Research Note · Cloud · Cost Comparison

AWS vs Azure vs GCP: 2026 enterprise pricing comparison.

List pricing favours GCP marginally on most compute line items, Azure on storage, and AWS on egress at low tiers. Net of negotiated commitments — EDP, MACC, CUD — the realised cost gap for equivalent workloads is typically under 15%. The real decision is which commitment vehicle fits your workload profile, and how to use multi-cloud presence as leverage.

By James Hill-WoodUpdated Dec 20249 min readCloud research cluster
Bottom line

There is no single "cheapest" hyperscaler. For a 100,000 vCPU-hour/month general-purpose Linux workload, on-demand list cost runs $8,400–$11,200 on AWS, $7,900–$10,600 on Azure, and $7,600–$10,200 on GCP — but net of EDP, MACC and CUD commitments the realised gap is typically under 15%. Choose Azure for Microsoft-heavy estates, AWS for breadth and predictable terms, GCP for compute- and analytics-led workloads. Documented multi-cloud presence is worth 10–25 points of extra renewal discount.

01 Key findings

  1. List price is not the decision. GCP leads marginally on most compute line items, Azure on storage, AWS on low-tier egress — but after commitment discounts the realised cost gap for equivalent workloads is typically less than 15%.

  2. Three different commitment models. AWS EDP and Azure MACC are similar multi-year monetary commitments (5–25% off retail); GCP uses per-service committed use discounts (CUDs) that reach up to 70% on Compute Engine but with far less flexibility.

  3. GCP's deeper headline discounts are not always deeper realised discounts. Workloads that do not fit cleanly into Compute Engine CUDs — variable demand, frequent machine-type changes, multi-region — realise lower effective savings than the 40–65% headline suggests.

  4. Egress is the most distorting line item. AWS/Azure charge $0.087–$0.09/GB at the first tier; GCP charges $0.11–$0.12/GB. For content- and data-distribution workloads egress can exceed compute cost, and commitment vehicles deliver almost no relief.

  5. Multi-cloud presence is the highest-value lever. A credible, portable 15–25% secondary-cloud footprint delivers 10–25% improvement over the standard EDP/MACC/CUD renewal trajectory.

02 Cost scorecard

Relative commercial strength across the dimensions that move enterprise cloud cost. Five dots = strongest; scoring reflects commercial and cost posture, not raw technical capability.

Dimension
AWS
Azure
Google Cloud
Compute list price
Discount programs
Egress economics
Ecosystem leverage
Enterprise terms & predictability

03 List pricing for equivalent compute and storage

Equivalent general-purpose configurations across the three clouds. List prices favour GCP marginally on most line items, Azure on storage, and AWS on low-tier egress. After commitment discounts the realised gap narrows substantially.

ServiceAWS (us-east-1)Azure (East US)GCP (us-central1)
General-purpose 4 vCPU/16GB Linux (on-demand, per hour)$0.166 (m6i.xlarge)$0.152 (D4s_v5)$0.149 (n2-standard-4)
1-year reserved/savings (no upfront)$0.105$0.097$0.094
3-year reserved/savings (all upfront)$0.061$0.057$0.056
Block storage (GP3/SSD, per GB/month)$0.080$0.075 (Premium SSD v2)$0.068 (pd-balanced)
Object storage standard tier (per GB/month)$0.023$0.0184$0.020
Egress to internet (first 10 TB/month, per GB)$0.09$0.087$0.12
Egress to internet (next 40 TB/month, per GB)$0.085$0.083$0.11
NAT gateway (per hour)$0.045$0.045 (NAT Gateway)$0.044
Load balancer (per hour)$0.0225 (ALB)$0.025 (Standard LB)$0.025 (HTTPS LB)

The structurally important differences are not in headline rates but in the commitment vehicles and in service-specific pricing for managed databases, AI and analytics.

04 Commitment vehicles: EDP, MACC, CUD

AWS EDP and Azure MACC are commercially similar multi-year monetary commitments drawn down through normal consumption. GCP's model is more granular — per-service committed use contracts, with Compute Engine CUDs offering the deepest discounts but the least flexibility.

Commitment vehicleAWS EDPAzure MACCGCP Commitments
FormMulti-year spend commitment (typically 3-year)Multi-year monetary commitment (typically 3-year)Per-service committed use discounts
Minimum size$1M to $3M per year$1M per yearNo minimum; per-service
Typical discount5 to 20% off retail5 to 15% off retail (plus MACC-eligible service discounts)20 to 70% off (compute CUDs); 25% off (flex CUDs)
CoverageMost AWS services credit toward commitMost Azure services credit; some PaaS excludedPer-service; less holistic than AWS/Azure
FlexibilityAnnual ramp; renegotiation at term-endMonthly drawdown; shortfall billed at term-endConvertible between machine types (CUDs); per-service contracts otherwise
Exit termsPenalty for unused commitPenalty for unused commitPenalty for unused per-service commit

Discount realisation scales with deal size, but GCP's deeper headline numbers reflect aggressive positioning and the per-service CUD model — not necessarily a lower realised cost. AWS EDP and Azure MACC discounts are smaller but apply more uniformly.

Annual cloud spendAWS EDP discountAzure MACC discountGCP achievable
$1M to $3M5 to 10%5 to 10%15 to 30% (with CUDs)
$3M to $10M8 to 15%8 to 12%25 to 45%
$10M to $50M12 to 20%10 to 15%35 to 55%
$50M+15 to 25% (custom)12 to 20% (custom)40 to 65% (custom)

05 Service-specific pricing differences

Headline compute and storage rates obscure the more important pricing decisions for enterprise workloads, which sit in managed services. The notable per-service differences:

Service categoryPricing characteristic
Managed PostgreSQL (RDS / Azure Database / Cloud SQL)AWS RDS PostgreSQL ~$0.193/hr (db.m6i.xlarge). Azure Database ~$0.232/hr (GP_Standard_D4ds_v5). Cloud SQL ~$0.150/hr (db-custom-4-16384). GCP lower; Azure higher.
Managed Kubernetes (EKS / AKS / GKE)EKS control plane $0.10/hr + cluster cost. AKS control plane free (standard tier). GKE Autopilot per-pod billing or $0.10/hr per cluster. AKS cheapest for many workloads.
AI/ML training (GPU instances)NVIDIA H100 hourly: AWS $98 list (p5.48xlarge), Azure $98 list, GCP $88 list. Realised pricing varies 30 to 60% below list with capacity reservations.
Object storage long-term archiveGlacier Deep Archive $0.00099/GB. Azure Archive $0.00099/GB. GCP Coldline $0.004/GB, Archive $0.0012/GB. AWS/Azure cheaper for very long retention.
Data warehouseSnowflake (multi-cloud) $2 to $4 per credit. Redshift $0.25 to $13.04 per node-hour. Synapse Dedicated Pool $1.20/DWU100/hr. BigQuery on-demand $6.25/TB scanned or flat-rate slot reservations. BigQuery typically cheapest for variable-demand analytics.

06 Provider profiles

AWS
Broadest catalogue
Best for: greenfield modernisation, the widest service portfolio, and buyers who value predictable, benchmarkable EDP terms.
Strengths
  • Largest service catalogue and partner ecosystem
  • Mature Reserved Instance & Savings Plan instruments
  • EDP credits most services; benchmarks reliably
Limitations
  • Highest compute list rates of the three
  • Commitment structures penalise underspend
  • Egress high; limited relief from commitments
Azure
Microsoft ecosystem leader
Best for: Microsoft-heavy estates — Office 365, Dynamics, Windows Server and SQL Server workloads.
Strengths
  • Azure Hybrid Benefit for Windows & SQL licences
  • Cheapest storage; free AKS control plane (standard)
  • MACC credits Marketplace and eligible services
Limitations
  • Higher managed-database rates (~$0.232/hr PostgreSQL)
  • Complex discount stacking obscures net price
  • Shortfall billed at term-end on MACC
Google Cloud
Data & compute-led value
Best for: compute-intensive and analytics-led workloads that fit CUDs and can model economics net of incentives.
Strengths
  • Lowest compute list rates on most line items
  • Compute Engine CUDs up to 70% off
  • BigQuery economics; Sustained Use Discounts apply automatically
Limitations
  • Highest egress ($0.11–$0.12/GB)
  • Per-service CUDs less flexible, less holistic
  • Headline discounts can mask weaker realised savings
Discount-program trap

GCP's headline is not the realised number. Compute Engine CUDs advertise up to 70% off, but they lock to specific machine types or regions unless you use flex CUDs (25%). Workloads with variable demand, frequent machine-type changes, or multi-region deployments realise materially lower effective discounts — model the blended rate net of the portion of spend that will actually sit under a CUD, not the headline maximum.

07 The egress trap

Egress is the single most distorting line item in enterprise cloud bills and the area where commitment vehicles deliver the least relief. AWS and Azure charge $0.087–$0.09/GB to the internet at the first tier, dropping with volume; GCP charges $0.11–$0.12/GB. For content- and data-distribution workloads, egress frequently exceeds compute cost. First-10TB internet-egress list rates:

Azure
$0.087/GB
AWS
$0.090/GB
Google Cloud
$0.120/GB
Mitigation

Reduce effective egress with CDN integration (Cloudflare and similar), private interconnect (AWS Direct Connect, Azure ExpressRoute, GCP Cloud Interconnect), and architectures that minimise cross-cloud and cross-region movement. The 2024 EU Data Act egress waivers are scoped — typically 100 GB free per month for active customers, full waiver only on customer-initiated migration — and do not reduce egress cost during normal operation.

08 Decision framework

Match the primary-provider decision to your estate, workload profile and certification needs before committing.

Customer situationPrimary cloud recommendation
Microsoft-heavy estate — Office 365, Dynamics, Power PlatformAzure (deep MACC discounting; Azure Hybrid Benefit for Windows and SQL)
Data- and analytics-led workload, large BigQuery candidateGCP (BigQuery economics; Vertex AI; data residency in 35+ regions)
Broadest service portfolio, largest partner ecosystemAWS (default for greenfield enterprise modernisation)
Regulated industry with specific certificationsRegion/service-specific: AWS GovCloud, Azure Government, GCP Assured Workloads
SAP-led estateAWS or Azure (both certified for SAP HANA capacity; GCP certified but smaller footprint)
Oracle-licensed workload using Oracle supportOCI or AWS/Azure with BYOL; see AWS vs Oracle Cloud pricing

09 Our recommendation

Choose AWS
When breadth & predictability win

You need the widest catalogue and formulaic EDP terms you can benchmark. Accept the higher compute list rates, then push hard on M&A flexibility and egress credits — the two default weak points.

Choose Azure
When Microsoft-heavy

You run material M365, Windows or SQL estates. Quantify Azure Hybrid Benefit, take the cheaper storage, and force a single joint software-plus-cloud MACC negotiation rather than two separate ones.

Choose GCP
When compute & analytics lead

You have compute-intensive or BigQuery-heavy workloads and can model economics net of incentives. Capture CUD depth — but confirm the workload actually fits a CUD and budget for higher egress.

The most expensive cloud relationship is a single-cloud relationship with no credible alternative. Maintain a meaningful, architecturally portable secondary footprint (typically 15–25% of spend); at EDP/MACC/CUD renewal, present the primary provider with the choice of improved terms or migrated workloads. Outcomes during 2024–2026: 10 to 25% improvement over the standard renewal trajectory.

Run a competitive multi-cloud process

Our Cloud & FinOps practice coordinates benchmarking, commitment sizing and timing across all three providers at once.

Request cloud advisory →

The Licensing Edge

Weekly cloud and licensing intelligence for enterprise IT leaders. 3,000+ subscribers.