AI pulls the data stack into one lane

Databricks owned the fortnight's tempo. On 27 Oct it announced general availability of AI Runtime for customising open models with enterprise data. On 30 Oct it announced general availability of Real-Time Mode for Spark Declarative Pipelines and smarter cost controls for serverless. On 1 Nov it expanded its SAP partnership with three products, Genie, Lakeflow and Lakebase. That is a deliberate sequence: models, pipelines, cost governance and enterprise application adjacency all pushed in one run.

The developer surface moved just as quickly. On 27 Oct, databricks-sql-go 1.18.0 addressed security findings and required Go 1.27.3 or newer. On 29 Oct, terraform-provider-databricks 1.140.0 updated the Go toolchain for certificates. On 31 Oct, CLI v1.23.0 fixed GitHub API rate-limit errors and added functionality for `databricks ssh connect` to make `claude` and `codex` available on dedicated clusters. On 1 Nov, databricks-sql-python 4.8.0 upgraded the SQL Kernel, updated dependencies and fixed Arrow result fetching. On 2 Nov, dbt-databricks 1.13.0 added additive merging of `tblproperties` and opt-in invocation telemetry.

Go-to-market followed the product line. On 30 Oct, Simon Barnes was appointed Genie GTM Lead for Asia-Pacific and Japan. The same period brought postings for a Lakebase Sales Specialist across China, Singapore and India, sales roles in Singapore, Seoul and the public sector, and AI or ML engineering roles. The signal is simple: Databricks is selling AI as a platform motion, not as a feature lane.

Snowflake's fortnight was about surface area. On 29 Oct it completed a $479mn funding round, reported as aimed at boosting its cloud data warehouse offerings. The same day its GitHub organisation introduced `sar-sdk`, `drivers` and `snowflake-connector-net`. On 28 Oct it launched `snowflake-ml-python`, `terraform-provider-snowflake`, `snowflake-actions` and `gosnowflake`; on 30 Oct it created `snowflake-ingest-java`; and on 2 Nov a new `snowflake-ml-python` repository was again observed under the Snowflake organisation. On 29 Oct, Snowpark Python API 1.56.0 added `DataFrameReader.documents()` for reading staged documents into DataFrames.

The paid channel was unusually active. On 28 Oct Snowflake added 15 new Google ad creatives. On 29 Oct it added 10, on 30 Oct seven, on 1 Nov 21 and on 2 Nov 14, with 40 creatives running in each observed set. The important detail is not the individual advert. It is the rotation: the creative ceiling stayed full while the message set was being refreshed repeatedly.

The market-facing proof points also moved. On 28 Oct, Under Armour appeared on Snowflake's customer wall and Flix Mobility was removed. Hiring kept a global AI and engineering shape, with observed roles across London, Zurich, Menlo Park, Singapore, Bangalore, Melbourne, Sydney and Wellington. The operational footnote was less helpful: on 31 Oct Snowflake reported two minor status incidents, one lasting about 206 minutes and another about 1,288 minutes before resolution.

Fivetran spent the fortnight repositioning the connector layer around AI. On 27 Oct its homepage section changed from 'Open Data Infrastructure' to 'Connect your AI to data'. On 28 Oct that line became 'Connect your AI to your data', and the website navigation added a Japanese language option. The extra word matters because the pitch shifts from generic AI connectivity to governed access to a customer's own data estate.

The documentation trail made the same point. On 27 Oct, Fivetran added pages for its connector SDK and applications including Fireworks AI, while removing seven PostgreSQL destination troubleshooting pages. On 29 Oct, it added dbt platform, Salesforce and Zendesk troubleshooting pages while removing dbt Cloud and PostgreSQL guides. On 30 Oct, it added troubleshooting for Coupa OAuth scopes, a Cursor changelog and GitHub long-sync guidance, while removing Cursor AI video and PostgreSQL troubleshooting pages.

GitHub activity backed the repositioning. On 27 Oct, new repositories included `connector_sdk`, `lake_compute_quickstart`, `dbt_openai`, `terraform-provider-fivetran` and community connector work. On 28 Oct, more dbt integrations appeared, including Instagram Business, Facebook Ads, Twilio, Zuora, Workday Financial Management, Xero, YouTube Analytics and TikTok Ads. On 31 Oct, Fivetran released dbt_salesforce v2.4.2-a1 with configurable `salesforce_naming_convention`, and on 1 Nov its great_expectations library moved to 1.24.1 with LIKE-pattern Expectations for Oracle and SingleStore. This is not just source coverage. It is an attempt to own the operating layer around data movement, validation and downstream modelling.

dbt Labs chose content velocity. On 27 Oct it introduced 21 blog pages focused on AI, governance and data quality, removed 22 older blog pages, added on-demand dbt Summit content and created a new competitive page. On 28 Oct it added 38 more blog posts, a Zendesk case study and three product-related pages. On 29 Oct it published a Fanatics case study and a decision-models product page. On 30 Oct it added healthcare AI and dashboard product pages, plus a blog on decision models in the data ecosystem. That is a narrative reset around governance, quality and decisioning.

The reliability note landed on 1 Nov. dbt Labs reported a major incident involving SQL execution failures in dbt Wizard for about 87 minutes, followed by a minor incident with degraded performance across the dbt platform for about 262 minutes. In this part of the stack, reliability is not a back-office metric. It sits directly beside the governance and trust story dbt Labs was publishing all week.

Sigma's moves were quieter but precise. On 27 Oct it published a blog on its architecture runtime for AI. On 28 Oct it added a Sigmaverse App Game Storefront page, added Double Wood Supplements to its customer presence, removed Armstrong Transport Group from the public customer wall, and removed 49 legal product pages. The same day it published a Double Wood case study saying dashboard rebuild time fell from weeks to hours using code-backed Workbooks. On 29 Oct it added a PCI DSS Level 1 compliance blog and removed two Forrester Total Economic Impact pages. On 2 Nov it moved 'Get Started' out of the button position and into navigation as 'Get started', while tightening the security heading from 'Security that follows every user' to 'Security that follows the user.' The pattern was trust proof, applied AI, customer proof and a cleaner web estate.