AI Storage Cost Index

Methodology

Overview

The AI Storage Cost Index (ASCI) is a composite index tracking cloud storage costs relevant to AI/ML workloads. It aggregates daily pricing data from AWS, Azure, and GCP across 114 regions and 9 storage types, weighted by provider market share and storage tier relevance to AI infrastructure.

Data Collection

Storage pricing data is collected daily from all three major hyperscalers — approximately 5,000 records per day covering 9 standardised storage types across 114 cloud regions. Records with unmapped storage types or prices exceeding $10/GB/mo are excluded as likely data quality issues.

Provider Weighting

Providers are weighted by their share of the AI compute market (c.63% combined):

• AWS — 30% market share (47.6% normalised weight)
• Azure — 20% market share (31.7% normalised weight)
• GCP — 13% market share (20.6% normalised weight)

For each sub-index, the median price is computed within each provider's regions first (avoiding bias from providers with more regions), then a market-share-weighted average is computed across participating providers. When a provider has no data for a given sub-index, its weight is redistributed proportionally.

Sub-Index Composition

The composite ASCI is a weighted average of four sub-indices, each tracking a storage tier relevant to AI workloads:

• Performance SSD (40%) — SSD Premium, SSD Ultra, SSD Hyperdisk. Primary storage for training data, model checkpoints, and weight files requiring high IOPS and throughput.
• General Purpose SSD (30%) — SSD GP. General dataset storage, model repositories, and inference asset hosting.
• Archive / HDD (20%) — HDD Standard, HDD Cold, HDD Throughput, Standard. Archival storage for large datasets, training logs, and cold backups of model checkpoints.
• Local SSD (10%) — Local SSD. Ephemeral scratch storage attached directly to GPU instances during active training runs.

Index Normalisation

All sub-indices are normalised to a base value of 100 on January 11, 2026 — the first date with complete three-provider coverage. Values above 100 indicate prices have risen relative to the base date; values below 100 indicate prices have fallen. Days with data from fewer than two providers are excluded from the chart.

Skew Adjustments

• Region count imbalance — Azure has approximately twice as many regions as AWS or GCP. Computing medians within each provider before weighting prevents Azure from dominating the index.
• Provider-specific storage types — Some storage types (e.g. SSD Hyperdisk) are offered by only one provider. Sub-index weights are renormalised across participating providers to maintain representativeness.
• Outlier exclusion — Records with unmapped storage types (“-”) and prices exceeding $10/GB/mo are excluded. The unmapped category contained prices up to $397/GB/mo, likely reflecting different billing units or data quality issues.
• Incomplete coverage days — Days where only a single provider reported data are flagged and excluded from the trend line to prevent single-provider bias from affecting the composite value.

Last updated: March 2026. Data sourced from 60 complete collection days.