By Microsoft just served notice to Snowflake and Databricks. GPU-accelerated query execution is coming to Fabric Data Warehouse — up to 7x faster SQL performance, zero code changes, one workspace toggle. Early access preview opens in July 2026. If you are running Fabric workloads, this is the most significant performance upgrade since the platform launched.
What Microsoft Announced
At Build 2026, Microsoft revealed that Fabric Data Warehouse now runs eligible queries directly on NVIDIA accelerated computing. The engine handling this is called CoddSpeed — a GPU-accelerated query execution system that began as a Microsoft Research prototype called TQP (Tensor Query Processor) and spent years evolving into production infrastructure.
The SIGMOD 2026 database research conference just awarded CoddSpeed its Best Industry Paper award. That is the flagship peer-reviewed venue for data management research — not a marketing badge, an academic validation. The paper describes an architecture designed to route queries across multiple hardware accelerators: GPUs today, FPGAs and ASICs on the roadmap.
The practical upshot: you run the same SQL you have always run. Fabric detects eligible queries and routes them to GPU execution automatically. No schema changes. No query rewrites. No new tooling to learn.
The Performance Numbers
Microsoft internal benchmark, run at 64-user concurrency in May 2026, shows up to 7x faster performance compared to three major cloud data warehouse competitors across reporting and application workloads. UNC Health, one of the early production adopters, reports 5x query speed improvement on existing workloads.
There is a third number worth noting: teams using Fabric data agents — AI agents that query the data warehouse in real-time — saw 50% reduction in end-to-end response times. That is not a coincidence. AI agents are the reason GPU acceleration in data warehousing is suddenly urgent.
Why AI Agents Changed the Math
Traditional analytics worked on human time: a dashboard refreshes in 10 seconds and nobody complains. AI agents operate on machine time: a query that takes 10 seconds blocks an entire reasoning loop. GPU acceleration brings analytical queries into the sub-second range that agentic workflows require.
Microsoft announced Fabric IQ alongside the GPU data warehouse — a semantic context layer that grounds AI agents in organizational data: business rules, entity relationships, real-time signals. GPU-accelerated Fabric Data Warehouse is the performance foundation that makes Fabric IQ viable at scale. The two announcements are a package deal.
What This Means for Snowflake and Databricks Users
Fabric has historically trailed Snowflake on large-scale, high-concurrency analytical performance. GPU acceleration does not just close that gap — it establishes Fabric as the first fully managed data warehouse with native GPU support.
Snowflake architecture does not include GPU-native query acceleration. Databricks has Photon (a CPU-vectorized engine claiming 2-8x speedup), which is excellent for mixed AI/SQL workloads but not GPU-accelerated. If 7x proves out in production, Microsoft has built a real moat.
If you are currently evaluating Fabric versus Snowflake or Databricks, GPU acceleration should be in the conversation now. Run the preview, benchmark your own workloads — Microsoft internal numbers will not match every use case, but the direction is clear.
How to Get Access
The GPU-accelerated Data Warehouse enters early access preview in July 2026. To enable it once available:
- Go to Workspace Settings in Microsoft Fabric
- Enable the GPU acceleration toggle
- Run queries as normal — eligible queries are automatically routed to GPU execution
The toggle applies to all SQL Analytics Endpoints and Data Warehouses in the workspace. Sign up for early access through the Microsoft Fabric Community announcement.
GPU acceleration in data warehousing is not going to stay a differentiator — it will become table stakes within two years, just as SSD storage and columnar formats did before it. The question is who gets the performance advantage during the window when it matters. For Fabric users, that window opens in July.
