AI Usage Pressuring Databricks Margins

Databricks is trading some software like economics for land grab economics in AI. The old model was mostly high margin analytics and data processing. The new model adds GPU heavy training, inference, and agent workloads that drive far more cloud spend per dollar of revenue. That is why AI can lift growth and retention at the same time that gross margins come under pressure from the roughly 80 percent level seen in June 2024.1 2 3 4

• The mix shift is large enough to matter. AI products reached a $1B plus revenue run rate within 18 months by August 2025, then $1.4B in Q4 2025, while Data Warehousing also passed $1B by December 2025 and $1.5B ARR by June 2026. Fast growth is increasingly coming from newer workloads, not just classic Spark jobs.2 5 1
• These newer workloads are physically more expensive to serve. Databricks now sells model training, inference, vector search, and AI assistants, all of which lean harder on GPUs, memory, and always on serving than batch analytics. The company has also built tooling to unify Databricks and cloud infrastructure costs, which signals that AI era spend is a first order operating issue.2 6 7 8
• There is precedent inside Databricks for margin recovery once products mature. Databricks SQL grew from $100M ARR in 2023 to $400M in 2024, and by late 2025 the company reported 85 percent gross margins with positive free cash flow. The pattern is near term margin compression during product buildout, followed by better utilization and pricing as products harden.3 2 5

The next phase is about turning AI from a high growth compute reseller into a higher margin control point. If Databricks can keep owning governance, orchestration, and the full workflow around AI workloads, margin expansion should follow as infrastructure gets optimized and more revenue comes from the software layer wrapped around that compute.2 7 8