) spent 3.2 million USD per year on AWS in 2022. After moving seven applications onto its own hardware, the 2024 cloud bill fell to about 1.3 million USD per year, with infrastructure spend projected to drop well under 1 million USD per year and total projected savings exceeding 10 million USD over five years, no staff added.

What on-premises actually changes

On-premises does not magically eliminate cost growth. What it eliminates is the volatility: a single platform line, a known compute and storage envelope, a 12-month-forward number a CFO can defend at the next budget review. Less fragmentation, more control at scale, and a platform line you can take to the CFO twelve months forward.

For the architectural side of the same problem, see The Cost of Fragmentation.

Solutions

Platform

Articles

Coelion home