Should I build a new RDS farm on Server 2022 or 2025?

If you want the lower-risk choice, I’d lean toward 2022 unless you have a specific requirement that points to 2025. The feedback is mixed: some deployments report no trouble with 2025, while others describe more friction. That’s not enough to call one version universally faster or more reliable.

Check application support, licensing, and your organization’s patching and support window before committing. If 2025 fits those requirements, test it with your actual user workload rather than relying on broad performance claims.

I wouldn’t treat a direct logon to a Session Host as proof that Connection Broker is broken. Broker-based access and the host’s own sign-in permissions are separate things. Review which users or groups are allowed to log on locally or through Remote Desktop Services, and make sure ordinary users have only the access you intend.

Then test both paths: connect through the published collection and try the host directly. Check the collection, deployment, DNS, and any firewall or gateway rules too. If users should never reach a host directly, restrict that route rather than assuming the Broker will prevent it.

Performance reports are hard to compare without the same apps, profile setup, storage, and user load. A newer server version won’t automatically make an RDS workload feel faster. I’d build a small pilot, measure login time and the apps people actually use, and include a busy period in the test.

Also verify compatibility for your key applications and management tools. If the pilot is clean and 2025 gives you a concrete benefit, use it. If not, choosing the more familiar 2022 build is a reasonable way to keep the project simpler.

Check Microsoft 365 Apps support as part of the decision. Server-version support can affect how long that combination remains viable, so confirm the current requirements for the exact apps and deployment method you plan to use. That may matter more than vague claims that one OS performs better.