Joey on SQL Server

Why Microsoft Is Still Betting Big on SQL Server in a Cloud-First World

Microsoft's continued focus on SQL Server 2025 highlights the lasting demand for on-premises and hybrid deployments that prioritize performance, control and flexibility for enterprise workloads.

• By Joey D'Antoni
• 05/23/2025

In the cloud era, you may wonder why Microsoft is investing big into an "on-premises" product like SQL Server. It is a complex question I wanted to answer through the lens of some of the new features in SQL Server 2025. Another aspect of the development process is that when features get added to one service, they will ultimately (in nearly all cases) end up in all versions of the product. Some services start in Azure SQL and move into the boxed product, which is most common nowadays. Likewise, we have seen other features initially released in an SQL Server release that later appear in the various Azure database services.

I wanted to examine some of the reasons why Microsoft needs to continue to invest in SQL Server -- both from the perspective of large and small customers and their own shareholders. As I mentioned earlier, at least ten products probably share the source code of SQL Server so that developments can cascade across the product line. Not to channel my inner Steve Ballmer, but it also means software developers writing code for Fabric, SQL Server or Azure SQL Database can all write the same dialect of SQL and T-SQL (mostly -- one of the complaints about Fabric DW and Lakehouses that I frequently hear is that they don't support the full surface area of T-SQL). Microsoft's investments pay off in more robust cloud data services, common developer experiences and better on-premises offerings.

Why would I want to deploy a database server to a cloud VM, an on-premises VM or, heaven forbid, a physical server? There are many reasons—the first being concerns about being locked into a cloud platform, given some of the political headwinds the cloud faces -- portability is a bigger problem than ever. I love Azure SQL Database and think it's an excellent service. Still, for the most part, once your data is there beyond a nominal size (let's call it 250 gigabytes), there is no way to get it out without extended downtime and administrative hassle. Suppose instead you were running SQL Server in an Azure VM. In that case, restoring your backup to your target environment is simply a matter of executing a RESTORE statement, and you can log ship with near-zero downtime.

Another reason why you might not want to use a cloud service is the cost. Now and again I will see someone state how "the cloud is cheaper than on-premises," which is a complicated, oversimplified statement that I hate. There are places where the cloud is cheaper (usually bulk storage and networking "hardware"). There are places where on-premises will be more cost-effective because the cause of leasing large amounts of hardware used over long periods is often higher than the capital expense of buying that hardware, amongst other reasons which we'll discuss below.
On-premises solutions are typically more cost-effective for large database workloads that use large amounts of ultra-fast storage and memory and run at high utilization over long periods. Suppose you have small databases that aren't busy. In that case, they can be an excellent fit for PaaS – this is even truer for larger databases that don't require massive amounts of hardware resources. The challenge with nearly every PaaS database service (Azure SQL DB Hyperscale is the only exception I am aware of here) is that storage volume and compute are connected dimensions in the size of your PaaS compute.

That means as your database grows in data volume, even if the compute demands are not high, you must increase the compute capacity of your PaaS resources to get more storage. When using core-licensed software like SQL Server or Oracle, this means your costs can increase exponentially if data volume increases.

On-Premises Demands
Likewise, you usually can't adjust storage performance without moving to another tier. Cloud VMs provide the flexibility to adjust storage volume and performance without increasing the size of your VM, within some limits, and offer way more flexibility than most PaaS offerings. Ultimately, as much as cloud storage has matured, the fastest storage is still either locally attached or connected via a high-speed, dedicated network with higher bandwidth than is commonly in the cloud.
Across my years of consulting, I've seen three key demands from enterprises: security, performance and availability. Cloud providers have largely mitigated security by improving processes like private networking for nearly all major service offers. However, AI introduces new security concerns, and SQL Server 2025's "bring your model" addresses that. If you want to experiment, my MVP colleague Anthony Nocentino has built a Docker compose file that deploys a SQL Server 2025 container alongside an Ollama container to test SQL Server's new AI features.

Performance is the next concern. I have worked with some customers who pushed the absolute envelope of database performance (for example, 13 million transactions per second on a single server for a gambling site). These customers work closely with Microsoft and are likely the reason for some of the interesting features you have seen added to SQL Server. While cloud VMs can offer excellent performance, these are the kinds of customers.
If you have ever worked on mission-critical apps, you know the importance of high availability. While open-source databases like PostgreSQL and MySQL can be excellent development platforms, even for large-scale production they require a lot of engineering work to keep them highly available. SQL Server's high availability options, log shipping (yes, log shipping is HA and DR!), availability groups and failover cluster instances work. Availability groups have some challenges, especially under high loads with inadequate hardware. However, Microsoft has invested a lot in SQL Server 2025 to improve the robustness of availability groups. Having sturdy availability solutions that work out of the box is one of the biggest reasons customers continue to deploy SQL Server on-premises and IaaS. And because DBAs are control freaks (just kidding, except not really).

Even in an era where we're too focused on AI and analytics, operational databases still record business-critical transactions like sales orders, stock trades, and bank transactions. Database revenue at Microsoft is alleged (by Gartner) to be around $14 billion a year, with SQL Server (and Azure SQL services) making up the vast majority of that revenue. Databases that perform, are reliable and are relatively easy to manage have been Microsoft's calling card since SQL Server 7, and that's why it continues to make significant investments in the product.