Data platforms become the AI layer: zero-copy SAP data, AI inside Iceberg tables, and a CDP fight
SAP data becomes available in Databricks without copying it, AI functions run incrementally inside Iceberg tables, and Databricks' CDP launch raises a build-versus-buy question.
- The companies say SAP data products can be shared with SAP BDC Connect for zero-copy access, and SAF-Holland uses SAP BDC and SAP Databricks to unify more than 100 million records, generating analyses 5x faster than with traditional reporting. Self-reported m:…7080
- Snowflake dynamic Iceberg tables support AI functions directly within the SELECT statement, processing only new or changed rows, with FROZEN regions preserving old AI outputs when prompts change. m:…0415
- Databricks announced a CDP product, but a commentator argues it lacks the breadth of traditional CDPs, particularly in real-time use, and that its user will likely be in IT rather than marketing. m:…3432
SAP and Databricks remove the copy step
SAP and Databricks are announcing the next phase of their partnership at SAP TechEd, the companies say, building on SAP Databricks, which they introduced last year. The core idea is to stop copying enterprise data before AI can use it: SAP data products can be shared with SAP BDC Connect for zero-copy access, and SAP Data Foundation makes data from SAP applications available for consumption in the SAP Data Cloud. The companies say the aim is to lower total cost of ownership and reduce time to value, and note that over 90% of Fortune 500 companies use SAP solutions to manage key business processes. m:…7080
The customer example offered is SAF-Holland, which the companies say uses SAP BDC and SAP Databricks to unify more than 100 million records and generates analyses 5x faster than with traditional reporting approaches. Adidas, Zalando, Mondi and CAF are named as customers using SAP Databricks and SAP BDC Connect for Databricks. On the AI side, the companies describe Genie as an AI coworker providing access to governed data and AI with enterprise context, with a planned bi-directional integration between Genie and Joule in development, plus Unity Gateway for multi-AI governance and cost control across agents, tools, models and MCPs, Lakeflow for ingestion, transformation and orchestration, and Lakebase, a serverless Postgres database built for AI apps and agents. m:…7080
Snowflake puts AI functions inside the table
A separate write-up on Snowflake's Iceberg support describes a similar consolidation from the other direction: instead of moving data to the AI, move the AI into the table. Dynamic Iceberg tables support AI functions directly within the SELECT statement, and those functions only process new or changed rows, keeping enrichment incremental rather than recomputing everything on each refresh. Change tracking must be enabled for incremental reads, the author notes. m:…0415
The detail that makes this workable in practice is the FROZEN region: when prompts change, old AI outputs can be kept instead of triggering a full re-computation of historical data, and filtering on AI flags in a downstream table maintains incremental refresh mode. Failed AI function calls return NULL rather than throwing errors, so data metric functions can monitor for NULL outputs in AI-enriched tables. External engines can also read Snowflake-managed Iceberg tables via the Iceberg REST protocol, preserving the open-table promise while the AI enrichment lives inside. m:…0415
The CDP launch tests the consolidation thesis
Databricks' announcement of a CDP product is where the consolidation story meets its skeptic. A commentator argues the new CDP lacks the breadth of capabilities of traditional CDPs, particularly in real-time use, and estimates an 80% overlap with Hightouch's capabilities. Bloomreach, whose foundation is the Exponea CDP, is a launch partner. The commentator also notes that traditional CDPs have struggled to deliver clear ROI, which is part of why a platform-native alternative has an opening at all. m:…3432
The commentator's sharpest observation is about who will actually use the product: the user of Databricks' CDP will likely be in IT rather than marketing, and the piece contrasts that with Snowflake's approach of prioritizing its partner network. Databricks' Genie, which translates natural-language briefs into SQL queries, is cited as part of the same pattern of moving work into the data platform. The unresolved question for buyers is whether the convenience of native features outweighs the breadth of specialists, and whether a tool built for IT will ever feel like one built for marketers. m:…3432
What is still unresolved
- Will platform-native features like Databricks' CDP displace specialist vendors, or will breadth win?
- Does moving AI enrichment into tables change who owns data quality when prompts change?
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.