Cloud marketplace economics: the offer type changes the answer
AWS, Microsoft and Google fee schedules revisited, monthly private-offer payments clarified, and a transparent break-even model for packaged software.
This research continues The 100 open-source projects that matter in 2026 (September baseline) and What we found looking at 300+ cloud marketplace listings (August fieldwork). Those articles establish the cohort and the commercial questions; the evidence below states what we checked again.
The comparison starts with the offer type
Our August article compressed AWS pricing into a 20% headline. That describes public server products, including AMIs, containers and machine-learning products. It does not describe public SaaS, whose standard listing fee is 3%. This distinction changes the commercial comparison before support costs enter the model.
For this follow-up we read the current provider fee schedules, AWS private-offer documentation and installment-plan documentation. We did not repeat the listing census, inspect new company accounts or measure seller conversion. The contribution here is a corrected comparison and an explicit operating-cost model.
Current documented base rates
| Transaction | AWS | Microsoft Marketplace | Google Cloud Marketplace |
|---|---|---|---|
| Public / standard software offer | 3% SaaS; 20% server products | 3% standard transact fee | 3% standard offer |
| New private deal below $1M | 3% | 3% standard fee | 3% |
| New private deal $1M to below $10M | 2% | No equivalent TCV tier stated in the cited standard fee page | 2% |
| New private deal $10M or more | 1.5% | No equivalent TCV tier stated in the cited standard fee page | 1.5% |
| Renewal | 1.5% private-offer renewals | 50% discount on existing transaction fee for qualifying private-offer customer renewals | 1.5% for qualifying native renewals; also qualifying channel shifts and migrations |
Sources: AWS listing fees, Microsoft transact capabilities, and Google's vendor net revenue schedule. Google's schedule is effective 21 April 2025 and has eligibility definitions; the retained vendor share is converted here into the corresponding platform share. Microsoft's renewal discount has qualification requirements. These rows are not interchangeable eligibility promises.
AWS also documents a 0.5 percentage-point channel partner private-offer uplift and a 1 percentage-point regional uplift for South Korea. Such additions can change the comparison. The arithmetic below deliberately excludes them, taxes, refunds, currency conversion and negotiated exceptions.
Monthly payments are not the AMI Monthly product model
AWS excludes AMI Monthly from private offers. Separately, eligible contract private offers support installment plans with monthly, quarterly, annual or custom frequencies. Our earlier wording incorrectly extended the AMI restriction to monthly billing in general. A payment every month under a committed contract is also not the same commercial promise as a cancellable monthly subscription.
Sources: private-offer considerations and installment plans. Sellers still need to check the exact product and pricing type before choosing a schedule.
A $99 example, with the missing costs restored
At an assumed $99 monthly software charge, a 20% fee leaves $79.20 and a 3% fee leaves $96.03. The difference is $16.83 per customer-month, or $201.96 over twelve equal payments. These are receipts after the platform fee, not profit. Compute may be billed separately to the buyer for a VM offer, while a SaaS seller may carry hosting costs inside its own price.
For a deliberately simple sensitivity model, assume $30 per customer-month for variable support and operations, plus $600 per month of fixed maintenance. These are illustrative inputs, not measured seller costs. Break-even customers equal fixed cost divided by contribution per customer, rounded up.
| Assumed fee | After platform fee | After $30 variable cost | Customers covering $600 fixed cost |
|---|---|---|---|
| 20% | $79.20 | $49.20 | 13 |
| 3% | $96.03 | $66.03 | 10 |
At $60 variable cost, the same calculation requires 32 customers at 20% and 17 at 3%. That spread shows why an estimate of support load belongs next to the fee schedule. It does not prove that one delivery model is better: hosting a service, shipping an image and supporting a deployment can have different costs and different prices.
monthly contribution = price × (1 − fee rate) − variable cost
break-even customers = ceil(fixed monthly cost / monthly contribution)
If contribution is zero or negative, no customer count covers fixed cost in this model. It also assumes stable costs per customer, full collection and no acquisition cost. Real staffing often grows in steps rather than smoothly.
What has changed since the original fieldwork
AWS now documents a zero base listing fee for qualifying professional-services agreements in multi-product solution offer sets accepted on or after 1 September 2026. Qualification is checked at invoice generation, and regional or channel uplifts can still apply. This is a targeted rule for services agreements, not a removal of the software listing fee.
AWS also documents express private offers: rate-card-based offer generation for SaaS Contract and SaaS Contract with Consumption Pricing. We verified the current capability, not its launch date or its conversion impact. It may reduce offer preparation work for eligible SaaS products; it should not be assumed to apply to an AMI catalog.
What to measure before expanding a catalog
For each offer, record software receipts, the actual fee and adjustments, support time, update work and acquisition source. Separate a new customer discovered through the marketplace from an existing deal routed through marketplace procurement. A growing SKU count cannot substitute for either measure.
The November OSS refresh helps identify where developer attention is moving. The August marketplace study supplies historical observations about listings and reviews. Neither establishes the contribution margin of a new offer. That remains a question for a small operating trial with recorded costs and paid customers.
Our next empirical question is therefore narrower than “does marketplace work?”: for one supported product and one offer type, how much collected contribution remains after updates, support and acquisition? The model above makes the missing evidence visible without filling it with an industry-wide revenue claim.