Databricks ships Real‑Time Mode
Databricks put Apache Spark Real‑Time Mode into production and announced a partnership with Accenture to accelerate enterprise agent and real‑time AI applications, while also expanding hiring and its San Francisco footprint. Real‑time mode and agent use cases explicitly push low‑latency inference and autoscaling requirements for GPU‑optimized serving. ( )
Apache Spark 4.1 lists "Real-time Mode" in its release notes as an SPIP feature (Real-time Mode in Structured Streaming), marking the feature's inclusion in the upstream Spark 4.1 release. (spark.apache.org) (spark.apache.org) Databricks’ March 16, 2026 architecture post names RTM authors Jerry Peng, Siying Dong and Indrajit Roy and states RTM delivers millisecond-level latency and "sub-100ms responsiveness" for operational workloads. (databricks.com) (databricks.com) Azure Databricks documentation reports RTM can achieve end-to-end latency "as low as five milliseconds" and requires Databricks Runtime 17.1 or above for real-time features. (learn.microsoft.com) (learn.microsoft.com) Databricks describes RTM’s hybrid execution as scheduling stages concurrently with non-blocking operators to avoid microbatch boundaries, and Databricks’ evaluation claims RTM can outperform Apache Flink on common feature-engineering workloads. (databricks.com) (databricks.com) Azure and Databricks docs require RTM clusters be created as fixed "classic compute" instances (autoscaling and spot instances must be disabled) and instruct users to clear Photon acceleration for RTM clusters. (learn.microsoft.com) (learn.microsoft.com) The Accenture–Databricks announcement on March 17, 2026 creates an "Accenture Databricks Business Group" supported by more than 25,000 Databricks‑trained professionals and names Lakebase, Genie and Agent Bricks as joint offerings with pilot customers including Albertsons, BASF and Kyowa Kirin International. (newsroom.accenture.com) (newsroom.accenture.com) Databricks’ Bay Area footprint includes a roughly 150,000‑square‑foot lease at One Sansome in downtown San Francisco announced in 2025, and the company disclosed a $250 million investment to expand R&D and hire more than 50% additional staff in India (aiming for over 750 employees there). (connectcre.com) (connectcre.com) (economictimes.indiatimes.com) (economictimes.indiatimes.com) Databricks reported crossing a $5.4 billion revenue run‑rate with >65% year‑over‑year growth in a February 9, 2026 company release, providing the financial scale behind investments in Lakebase, Genie and RTM adoption. (databricks.com) (databricks.com)
