Azure ND H100 v5 vs Google Cloud A3

Azure ND H100 v5 and Google Cloud A3 cannot be scored on the same axes — one of them does not publish a rate for everything gpu cloud bills on.

List prices checked

Rate cards side by side

Only the axes at least one of them bills on. A dash means that provider does not charge for it.

Billing axisAzure ND H100 v5Google Cloud A3Difference
Free tierNoNo

Source: Microsoft pricing (May 2026) · Google pricing (May 2026). Published list prices only — committed-use discounts, regional variation, and per-request charges are not included. Not yet independently re-checked against the vendor’s page.

Monthly cost for three real workloads

The rate cards applied to identical usage. This is where the winner changes.

WorkloadAzure ND H100 v5Google Cloud A3Cheaper
Side projectA personal site or small app with modest traffic, a few gigabytes of assets, and one small always-on process.$5,095$5,095Not comparable
Growing SaaSA funded product serving a few thousand daily users, with an image-heavy frontend and two application servers.$10,191$10,191Not comparable
Media-heavy siteA publisher or video product where delivery dominates the bill and compute is almost an afterthought.$5,095$5,095Not comparable

Workload definitions are identical on every comparison page, so figures here can be read against any other pair in the same category.

What each one is for

Azure ND H100 v5

by Microsoft

Microsoft's H100 instances, sold mostly through enterprise agreements where the list rate is not what anyone actually pays.

Google Cloud A3

by Google

H100 instances on Compute Engine with fast interconnect, priced in line with the other hyperscalers and well above the specialists.

Common questions

Is Azure ND H100 v5 or Google Cloud A3 cheaper in 2026?
Neither can be scored across all three workloads, because at least one of them does not publish a rate for every axis this category bills on.
Which has the better free tier, Azure ND H100 v5 or Google Cloud A3?
Neither publishes a free tier; both bill from the first unit of usage.
What do these estimates not include?
Committed-use and reserved discounts, regional price variation, per-request API charges, support plans, and minimum retention periods. All of those move real invoices, sometimes by a lot. These figures compare published list prices for the same workload, which is the comparison the rate cards alone cannot give you.

Other comparisons in this category