One unit of API Management is two compute resources. Microsoft's reliability documentation describes automatic zone support distributing those two resources across two zones, with no downtime expected during a zone outage because the resource in the unaffected zone keeps working [14]. That is why $700 a month buys anything here at all: the money pays for the two halves of one unit to sit apart, not for a second unit [4][14].
The two-unit requirement lives in the contract, not the platform. Read the Premium footnote as two units and the like-for-like bill becomes 2 x $2,801, or $5,602, eight times the Standard v2 unit price [5][2]. Per unit the gap is $2,101 a month [1]. What that gap buys is a service credit threshold: on a 730-hour month, 99.95% permits 21.9 minutes of downtime and 99.99% permits 4.38 [3]. The 17.5-minute difference works out at roughly $120 per minute per month [4]. Nobody purchases SLAs by the minute, but the division is the only way to see what the tier premium is denominated in.
Surviving a zone loss and surviving it at full capacity are priced separately. Microsoft gives no guarantee that requests for more capacity in another availability zone succeed during a zone-down scenario, calls backfilling lost units best-effort, and tells teams that need guaranteed capacity to over-provision [15]. Over-provisioning is where the price advantage erodes. Three Standard v2 units cost $2,100 a month, within $701 of a single Premium v2 unit [5].
For anyone already running Standard v2, this comes down to migration, not price. The feature is creation-time only, as Azure integration architect Stephen W. Thomas noted in response to the announcement [8], and Microsoft's documentation imposes the same restriction on Premium v2 [9]. Premium classic is the one tier where zones can be added to an existing location without downtime, applying in 15 to 45 minutes while the gateway continues serving requests [12]. It charges for that in addressing: the public virtual IP changes, the private VIP as well in internal virtual network mode, and DNS records, private DNS zones, firewall rules, allowlists, routing rules and Application Gateway backend settings all have to follow [13].
Moving down from Premium to reach the cheaper zone spread only works if the Premium feature list is dead weight. The pricing table keeps multi-region deployment, the self-hosted gateway, VNET injection, multiple custom domain names and workspaces on the Premium side [17]. Standard v2 does get VNet integration, so the gateway can reach backends isolated in a connected virtual network, but not injection [18].
The trade I would defend: a single-region estate with one custom domain and no self-hosted gateways should take the $700 unit and turn zones on at creation, where the mechanism is real and the migration cost is zero [4][14]. The number that does not come with it is 99.99% [6]. If an internal customer is holding you to four nines, the tier itself is what satisfies that requirement, not the physical topology [5].
