At a glance
Global cloud market
Exhibit 1The big three are growing faster than ever
Cloud revenue growth, Q3 2026 vs Q3 2025
Google Cloud includes Workspace; figures are not strictly comparable.
The category on a page
Exhibit 2Where cloud sits
Profit impact against supply risk, technology subcategories
Exhibit 3Cloud waste is rising again
Estimated share of IaaS/PaaS spend wasted
Exhibit 4Market share
Share of cloud infrastructure services, Q3 2026
Exhibit 5GPU prices are going up
AWS p5e.48xlarge (8× H200) Capacity Block, $ per hour
A further ~20% rise on Capacity Blocks took effect on 1 July 2026.
Exhibit 6Contracted backlog is huge
Committed future revenue, latest disclosed, $bn
Providers are capacity-constrained and have little need to discount.
Typical savings from FinOps and commitments
On IaaS/PaaS spend
Value levers
In order of value.
Right-size, shut idle resources and schedule non-production. AI workloads are making waste worse.
Use savings plans and reservations for steady usage; keep spiky and experimental AI on demand or spot.
GPU prices differ widely by provider and model; neoclouds and spot can be far cheaper.
Commit total annual spend in exchange for a % discount, credits and migration support.
Arm-based instances (Graviton) and custom AI chips (TPU, Trainium) often cost less per unit of work.
How to buy it
KPIs to track
For the monthly review
Where most buyers are, and what good looks like
| Most buyers today | Best practice | ||
|---|---|---|---|
| Visibility | Monthly invoice | → | Every resource tagged to an owner |
| Commitments | Bought once a year | → | Managed continuously, matched to baseline |
| AI | Pooled in IaaS line | → | AI cost forecast and owned separately |
| Architecture | Lift and shift | → | Right-sized, scheduled, cheaper chips |
| Contract | Standard terms | → | Commit flexibility, egress and exit terms |
What AI changes
More in the AI disruption tab.
Full written analysis
Written live for Cloud & hosting and your region, grounded on the figures above. Public data only.
Written live for this category and region when the page opens in a browser.
Negotiation playbook
Negotiation playbook
Providers are capacity-constrained and growing 39-103%, so they don't need to discount. Your leverage is committed spend, multi-cloud credibility and timing at their quarter-end.
Pick the right commercial model
| Enterprise commit (EDP / MACC / CUD)Usual best | Commit annual spend for a discount and credits. Best for stable baseline. |
| Savings plans / reservations | 1-3 year commitments on compute for 20-60% off on demand. |
| On demand | Full flexibility; most expensive. |
| Spot / preemptible | Deep discounts for interruptible work; needs checkpointing. |
Questions to ask suppliers
- Commit flexibility
If we don't use our full commitment, can we roll it over or apply it to other services?
Good answer: roll over unused commitment; apply across services. - Credits
What migration and AI credits will you give in year one?
Good answer: migration and AI credits in year one. - Egress
Will you cap or waive data transfer (egress) costs if we leave?
Good answer: capped or waived data transfer costs on exit. - GPU capacity
Can you reserve GPU capacity for us at an agreed price, and for how long?
Good answer: reserved capacity at an agreed price. - Price protection
Will you guarantee no list-price increases on the services we commit to?
Good answer: no list-price increases on committed services.
Your leverage
- Total annual commit and growth
- Multi-cloud or neocloud alternative for AI
- Provider quarter-end timing
- Reference value of your migration
Watch out for
- Commitments sized on optimistic forecasts
- GPU price rises on reservations
- Egress costs that trap data
AI disruption
Tailwinds for buyers
- AI FinOps tools find idle and oversized resources automatically.
- Custom AI chips (TPU, Trainium) offer cheaper alternatives to Nvidia GPUs.
- AI agents can manage commitments continuously.
- AI-powered FinOps tools cut wasted cloud spend.
- Rightsizing and auto-scaling become easier to automate.
- Better forecasting improves commitment purchases.
Headwinds for buyers
- GPU prices rising: AWS up 20-50% on reservations in 2026.
- AI workloads are hard to right-size, raising waste to 29%.
- Proprietary AI services deepen lock-in.
- GPU and AI capacity is scarce and priced high.
- AI workloads drive cloud bills up fast.
- Hyperscalers prioritise large AI customers for capacity.
How to capture it
Give AI workloads their own budget and owner, use AI FinOps tooling, and compare GPU prices across providers and neoclouds before committing.
Who is changing the market
- GPU cloud gives an alternative to the big three for AI work
- Competition on GPU price and availability
- AI cloud capacity in Europe and the US
- Another supplier to tender for GPU capacity
- Automates savings plans and reserved instances
- Cuts wasted commitment without manual work
- Shows cloud and software spend across vendors
- Identifies unused licences and resources to cut
Risk radar
Risk radar
Exposure by dimension, 0-100, Global
Risk heat map
Impact against likelihood; numbers match the register below
- Waste and cost overruns
- GPU price and capacity
- Unused commitments
- Lock-in and egress
- Data residency and sovereignty
- GPU and capacity shortages
- Egress and lock-in costs
- Data sovereignty rules
- Hyperscaler outages
Risk register
29% waste and rising with AI workloads.
AWS raised GPU reservation prices twice in 2026; capacity is tight.
Commitments sized on optimistic forecasts become waste.
Proprietary services and data transfer costs make exit expensive.
Rules on where data sits are tightening in Europe and India.
AI demand has pushed GPU prices up and limited capacity in some regions.
Data transfer fees make switching expensive.
Rules on where data is stored keep tightening.
One provider's regional outage can stop many services.
Market pulse
What the biggest suppliers are reporting
From their latest results and investor material.
Q3 2026. $46bn revenue, $578bn backlog; growth guided above 40% for 2027 and 2028.
FY27 Q1. Azure $33.2bn; Microsoft Cloud $63.9bn (+30%); commercial backlog $677bn.
Q3 2026. $30.7bn revenue; 36% margin; $561bn backlog.
Q3 2026. Cloud infrastructure services grew 55%; AWS 26% share, Azure 19%, Google 16%.
Q3 2026. First €10bn+ quarter; current cloud backlog €24.1bn (+26% cc); licence revenue -9%, support -5% as customers move to cloud.
Latest developments
Dated, newest first. Checked 16 Nov 2026.
Revenue $46bn; operating margin 38.7%; capex about $220bn in 2026 and more in 2027.
So what: AWS is selling all it can build: expect firm pricing.
Constellation Research →Operating margin 36.2%; backlog about $561bn; TPU system sales contributing.
So what: Google is the aggressive challenger: use it as leverage.
Company Q3 2026 results and calls (Microsoft, Amazon, Alphabet) →Microsoft discloses Azure revenue in dollars for the first time.
So what: Azure growth is outpacing AWS: Microsoft will push bundled commits.
SiliconANGLE →AI-specific services now a third of the market; neoclouds gaining share.
So what: Neoclouds are a real alternative for AI capacity.
CIO Dive →Second rise of 2026, across P6, P5 and P4 families.
So what: Don't assume GPU costs will fall; lock prices where you can.
The Procurement Lens →What this means for your next negotiation
- Cloud vendors are growing 39-103% and capacity is tight: commit early for GPU capacity.
- Waste is ~29% of cloud spend: FinOps first, then negotiate.
- Negotiate egress waivers and exit terms now.
Signals to watch next quarter
Capacity and pricing.
Supply vs demand.
Region requirements.
Sources
- Company Q3 2026 results: AWS, Microsoft, Google (via Techzine) · Oct 2026
- Flexera 2026 State of the Cloud Report · Mar 2026
- Synergy Research Q3 2026 (via CIO Dive) · Oct 2026
- AWS GPU pricing changes (via eCorpIT) · Jul 2026
Figures checked 16 Nov 2026. Tags: Official = published by the source; Derived = calculated from published figures; Judgement = The Procurement Lens view.