Nine months after Microsoft first showed off HorizonDB, its cloud-native, PostgreSQL-compatible database built around disaggregated compute and storage, the service is still in public preview with no committed general-availability date beyond a vague target of "second half of 2026." Cloud analysts interviewed for this story argue that window matters a lot right now, because Microsoft is no longer the only hyperscaler chasing AI-workload database business: AWS Aurora PostgreSQL has had vector search and Bedrock integration for a while, Google's AlloyDB AI already offers embeddings and natural-language query support, and both Databricks (via Lakebase) and Snowflake (via its Crunchy Data acquisition) have shipped competing Postgres-compatible services this year. The practical takeaway for engineering and platform teams evaluating databases is that a long, open-ended preview creates real risk: no committed SLAs, unclear pricing, and no certainty about which PostgreSQL extensions will ultimately be supported, since HorizonDB currently only supports an approved subset rather than arbitrary extensions. Analysts also flagged that HorizonDB uses provisioned compute rather than scale-to-zero serverless pricing, meaning idle or intermittent workloads keep accruing charges, and that Microsoft has published throughput claims against open-source Postgres but not head-to-head benchmarks against Aurora or AlloyDB, leaving real-world performance comparisons unresolved. The consensus view is that most enterprises with near-term database needs should not wait on HorizonDB and should default to an already-GA option, reserving HorizonDB for experimental workloads unless they are deeply committed to Azure and willing to bet on Microsoft catching up on maturity and ecosystem support.