Joey on SQL Server

Azure Under the Hood: Key Infrastructure Updates from Ignite 2025

Amid the AI-heavy headlines at Microsoft's annual conference, several behind-the-scenes updates landed that could have a major impact on how IT teams build, run and scale core infrastructure in Azure.

• By Joey D'Antoni
• 11/21/2025

In my effort to find some of the interesting things that happened at Ignite this week that didn't quite make the cut from Microsoft marketing for the keynote, I talked to several product managers, read a whole lot of press releases, and tried to distill some interesting information that didn't make the latest headlines about AI. With that in mind, here are the most interesting things I found in my research.

Managed Instance on Azure App Service. As IT professionals, we all know naming things is hard. This service is not directly related to Azure SQL Managed Instance; however, the two services do share some features, much like Managed Instance, which differs from Azure SQL Database by providing an experience closer to on-premises SQL Server. Managed Instance on App Service enables Internet Information Services (IIS) apps that previously had requirements Azure App Service didn't support. These apps would represent older .NET apps that depend on libraries installed on the OS, SMTP servers, or legacy MSMQ middleware. This service is only available for Windows apps and is in public preview.

Smart tier for Azure Blob and Azure Data Lake Storage is a new service built on top of the Azure storage stack. Smart tier represents what many administrators have scripted themselves using storage policies, with a couple of key differences. Smart tier will keep regularly accessed storage objects on the hot tier, then move inactive objects to the cool tier after 30 days, and to the cold tier after 60 days. The archive tier is not included in this offering. The key difference between the smart tier and doing this manually is that, in the smart tier, there are no data egress costs when moving data from tier to tier. There is a small monitoring fee for orchestrating the storage account.

Database professionals know that one of the biggest challenges of moving to the cloud is storage performance. The problem typically isn't that the disks can't perform well enough; it's that there isn't enough bandwidth between the virtual machines and the storage. Azure Boost has been an ongoing effort to address this problem, and Microsoft has announced the latest version, which offloads virtualization workloads from the physical host to purpose-built hardware and software. The first wave of this will become available on the following VM series: Dlsv7, Dsv7, and Esv7, offering up to:

• Up to 400 Gbps networking bandwidth with the largest size.
• Up to 800k IOPS and 20 GBps throughput to Premium v2 and Ultra Disk remote storage with the largest size.

That should help solve the storage problem and also go up to 1,000,000 I/O operations per second (IOPs). While this would be very expensive to provision, if you have workloads running a hedge fund or a stock exchange, it could be worth it, which leads us to our next bit of news.

Azure Ultra Disk has been generally available since 2019 and has primarily maintained the same pricing model. At the same time, Ultra Disk offers significant performance gains over Premium storage V1, the introduction of Premium V2, which came with similar flexibility to Ultra Disk, and very similar performance levels. Ultra Disk and Premium V2 use the same underlying storage platform. Still, Ultra Disk offers better performance at the right tail of the I/O curve. What do I mean by that? In cloud storage, as you approach IOPs and bandwidth limits, you get throttled, which means a governor starts backing off your IO. This throttling manifests itself in increased storage latency. You can see the benchmark below, which I did the other day for a presentation. Note: Ultra Disk is not in this group of disks:

This chart is not a reference (I actually screwed up the config of my Premium V2 disk). The chart shows the relative increase in latency as IO approaches its maximum. Ultra Disk would provide you with better performance as you approach the right-hand side of that chart. The problem is that Ultra Disk has been approximately three to four more expensive than Premium storage, making it cost-prohibitive for all but the most critical workloads. While Ultra Disk remains costly, the performance dimensions have become more granular, with storage billed in 1 GiB increments, increasing available IOPs per GiB from 300 to 1000, and introducing minimum IOPs and MB/second. The example Microsoft used showed roughly a 22 percent decrease in costs for a 12.5 TB disk with 5000 IOPs and 200 MB/s, going from $2324/month to $1815/month.
Related to storage, some big news is that Azure Site Recovery (ASR), Microsoft's disaster recovery-as-a-service offering, has a couple of significant enhancements. There is now support for both Ultra Disk and Premium Disk v2, which is one of the considerable limitations around using those newer disks with VMs. That change, along with capacity guidance for ASR, is generally available. Another ASR change that will affect database workloads is that ASR now supports up to 5x churn -- meaning it should support up to about 500 MB/s of disk activity in a protected VM, which opens ASR to a much wider range of database and busy I/O workloads. Another important GA announcement was support for Azure Backup for Azure Data Lake Storage Gen 2 accounts, along with a public preview of Azure Backup for Elastic SAN. There are also a slew of resiliency features around recovery from cyberattacks, including soft delete for backup vaults and SAP HANA and SQL Server VMs.

One other thing I wanted to include, because I had some nice help from a friendly product manager (thanks, Scott!), was about a demo in Mark Russinovich's Azure Innovations session. The Azure Storage teams have been working towards “scaled account,” which offers customers more throughput and IOPs (you might notice this is always a theme around storage). At Ignite, the Azure storage team demoed reading and writing from client nodes to Azure Blob Storage at 51 terabits per second and 22 terabits per second, respectively. The required bandwidth of a 4k video stream without buffering is about 50Mbps. This means 51 Tbps should be roughly 1M simultaneous streams. This demo is part of some research work. If you need this kind of throughput for blob storage, your firm is probably already in contact with Microsoft. If not, feel free to reach out.

AI continues to be the overwhelming theme of the tech industry, and that shows in much of the announcements at Ignite. However, even the latest AI workloads, have dependencies on building core infrastructure that is robust and can handle the massive workloads that AI demands. I put together this column to try to highlight those infrastructure enhancements, and to showcase teams that are doing good work, that doesn't make the headlines.