Negotiating cloud commitments when capacity is scarce
In brief
AWS and Google say demand outruns supply, and all three report backlogs over $550B. What that does to your bargaining power and which terms still move.
Steve Elliott · · 7 min read
On this page
We think that when hyperscalers are short of capacity, a buyer loses bargaining power on the headline discount but not on contract terms. The enterprise commitment worth signing in late 2026, in our view, gives up less to chase a bigger rate and asks more about ramps, capacity, AI services, marketplace spend and what happens if you fall short.
The numbers behind that view come from the companies' Q3 2026 results, filings and earnings calls, published in the last week of October. The terms come from their documentation for AWS private pricing (often called an EDP), the Microsoft Azure Consumption Commitment (MACC) and Google Cloud commitments. Nothing here is legal advice: read your own agreement with counsel and your procurement team.
What do the Q3 2026 results say about supply and demand?
All three providers reported cloud growth far above their overall business, very large contracted backlogs and sharply higher capital spending.
| As reported for the quarter ended September 30, 2026 | AWS (Amazon) | Microsoft | Google Cloud (Alphabet) |
|---|---|---|---|
| Cloud revenue | AWS $46.0B, up 39% | Azure $33.2B, up 48% | $30.7B, up 103% |
| Contracted backlog | AWS backlog $578B | Commercial RPO $677B, up 73% | Google Cloud backlog $561B |
| Quarterly capital spend (company-wide) | $77.3B property and equipment purchases, from $35.1B | $39.4B additions to property and equipment, from $19.4B | $51.4B property and equipment purchases, from $24.0B |
| Full-year signal | About $220B cash capex expected for 2026 | Not given in the release | $195B–$205B capex guidance for 2026, with a significant increase expected in 2027 |
Three caveats matter before reading bargaining power into this table. Microsoft's commercial remaining performance obligation is the commercial portion of all its contracted future revenue, not an Azure-only figure. Capital spending is company-wide, not cloud-only. And the three backlog figures are defined by each company, so they are not like for like.
Amazon and Alphabet executives were plain about supply. Andy Jassy said AWS backlog was growing triple digits year over year and that "almost as fast as we install capacity, customers are gobbling it up." Alphabet's CFO said Google Cloud began using third-party compute capacity at the end of the quarter because of supply constraints. Alphabet's 10-Q puts its total backlog at $567.7 billion, $560.8 billion of it Google Cloud, and expects just over 55% to become revenue within 24 months. Google Cloud CEO Thomas Kurian said "we could obviously serve a lot more demand with capacity," adding that Google is balancing growth carefully.
Does scarcity weaken your negotiating position?
On rate, we think yes. By our arithmetic, AWS backlog is about 3.1 times its $184 billion annualized revenue run rate, and Google Cloud's is about 4.6 times its Q3 revenue annualized. When a provider sells capacity almost as fast as it installs it, a marginal enterprise commitment is a smaller share of its future revenue than it was when growth was slower. We would not expect headline discount percentages to improve much in this cycle.
On terms, we think no. All three roughly doubled quarterly capital spending year on year, and in our view a well-forecast multi-year commitment still has planning value to a provider building that fast. The negotiable surface moves from the size of the discount to what exactly you are buying, when, and with what flexibility.
Why a discount commitment is not a capacity commitment
In our reading, enterprise spend commitments are billing constructs, and the discount instruments underneath them don't reserve hardware by themselves, with one exception at Google.
- AWS. The savings plans FAQ says plainly: "No, Savings Plans does not provide a capacity reservation." Capacity comes from On-Demand Capacity Reservations, which you can cancel at any time, or future-dated reservations, where you choose a commitment duration, can't reduce it, and may pay a charge to cancel during it (cancellation arrived in June 2026). GPU Capacity Blocks for ML can't be cancelled.
- Azure. Microsoft's capacity reservation documentation draws the same line: "Reserved Instances provide a billing discount only and don't guarantee capacity." On-demand capacity reservations are billed at the VM rate whether used or not, and both used and unused reservations can take reservation and savings plan discounts.
- Google Cloud. Google says commitments don't reserve zonal capacity on their own. Resource-based commitments can cover GPUs, but Google requires you to attach reservations for them, so a GPU commitment there does come with capacity. Spend-based commitments cover Cloud GPUs only for the G2 and G4 series, as of November 2026.
The practical risk: you sign a commitment sized for an AI roadmap, the GPUs you need aren't available in your region, and you end the year short on spend for reasons you didn't control. In our view, any commitment that assumes accelerator growth should say what happens to the commitment if capacity is not available.
What should you ask for?
These are our suggestions, ranked by how much exposure they remove in a supply-constrained market.
- A ramp, not a flat line. Azure's documentation shows that a MACC can carry milestones, and a missed milestone triggers a shortfall charge, just like a missed total. Shape milestones to your migration and AI plans, not to equal annual slices. In our view the same logic applies to AWS private pricing, which AWS describes as a "structured annual commitment", and to Google contracts with minimum commitments.
- Capacity linkage. Ask for named capacity (regions, instance families, quantities) or relief from the commitment if the provider can't supply it. Google's GPU commitments already tie discount to reservation, so the idea isn't foreign to providers.
- AI services inside the commitment. Confirm in writing which AI services decrement the commitment. Some are explicit: Microsoft says spend on Claude models in Foundry under its CCU meter decrements MACC like other Azure Marketplace consumption. Google's Flexible Savings Plans cover generative AI models, but that is a separate instrument, so confirm whether AI spend draws down your enterprise minimum. AWS's private pricing overview does not mention AI or GPU services.
- Marketplace burn-down, with the fine print. AWS says qualifying AWS Marketplace purchases count toward a private pricing commitment, without stating a rate. Microsoft counts 100% of the pretax amount of "Azure benefit eligible" offers bought through the Azure portal, but not credit card purchases, purchases using Azure prepayment, or licenses deployed on-premises or in hybrid. Google counts 100% of eligible Marketplace spend toward minimum commitments, excluding Maps, Ads, professional services and services not running on Google Cloud, among others.
- Shortfall and true-up terms you can live with. On Azure, an unmet MACC is charged as an Azure prepayment credit for the remaining balance, and consumption covered by that credit doesn't count toward a MACC. Know where the shortfall money goes before you sign.
- Flexibility in the contract, as reservation exchanges narrow. From February 1, 2027, reservations for Azure services covered by savings plans can no longer be exchanged, with one final exchange for reservations bought before that date. Cancellation and instance size flexibility don't change, and Microsoft says the change gives customers a clear, predictable framework for choosing between commitment types. As self-service exchanges narrow, we think it's reasonable to negotiate some of that flexibility back at the agreement level.
An illustrative example: flat versus ramped
The numbers below are made up to show the mechanics.
A company signs a $30 million, three-year Azure commitment with three equal $10 million annual milestones. Its forecast is $7 million in year one, $10 million in year two and $14 million in year three, because most of its AI workload lands late. Total forecast spend is $31 million, so the three-year total is safe. The first milestone is not: year one ends $3 million short, and that $3 million is billed as prepayment credit, a year before the company planned to spend it. Per Microsoft's documentation, consumption paid for with that credit doesn't count toward the MACC, so it doesn't help with later milestones either. Check how your agreement handles it.
| Year 1 | Year 2 | Year 3 | Total | |
|---|---|---|---|---|
| Forecast consumption | $7M | $10M | $14M | $31M |
| Flat milestones | $10M | $10M | $10M | $30M |
| Shortfall against flat | $3M | $0 | $0 | $3M |
| Ramped milestones | $7M | $10M | $13M | $30M |
| Shortfall against ramp | $0 | $0 | $0 | $0 |
Now suppose the GPU capacity for the year-three workload arrives two quarters late and year three comes in at $11 million. Against the ramp, the company ends $2 million short in total. A capacity-relief clause would cover that gap. Without one, bringing forward eligible marketplace purchases might close part of it. That is why we rank ramp and capacity linkage above a slightly better rate.
How do you track a commitment once it's signed?
Burn-down needs its own reporting, separate from cost allocation. Azure sends billing account admins email reminders 90, 60 and 30 days before a MACC expires or a milestone ends. That is a late warning, not a plan.
The FOCUS specification has carried contract commitments since version 1.3: cost rows identify the commitment they apply to, and a separate Contract Commitment dataset holds each commitment's period, cost and type. Where providers supply it, that lets FinOps report burn-down by month, by service and against milestones.
The FinOps Foundation's rate optimization capability says procurement typically runs negotiated discount programs, with FinOps supplying historical and planned usage. It also notes negotiated discounts are often covered by non-disclosure rules, which is why we think published discount benchmarks deserve caution. Pair commitment burn-down with effective savings rate so finance sees both whether you're meeting the commitment and what it's saving.
A commitment is a multi-year funding decision tied to a product and AI roadmap, so it belongs in the same operating model as the roadmap it assumes.
Sources
- Amazon, Amazon.com Announces Third Quarter Results, October 29, 2026
- Amazon, AWS growth and AI demand: Q3 2026 earnings remarks, October 2026
- Microsoft, FY27 Q1 earnings press release, October 28, 2026
- Microsoft, FY27 Q1 investor metrics
- Amazon, Amazon.com Announces Third Quarter Results (2025), Q3 2025
- Alphabet, Alphabet Announces Third Quarter 2026 Results (Form 8-K, Exhibit 99.1), October 28, 2026
- Alphabet, Form 10-Q for the quarter ended September 30, 2026
- Alphabet, Q3 2026 earnings call transcript, October 28, 2026
- AWS, AWS Private Pricing overview, 2026
- AWS, Savings Plans FAQs
- AWS, On-Demand Capacity Reservations
- AWS, Amazon EC2 Future-dated Capacity Reservations Now Supports Cancellation, June 10, 2026
- AWS, Capacity Blocks for ML
- Microsoft Learn, Track your Microsoft Azure Consumption Commitment (MACC), updated March 20, 2026
- Microsoft Learn, Azure Consumption Commitment benefit, updated September 25, 2025
- Microsoft Learn, On-demand capacity reservation, updated June 16, 2026
- Microsoft Learn, Claude models billing in Microsoft Foundry, updated October 27, 2026
- Microsoft Tech Community, Reservation exchanges for Azure services covered by savings plans end starting Feb. 1, 2027, July 30, 2026
- Google Cloud, Google Cloud Marketplace commitment drawdown rates, updated November 10, 2026
- Google Cloud, Committed use discounts, updated November 10, 2026
- Google Cloud, Combine reservations with committed use discounts
- Google Cloud, Purchase commitments with attached reservations
- FOCUS, Contract commitments feature (v1.4)
- FinOps Foundation, Rate Optimization capability
Frequently asked questions
Does a cloud spend commitment guarantee capacity?
Generally no. AWS says Savings Plans do not provide a capacity reservation, and Microsoft says reserved instances provide a billing discount only. Google is the exception for GPUs, where it requires reservations attached to the commitment. Otherwise capacity comes from separate products, such as AWS Capacity Reservations and Capacity Blocks for ML or Azure on-demand capacity reservations, or from terms you negotiate.
Do AWS Marketplace purchases count toward a private pricing commitment?
AWS says qualifying AWS Marketplace purchases count toward a private pricing commitment, but its public overview does not state a crediting rate. Microsoft and Google publish more detail: both count 100% of eligible marketplace spend, with exclusions. Confirm in your agreement which listings qualify and at what rate before you plan to use marketplace spend to close a gap.
What happens if you miss an Azure MACC?
Per Microsoft's documentation, if you don't meet the total by the end date, a shortfall charge for the remaining balance is applied as an Azure prepayment credit. Missed milestones are charged the same way. Consumption covered by that credit doesn't count toward a MACC. Your own agreement may differ, so confirm the terms with your Microsoft account team and advisers.
Is now a bad time to sign a cloud commitment?
Not necessarily. In our view scarcity weakens a buyer's bargaining power on the headline discount but not on structure. A commitment with a ramp that matches your roadmap, explicit AI service inclusion, clear marketplace rules and some capacity linkage can be worth more than a higher rate on a flat commitment you might not be able to consume.
Steve Elliott
Published November 22, 2026
More on FinOps & Cloud
FOCUS, commitments, shared cost.
Keep reading
Forecasting cloud spend when usage won't sit still
Split committed, usage-driven and AI cloud spend, tie usage to business drivers, forecast in ranges and measure variance by line, with a worked example.
Cloud repatriation: when the numbers support moving back
A TCO frame for cloud repatriation: what to count, what the exit-fee waivers cover, what 37signals' numbers show, and when moving back doesn't pay.
Should product teams own their cloud budgets?
Product teams should hold cloud budgets once allocation puts spend in their name and shared costs have an owner. Prerequisites, guardrails and failure modes.
The Ultimate Guide to Capacity Planning for Cloud Providers
Cloud capacity planning forecasts how much compute, storage and network you will need, then provisions and autoscales it so you meet demand without paying for idle resources.